Tool orientation

The Publication Timeline shows how article output is distributed across years and journals. Each colored band in a year's stacked bar represents one journal's contribution to that year's published total. Read horizontally, the chart traces each venue's publishing rhythm; read vertically, it reads as a profile of the field's annual scholarly throughput.

Run with default parameters against the full Pinakes index, the chart spans 1990 to 2026 and aggregates 35,108 articles across 51 indexed journals. College English contributes the tallest single column (2,622 indexed articles), followed by College Composition and Communication (2,455), IEEE Transactions on Professional Communication (1,918), Rhetorica (1,900), and Computers and Composition (1,543). The five-journal block accounts for roughly a third of the entire indexed corpus.

Use the tool to compare journal output trajectories, locate publication surges that often track editorial transitions or special-issue runs, and read off the relative scale at which each venue contributes to the field's record.

Methodology

Article records are drawn from CrossRef using the Metadata API, queried by journal ISSN.¹ For each journal Pinakes pulls every record that carries a DOI and stores the title, authors, abstract where available, and publication date. A small number of journals lack reliable CrossRef coverage and are indexed via the publisher's RSS feed or by direct web scraping; these records sometimes lack structured publication dates and either fall out of the timeline or are bucketed at the year level only.

The bucketing date is the CrossRef published-print or published-online field, whichever is earlier. Articles are then grouped by year and journal and stacked into the bar chart. Years with no contribution from a given journal appear as gaps in that journal's color band rather than as zero-height slices. The "Show top 8 journals only" toggle re-renders the same data restricted to the eight venues with the highest aggregate output across the full date range.

Two limitations bear on interpretation. First, the timeline reflects indexed output, not total scholarly output: journals added to the index more recently show data only from their indexing date forward, which can produce sharp leading edges that read as publication surges but are recovery artefacts of when Pinakes began crawling that venue. Second, some publishers backdate batch-released content or deposit metadata irregularly with CrossRef, which produces artificial spikes in the year the deposit lands rather than the year the article appeared. The Coverage Notes page at /coverage documents the per-journal indexing window and known coverage gaps.

Controls

Two controls govern the chart.

The Show top 8 journals only toggle restricts the stacked bars to the eight venues with the largest total indexed output, rolling the remaining forty-three into invisibility rather than into an "Other" band. Use it to compare the most prolific venues without the visual noise of long-tail journals; switch it off to see the full distribution and to locate smaller venues' publication windows.

The chart auto-renders on tab open and re-renders instantly when the toggle is flipped; no Compute click is required because the underlying aggregation is precomputed at index-load time and held in memory.

References

1. Hendricks, G., Tkaczyk, D., Lin, J., & Feeney, P. (2020). Crossref: The sustainable source of community-owned scholarly metadata. Quantitative Science Studies, 1(1), 414–427.

Tool orientation

The Topic Co-Occurrence heatmap shows how often pairs of disciplinary topics appear together on the same article. Each cell answers the question: of the articles in the index, how many were tagged with both topic A and topic B? Darker cells mark thematic intersections; faded cells mark topic pairs that rarely sit together in the same paper.

Run with default parameters against the full Pinakes index, the heatmap renders a 61×61 grid. The single heaviest pair is "teacher development" and "writing pedagogy," which co-occur on 1,394 indexed articles. "Rhetorical criticism" and "book reviews" follow at 1,307; rhetorical criticism also pairs heavily with classical rhetoric (1,163), discourse analysis (1,146), modern rhetorical theory (1,108), and argument (1,101), reflecting both the centrality of rhetorical criticism as a research mode and its high base rate inside the indexed corpus. The pedagogy/teacher-development pair is the only intersection in the heatmap that is not anchored to rhetorical criticism on at least one side.

Use the tool to locate dense thematic overlap, identify recognized sub-fields at the intersection of two terms, and notice asymmetries (a topic that appears in heavy pairs across the row but rarely on the diagonal of any other topic's row).

Methodology

Tags are assigned by a rule-based matcher run over each article's title and abstract. The vocabulary is a controlled list of 61 disciplinary topics developed for rhetoric and composition, with each topic backed by a hand-curated keyword pattern (a regular expression with word-boundary anchors and a small set of synonym alternations). The matcher prefers precision to recall: marginal matches are intentionally dropped rather than over-assigned. Articles that lack an abstract receive fewer tags than articles with full text, since the title alone exposes a smaller surface for keyword matching. Roughly 73% of indexed articles currently carry at least one tag, with the remaining 27% concentrated in journals indexed via RSS or scraping rather than CrossRef.

The heatmap reads from a symmetric co-occurrence matrix where cell (i, j) holds the count of articles tagged with both topic i and topic j. Cells are colored on a sequential white-to-brown scale anchored to the matrix maximum, so the darkest cell in any rendering is the heaviest pair in that rendering rather than a fixed threshold. Diagonal cells are not displayed: an article cannot co-occur with itself, and the per-topic article count is reported separately rather than as a self-edge.

The analytic move here is descriptive rather than inferential. Co-occurrence on its own does not measure association strength; it conflates genuine thematic overlap with the joint base rate of two common terms.¹ A topic with high base rate (rhetorical criticism, in this corpus) will appear in heavy pairs across many rows simply because it is common, which is why the dominant rhetorical-criticism pairs run from book reviews to classical rhetoric to argument without that pattern necessarily indexing a single coherent thematic cluster. To control for base rate, divide each cell by the marginal counts and inspect the residual; the heatmap as displayed is the unnormalized signal, which is more legible at a glance but requires this interpretive caveat.

Controls

The heatmap is rendered without user-tunable parameters. The vocabulary, the keyword patterns, and the article corpus are all fixed at the time the page loads. Hover any cell to read the article count for that topic pair; the matrix is symmetric, so cell (i, j) reports the same count as cell (j, i).

Two structural caveats bear on interpretation. The 61-term vocabulary was designed for rhetoric and composition specifically and will systematically under-tag interdisciplinary work that uses adjacent-field terminology (information science, applied linguistics, education theory). Articles indexed via RSS or web scraping rarely carry abstracts and are therefore tagged on title text alone, which makes those venues appear thinner in the heatmap than their actual thematic footprint warrants; the Coverage Notes page at /coverage documents which journals fall in this category.

References

1. Callon, M., Courtial, J.-P., Turner, W. A., & Bauin, S. (1983). From translations to problematic networks: An introduction to co-word analysis. Social Science Information, 22(2), 191–235.

Tool orientation

The Co-Authorship Network maps which authors have written articles together inside the indexed journals. Each circle is one author, sized by their total number of indexed publications; each line is a co-authorship tie, weighted by the number of articles the pair has written together. Authors who collaborate frequently sit close to each other; authors who collaborate rarely or not at all drift apart or, in the case of pure solo authors, sit alone at the periphery as isolates.

Run with default parameters — 150 most-published authors, minimum three indexed publications — the graph contains 266 co-authorship edges connecting 125 of the 150 displayed nodes; the remaining 25 are isolates (frequent solo publishers with no co-author inside the displayed set). The single heaviest tie is between Joe Erickson and Kevin Roozen at 65 shared articles, followed by Cynthia L. Selfe and Gail E. Hawisher at 58, then Jennifer L. Holberg and Marcy Taylor at 45. Selfe and Enos are the most prolific authors in the rendered set at 105 indexed articles each, though only Selfe sits inside one of the densest collaborative clusters; Enos's 105 articles are largely solo, which is a substantively different career profile.

Use the tool to identify collaborative communities, locate high-traffic mentor-student dyads (long-running pairs with double-digit shared output), and notice the difference between productive solo authors and productive collaborators — both are common in rhetoric and composition, and the graph distinguishes them visually.

Methodology

Co-authorship analysis treats each multi-author article as evidence of a collaborative relationship between every pair of authors on the byline.¹ Author names are pulled from CrossRef metadata, normalized by case-folding and whitespace-trimming, and aggregated. The resulting graph is undirected and weighted: each node is an author, each edge connects a pair of authors who share at least one indexed byline, and the edge weight equals the number of articles they have co-authored. Authors with only solo publications inside the indexed set carry no edges and appear as isolates if they meet the publication threshold.

The displayed graph is filtered. Authors must hold at least n indexed publications to be eligible (the "Min. publications" dropdown), and the eligible set is then capped at the top k authors by publication count (the "Authors shown" dropdown). Only edges between authors in the displayed set are rendered; an edge from a displayed author to a non-displayed collaborator is dropped, which means the apparent connectivity of any node is bounded by the cap. Layout is computed by a force-directed simulation in which nodes repel each other via simulated charge and edges act as springs whose pull is proportional to co-authorship weight. Heavier ties produce shorter, thicker lines; the simulation iterates until the configuration stabilizes, which produces visible clustering around long-running collaborative groups.

Author-name disambiguation is the standard limitation of bibliometric collaboration analysis.² Pinakes does not run a disambiguation pass: two scholars with the same name are merged into one node, and the same scholar appearing with a middle initial in some bylines and without it in others may split across two nodes. Coverage is the second limitation: the graph reflects only co-authored work in the indexed journals, so scholars who collaborate primarily in edited collections, monographs, or unindexed venues will appear less connected than their full collaborative record warrants. Solo work is invisible in the edges, though it does drive node size.

For the field's perceived intellectual relationships — which scholars are cited together regardless of whether they have ever collaborated — see the Author Co-Citation tool, which addresses a different question.

Controls

Three controls shape the network. The Reload button is required after changing any of them; the simulation does not auto-recompute on dropdown change.

The Min. publications dropdown sets the eligibility floor. At the default of 3+ indexed publications, the network surfaces consistently-publishing authors and excludes one-off contributors. Lowering to 2+ admits a long tail of newer or guest contributors, which usually adds isolates rather than edges. Raising to 5+ or 10+ filters down to the field's most established authors and tightens the visible clusters but loses several real collaborative relationships at the periphery.

The Authors shown dropdown caps the rendered set at the top-N most-published authors. The default of 150 balances readability against coverage; 50 or 100 produces a sparser, easier-to-read graph that risks dropping legitimate co-author connections, while 300 or 500 fills in the periphery at the cost of visual clutter. A node's apparent isolation often reverses at higher cap settings: an author who looks like an isolate at 150 may have several edges that connect them to authors ranked 151–300.

The Search author input does not refilter the graph; it locates and highlights a node already in the rendered set. Clicking any node highlights that author's connections and surfaces the author info bar with a link to the full author profile.

References

1. Newman, M. E. J. (2001). Scientific collaboration networks. I. Network construction and fundamental results. Physical Review E, 64(1), 016131.
2. Smalheiser, N. R., & Torvik, V. I. (2009). Author name disambiguation. Annual Review of Information Science and Technology, 43(1), 1–43.

Tool orientation

The Author Co-Citation graph maps which scholars the citing community treats as intellectually related. Two authors are co-cited when a later article's reference list includes at least one work by each. A pair's edge weight is the number of indexed articles that have cited both of them; node size reflects each author's total co-citation strength — the sum of weights across every co-citation tie that author holds. This is the field's mental map of who belongs together as an argument-frame, not who has ever collaborated.

Run with default parameters — minimum three co-citations, top 200 authors by total strength — the graph contains 4,870 edges across the 200 displayed nodes. The single heaviest tie pairs Natasha N. Jones with Rebecca Walton at 111 articles co-citing both. The heaviest cluster in the graph is the social-justice technical-communication group: of the top ten ties, eight involve Jones, Walton, Godwin Y. Agboka, Kristen R. Moore, or Angela M. Haas, with Jones–Agboka at 103, Walton–Agboka at 94, and Moore–Jones at 86. Ranked by total co-citation strength, Jones leads at 3,947 (across 11 indexed articles), followed by Jeffrey T. Grabill (3,689 across 17 articles) and Clay Spinuzzi (3,001 across 32 articles). The strength-per-article ratios reward an author whose smaller corpus is heavily co-cited with a tight neighborhood, which is what the Jones–Walton–Agboka pattern surfaces.

Use the tool to identify schools of thought (densely connected clusters), locate scholars who bridge clusters (nodes with edges into multiple groups), and distinguish heavily-published authors whose work is dispersed across multiple conversations from authors whose smaller corpus is concentrated in a single tightly-knit citation neighborhood.

Methodology

Author co-citation analysis was introduced by White and Griffith to map the intellectual structure of information science, treating the question "which authors does the citing community perceive as belonging together?" as a structural question answerable from reference-list co-occurrence.¹ The method extends Small's article-level co-citation measure by aggregating each cited article up to its author(s).² A high author co-citation count carries more inferential weight than direct citation, shared topic tags, or shared journal venue, because it reflects the collective judgment of many citing authors rather than the choice of any single author.

The aggregation proceeds article by article. For each indexed article C with a reference list deposited at CrossRef, the analysis identifies every pair of cited articles in C's reference list and maps each cited article to its author(s). A co-citation of cited articles X and Y by C produces one co-citation instance for every author pair (a, b) where a is an author of X and b is an author of Y. Multi-author articles therefore generate combinatorially more author-level instances than single-author articles. Pairs of authors who appear on the same cited article are excluded from the count: without this correction, every joint article a co-author pair has written would inflate their co-citation tie with every author cited alongside it, conflating the social fact of co-authorship with the perceptual fact of co-citation.

The resulting graph is undirected and weighted. Nodes are authors; edges connect pairs whose co-citation count meets the slider threshold. Node size scales to total co-citation strength, computed as the sum of edge weights incident to that node, which measures depth of embedding in the citation community rather than raw publication volume. Layout uses a force-directed simulation: nodes repel each other via simulated charge while edges act as springs whose pull is proportional to co-citation weight.

Three limitations bear on interpretation. The graph is corpus-bounded: a scholar whose work is heavily cited in books, edited collections, or journals outside the indexed set will appear less central than their wider record warrants. Author-name disambiguation is the standard bibliometric limitation — common names may merge distinct scholars and name-form variants may split a single scholar across nodes.³ The measure is retrospective: scholars early in their careers may not yet appear prominently because their work has had less time to be co-cited.

Controls

Five controls shape the graph. Filter changes do not auto-recompute; the Compute button re-runs the aggregation against the current filter state.

The minimum-co-citations slider filters which edges enter the graph. At the default of three co-citations a pair must appear together in at least three indexed reference lists to be drawn. Lowering to one admits speculative ties that often reflect a single citing author's idiosyncratic juxtaposition; raising to ten or twenty isolates the most established perceived kinships at the cost of pruning emerging conversations.

The Authors shown dropdown caps the rendered set at the top-N authors by total co-citation strength. The default of 200 produces a dense but legible graph; smaller caps surface only the most embedded scholars and risk dropping bridge nodes that connect clusters.

The From / To year filters restrict the set of citing articles whose reference lists feed the aggregation. Restricting to 2015–2025 reframes the question as "how has the field perceived these scholars in the last decade?" and tends to surface recent disciplinary turns; the social-justice TC cluster reads as much heavier under a 2018–present filter than under an all-time filter, which is a substantive finding about when those co-citation ties accumulated.

The Journals picker narrows the set of citing articles to a chosen subset. Restricting to the technical-communication venues asks "what does the TC sub-community perceive as related?" rather than "what does the wider field perceive?"; the cluster structure can shift considerably across these scopings.

For the field's collaborative relationships rather than its perceived intellectual ones, see the Co-Authorship Network tool. The two visualisations rest on disjoint signals and routinely disagree about which scholars belong together.

References

1. White, H. D., & Griffith, B. C. (1981). Author cocitation: A literature measure of intellectual structure. Journal of the American Society for Information Science, 32(3), 163–171.
2. Small, H. (1973). Co-citation in the scientific literature: A new measure of the relationship between two documents. Journal of the American Society for Information Science, 24(4), 265–269.
3. Smalheiser, N. R., & Torvik, V. I. (2009). Author name disambiguation. Annual Review of Information Science and Technology, 43(1), 1–43.

Tool orientation

The Citations tool ranks indexed articles by the number of other indexed articles that cite them. Each row shows one article and an internal citation count: the number of distinct indexed articles whose CrossRef reference list contains a DOI matching that article. The list is a frequency snapshot of internal influence, not a measure of an article's total scholarly reach.

Run with default parameters (no filters), the tool ranks all articles in the index by descending citation count. The single most-cited article inside the corpus is Jones, Moore, and Walton's 2016 "Disrupting the Past to Disrupt the Future: An Antenarrative of Technical Communication" in Technical Communication Quarterly, with 146 internal citations. Jones's solo 2016 piece "The Technical Communicator as Advocate" follows at 144, then Edbauer's 2005 "Unframing Models of Public Distribution" at 135 and Russell's 1997 "Rethinking Genre in School and Society" at 129. The distribution falls quickly: the fifth-ranked article carries 120 citations, the tenth 76, the twenty-fifth 51, and the fiftieth 42, which gives a quick read on how concentrated the field's internal citation traffic is at the top.

Use the tool to identify the articles the indexed journals are reading and re-citing, surface concentrations of influence inside particular sub-areas (filter by topic or by journal to narrow the scope), and locate the year-by-year shape of the citation distribution.

Methodology

Internal citation counts are derived from reference lists deposited by publishers with CrossRef. The fetcher pulls each indexed article's reference list, extracts the DOI of every cited work, and joins on the articles table to identify which references point to other indexed articles. A successful join produces one citation; failed joins are dropped, which is the correct behavior for an internal-only measure but means the count is silent about citations to non-indexed scholarship. Self-citations are retained if the publisher's deposit lists them in the reference array.

Ranking is by descending count of distinct citing articles, with ties broken alphabetically by title. The internal-citation distribution is a long-tail distribution dominated at the top by a small set of heavily-cited landmark articles and at the bottom by a long tail of articles cited once or not at all — a pattern consistent with citation-distribution literature dating to Lotka.¹ The top five articles in this index account for roughly 670 internal citations; the next forty-five account for around 2,500.

Three limitations bear on interpretation. The count reflects citations within the indexed journals only: many rhetoric and composition venues do not deposit reference lists with CrossRef and therefore contribute no citing signal, even when their articles do cite indexed work. The Coverage Notes page at /coverage documents per-journal deposit rates. Citations are also retrospective: older articles have had more years to accumulate citations than newer articles, which is why the top of the list skews toward landmark work from the 1980s through 2010s rather than the most recent publication years. Finally, internal citation count is not a quality measure: an article cited 100 times because it is contested or methodologically problematic registers identically to an article cited 100 times because it is foundational.

Controls

Four filters refine the ranking. They auto-recompute on change — no Compute button is required because the underlying query is fast.

The From / To year filters restrict the set of cited articles by publication year, not the set of citing articles. Restricting to 2015–2025 surfaces the most-cited recent work and lets the user read how rapidly the post-social-justice-turn citation pattern accumulated; restricting to pre-2000 returns a different ranked list dominated by foundational work.

The Journal filter narrows the cited set to a single venue. Selecting College Composition and Communication ranks only CCC articles by their internal citation count from the rest of the index; the resulting list is a quick read on which articles in that venue have made the deepest mark on the wider field.

The Topic filter restricts to articles carrying a chosen disciplinary tag. The dropdown is ordered by tag frequency; tag assignment is rule-based against title and abstract text, which means a topic filter applied to RSS-indexed venues without abstracts will under-recall.

Filters compose: From 2010 + Journal: CCC + Topic: writing centers returns CCC articles published in or after 2010 that carry the writing-centers tag, ranked by internal citations. Empty intersections produce an empty list rather than a fallback.

References

1. Lotka, A. J. (1926). The frequency distribution of scientific productivity. Journal of the Washington Academy of Sciences, 16(12), 317–323.

Tool orientation

The Citation Trends chart tracks how densely the indexed articles cite each other, year by year. The y-axis plots the average number of internal references per article — that is, on average, how many of each year's articles' reference-list entries point to other articles in the indexed corpus. The metric measures internal connectedness, not external impact.

Run with default parameters across the full index, the line rises monotonically across the indexed period. In 1990 the average article carries 0.47 internal references; the metric crosses 1.0 in 2007 (1.06), reaches 2.62 in 2022, 3.25 in 2024, and 3.33 in 2025. Total internal references peak in 2024 at 3,466 across 1,068 articles. The 2026 datapoint reads 4.01 references per article but rests on only 345 indexed articles — a partial-year sample whose mean is unstable and whose value will fall as the year fills out.

Use the tool to read the field's growing internal citation density, identify decades where indexed articles began citing each other in earnest, and compare a single journal's trajectory against the all-journals baseline.

Methodology

The metric plotted is outbound internal references per article: for each indexed article whose reference list has been fetched from CrossRef, the number of references pointing to another article in the indexed set is counted, and the result is averaged within publication year. Articles whose reference lists have not yet been fetched are excluded from the average rather than counted as zero, which would otherwise depress the curve in years with low fetch coverage. The chart series is restricted to years from 1990 forward, where the indexed corpus is dense enough for a stable mean.

The rising trend is not a measure of citation impact and should not be read as such. Two structural factors drive the increase. First, the indexed corpus has grown from a few thousand articles in the early 1990s to roughly 35,000 today, which means a 2024 article writing on the same topic as a 1995 article has many more eligible internal targets to cite. Second, citation behavior is biased toward recent work; an article published in any given year tends to cite work from the preceding decade more often than older work, so as the indexed corpus densifies year over year, the recent-work pool that articles preferentially cite grows in step.¹ What the chart measures is therefore internal citation density as a structural property of the index, not the rising or falling influence of any one cohort of articles. For inbound citation impact, see the Citations tab and the Half-Life tool.

Three limitations bear on interpretation. The averages are volatile in years with few articles — the partial 2026 datapoint demonstrates the issue. Journals that entered the index recently contribute a truncated time series, which can produce apparent year-on-year jumps that reflect indexing-coverage changes rather than citation behavior. And the metric is silent about citations to scholarship outside the indexed corpus: a journal whose articles cite mostly books and unindexed venues will register a low average that does not reflect the citation depth of its actual reference lists.

Controls

One control governs the chart.

The Journal filter restricts the chart to articles published in a single venue. Selecting Computers and Composition or Technical Communication Quarterly compares that journal's internal-reference density against the field-wide trend. Journals that engage heavily with the indexed corpus produce trajectories that sit above the all-journals line; journals that primarily cite outside the index sit below it. The shape of each journal's curve is also informative: a flat line suggests stable citation behavior, while a steep rise suggests an editorial or topical shift that drew the venue into closer dialogue with the rest of the indexed field.

The chart auto-renders on filter change; no Compute click is required.

References

1. Price, D. J. de S. (1965). Networks of scientific papers. Science, 149(3683), 510–515.

Tool orientation

The Citation Network displays the actual directed citation graph between indexed articles. Each circle is one article, sized by how many indexed articles cite it; each arrow points from a citing article to the article it cites. Density and clustering in the rendered graph reveal which articles are at the center of the field's internal conversation and which sub-areas form tight citation neighborhoods around shared landmark texts.

Run with default parameters — minimum five internal citations, all years, all journals — the graph displays 500 articles connected by 1,315 directed citation edges. Ranked by in-graph degree (citations received from other displayed nodes), the central node is Jones, Moore, and Walton's 2016 "Disrupting the Past to Disrupt the Future" with 38 incident edges, followed by Russell's 1997 "Rethinking Genre in School and Society" at 32, Carolyn D. Rude's "Contemporary Research Methodologies in Technical Communication" at 29, and Haas's 2012 "Race, Rhetoric, and Technology" at 27. The technical-communication and genre-theory clusters dominate the rendered structure; rhetoric-of-science and writing-center sub-areas appear as smaller, looser neighborhoods at the periphery.

Use the tool to identify landmark articles by visual centrality, locate sub-area clusters formed by shared internal citation, trace citation chains between specific articles by clicking and following edges, and read off the structural difference between heavily-interconnected sub-fields and sparsely-connected ones.

Methodology

The graph is built from CrossRef reference deposits. For each indexed article whose reference list has been fetched, every reference whose DOI matches another indexed article produces one directed edge from citing to cited. The resulting graph is filtered before display: nodes must hold at least n incident internal citations to be eligible (the minimum-citations slider), and the eligible set is capped at the top 500 articles by internal citation count to keep rendering responsive. Edges are retained only when both endpoints survive the node filter; a citation from a displayed article to a non-displayed article is dropped, which means the visible degree of any node is bounded by the cap rather than the full internal citation graph.

Layout uses a force-directed simulation. Nodes repel each other via simulated charge while edges act as springs whose pull is proportional to edge presence. The simulation runs until the configuration stabilizes; densely-interconnected article neighborhoods coalesce into clusters and lightly-cited articles drift to the periphery. Node size is proportional to internal-citation count and node color is assigned by source journal, which makes journal homogeneity within clusters visible without an explicit grouping step.

Two limitations bear on interpretation. The network shows only internal citation relationships; an article heavily cited outside the indexed corpus appears small or, if its in-graph degree falls below the slider threshold, absent.¹ Journals that do not deposit reference lists with CrossRef contribute no outbound edges, which means their citation behavior is structurally invisible even when their articles do receive citations from other indexed work. The 500-node cap is the second constraint: at low minimum-citation thresholds the cap binds tightly and excludes peripheral articles whose connectivity to the core is real but lightly weighted. Raising the slider tightens the visible core; lowering it admits more peripheral work but is constrained by the node cap.

Controls

Four controls scope the graph. Filter changes do not auto-recompute; the Compute button re-runs the query against the current filter state.

The minimum-citations slider filters which articles are eligible to appear. At the default of five internal citations, the graph carves to articles cited by at least five other indexed articles. Lowering to one or two admits a long tail of single-cited work and tends to fill the periphery with isolates and small components. Raising to ten or fifteen produces a sparser, more legible graph dominated by the field's most heavily-cited landmark texts.

The From / To year filters restrict articles by publication year. A tight range — 2015 to 2025, for instance — reframes the question from "what is the field's all-time citation structure?" to "what does the citation neighborhood of the last decade look like?", which often surfaces the social-justice technical-communication cluster as the dominant component.

The Journals picker narrows the corpus to a chosen subset. Restricting to a single journal exposes how that venue's articles cite each other internally; restricting to a journal cluster (the technical-communication venues, for example) shows the cluster's internal citation structure with cross-cluster edges removed.

The Search bar does not refilter the graph; it locates and highlights a node already in the rendered set by title or author. Click any node to open the full article page; hover any node to read its citation count and basic metadata.

References

1. Hicks, D., Wouters, P., Waltman, L., de Rijcke, S., & Rafols, I. (2015). Bibliometrics: The Leiden Manifesto for research metrics. Nature, 520(7548), 429–431.

Tool orientation

The Centrality tool computes two structural-importance scores on the indexed citation network: eigenvector centrality (implemented as PageRank), which rewards articles embedded in the most heavily-connected region of the field, and betweenness centrality, which rewards articles that sit on the shortest paths between sub-areas. The two metrics often disagree, and the disagreement is the analytic point: an article can be deeply embedded in a single cluster (high eigenvector, low betweenness) or sparsely connected but structurally indispensable as a bridge between clusters (low eigenvector, high betweenness).

Run with default parameters — minimum two internal citations, all years, all journals — the tool computes both metrics across 600 nodes connected by 1,624 edges. Stephen P. Witte's 1992 "Context, Text, Intertext" leads the eigenvector ranking at 1.0 (the normalized maximum), followed by Greg Myers's 1985 "The Social Construction of Two Biologists' Proposals" at 0.998 and Lucille Parkinson McCarthy's 1987 "A Stranger in Strange Lands" at 0.765. The betweenness ranking is led by Russell's 1997 "Rethinking Genre in School and Society" at 1.0, with Geisler et al.'s 2001 "IText" at 0.464 and Johndan Johnson-Eilola's 1996 "Relocating the Value of Work" at 0.449.

The two top-tens overlap on Witte and Russell but otherwise pull in different directions. Russell's "Rethinking Genre" is the field's heaviest bridge: a 1997 piece that sits on more shortest paths than any other article in the network, despite ranking only seventh on eigenvector. Witte's "Context, Text, Intertext" is the inverse signature — deeply embedded in a tightly connected neighborhood that pulls its eigenvector score toward 1.0 while sitting only fifth on betweenness.

Use the tool to identify intellectual epicentres (high-eigenvector articles) and bridge texts (high-betweenness articles), and to surface the analytic difference between an article that anchors a cluster and an article that connects clusters.

Methodology

The citation network is constructed as a directed graph G = (V, E) where each node v is an indexed article and each directed edge (u, v) means article u cites article v. The graph is filtered before centrality computation: nodes must hold at least n internal citations to be eligible (the minimum-citations slider) and the eligible set is capped at the top 600 articles by citation count to keep the eigenvector and betweenness solvers responsive.

What each metric measures. Before the implementation details, the conceptual difference is the analytically important thing. Eigenvector centrality propagates importance through edges: an article cited by five highly-cited articles scores higher than one cited by fifty peripheral articles, because the metric weights the score of each citing article by the citing article's own centrality. It rewards influence-by-association — being read by the work that other work reads. Betweenness centrality is structural rather than associative: it is high when removing the article from the network would lengthen the shortest paths between many pairs of other articles, regardless of those articles' citation counts. It rewards bridging — sitting on the routes between conversations that would otherwise be more distant. The two metrics often disagree, and the disagreement is the analytic point.³ The Pruitt (2026) bibliometric analysis of CCC 1950–2022 used the same two-metric framing on a single-journal corpus; this tool extends it across the full multi-journal index.⁴

How the metrics are computed. Eigenvector centrality is implemented as PageRank with damping factor α = 0.85.¹ The classical eigenvector definition assigns zero to every node outside the largest strongly-connected component, which is fatal for a citation graph that contains many small disconnected sub-clusters. PageRank's damping term models a random walker who follows citation edges 85% of the time and teleports to a random node 15% of the time, ensuring every node receives a nonzero score and producing meaningful rankings across the full graph. Betweenness centrality is computed using Brandes's O(|V|·|E|) algorithm, which counts for each node v the fraction of all-pairs shortest paths that pass through v.² Both metrics are normalized to a 0–100% scale relative to the highest-scoring article in the current filtered set, so the leader is always 1.0 and other scores are reported as fractions of that leader.

Two limitations bear on interpretation. The metrics reflect structural position within the indexed corpus only: an article that is foundational to the broader field but rarely cited inside the indexed journals will not score highly, and the rankings tilt toward whichever sub-areas have the densest internal citation chains in this corpus. The metrics are also sensitive to the time-accumulation effect: older articles have had more years to accumulate citation paths, and they therefore dominate both rankings. The year-range filters are the standard way to control for this effect; restricting the corpus to a 1995–2010 window, for example, surfaces a substantively different set of bridge articles than the all-time view.

Controls

Four controls scope the network. Filter changes do not auto-recompute; the Compute button is required because betweenness centrality is the slow operation in the pipeline and must run against the current filter state.

The minimum-citations slider sets the eligibility floor. At the default of two internal citations the graph admits 600 articles and surfaces a layered ranking in which the centrality leaders are landmark texts but the second tier surfaces less-canonical bridge articles. Lowering to one admits a long tail and inflates the betweenness scores of articles that sit on otherwise sparse paths. Raising to five or ten carves the graph to the most heavily-cited core and tightens both rankings around the field's most-canonical work.

The From / To year filters restrict articles by publication year. A tight range strips the network of older accumulators and surfaces the bridge articles operative within that period; restricting to 2010–2025 makes the social-justice technical-communication cluster much more prominent in both rankings than the all-time view.

The Journals picker narrows the corpus to a chosen subset. Restricting to a single venue or a journal cluster computes centrality on the smaller graph and produces venue-internal rankings; cross-cluster citations are dropped along with their endpoints.

The node-size and node-color toggles in the chart toolbar do not refilter the graph; they re-skin it. Node size can be toggled between eigenvector-centrality scaling, betweenness-centrality scaling, and raw citation count. Node color can be toggled between journal identity (each journal a distinct color) and a heatmap gradient on the active centrality metric. Switching toggles is instant and requires no recomputation.

References

1. Brin, S., & Page, L. (1998). The anatomy of a large-scale hypertextual web search engine. Computer Networks and ISDN Systems, 30(1–7), 107–117.
2. Freeman, L. C. (1977). A set of measures of centrality based on betweenness. Sociometry, 40(1), 35–41.
3. Brandes, U. (2001). A faster algorithm for betweenness centrality. Journal of Mathematical Sociology, 25(2), 163–177.
4. Pruitt, M. (2026). (Re)Considering centrality: A bibliometric network analysis of College Composition and Communication, 1950–2022. College Composition and Communication, 77(3). https://publicationsncte.org/content/journals/10.58680/ccc2026773430

Tool orientation

The Community Detection tool partitions the citation network into clusters of articles that cite each other more densely than they cite work elsewhere in the index. The algorithm finds the partition automatically rather than working from a predefined topic list, which means the resulting communities are emergent groupings driven by citation behavior rather than by editorial classification or subject tagging. Each community amounts to a research front: a group of articles whose authors have read each other and built on each other's work.

Run with default parameters — minimum two internal citations, resolution 1.0, all years and journals — the algorithm partitions 527 articles connected by 1,621 edges into 17 communities with a modularity Q of 0.587. The dominant cluster (C0, 91 articles) is the genre / activity-systems community led by Russell's 1997 "Rethinking Genre" and rooted in Written Communication, JBTC, and TCQ. The second cluster (C1, 77 articles) is the social-justice technical-communication community led by Jones, Moore, and Walton's 2016 "Disrupting the Past" and concentrated in TCQ. The third (C2, 71 articles) is the digital-rhetoric / computers-and-composition cluster led by Cooper and Selfe's 1990 "Computer Conferences and Learning" and rooted in Computers and Composition. Smaller communities surface the assessment cluster (C6, anchored by Flower and Hayes 1981 "A Cognitive Process Theory of Writing" in Assessing Writing), the intercultural TC cluster (C7), and the writing-center / service-learning cluster (C5).

Use the tool to identify the field's research fronts as the citation graph reveals them, locate which journals concentrate in which communities, and read off how stable or fragmented a sub-area is from the size of its cluster and the cohesion of its journal mix.

Methodology

Community detection partitions a network into groups such that within-group edge density is materially higher than between-group density.¹ The measure of fit is modularity Q, which scores a partition by comparing observed within-group edge weight against the expected weight under a degree-preserving random null model.² Values of Q above roughly 0.3 indicate meaningful community structure; the default-parameter Q of 0.587 sits well into the well-clustered range and reflects that the indexed citation network is genuinely modular rather than an undifferentiated mass.

The algorithm is the Louvain method — a greedy agglomerative procedure that iteratively moves each node to whichever neighboring community produces the largest gain in Q, then folds each community into a single super-node and repeats, terminating when no move improves Q.³ The Louvain algorithm is non-deterministic in its tie-breaking, so two runs on the same input may produce slightly different partitions for low-degree boundary nodes, though the overall cluster structure is stable. The directed citation graph is symmetrized before community detection because intellectual affinity, as opposed to citation flow, is undirected: a reciprocal citation between two articles produces an edge weight of 2 rather than two separate edges.

The resolution parameter γ modifies the modularity objective to favor finer or coarser partitions.⁴ At γ = 1.0 (the standard Newman-Girvan modularity), the algorithm produces 17 communities at this corpus size; values below 1.0 favor merging adjacent communities into larger blocs, and values above 1.0 favor splitting communities into smaller, more topically-uniform clusters. The trade-off is between communities that are large enough to be analytically informative and communities that are small enough to be topically coherent. Each community is summarized by its top articles (by internal citation count), top journals (by article count within the community), and top topics (extracted from the controlled-vocabulary tags assigned to its articles); these summaries appear in the legend and the per-community tables.

Two limitations bear on interpretation. The communities are corpus-bounded: a research front whose citation chains run primarily through unindexed venues will not appear as a coherent cluster here, and the visible structure tilts toward whichever sub-areas have the densest internal citation chains. The partition is also unstable at the boundary: a peripheral article that sits in roughly equal proximity to two communities may be assigned to either one across runs, which is why the per-community tables list the high-citation core articles rather than every member.

Controls

Five controls scope the partition. Filter changes do not auto-recompute; the Compute button is required because Louvain runs several seconds against a multi-thousand-node graph.

The minimum-citations slider sets the eligibility floor for inclusion in the network. At the default of two internal citations, peripheral articles enter the graph and the algorithm finds finer-grained communities at the cost of some boundary noise. Raising to five or ten produces fewer, larger communities anchored by the field's most-cited landmark articles.

The resolution slider controls cluster granularity directly, ranging 0.5 to 2.0. Lower values (0.7–0.9) merge the social-justice TC and intercultural TC clusters into a single technical-communication-of-difference community; higher values (1.3–1.7) split the genre cluster into separate sub-communities for activity systems, professional writing, and academic genre studies. Move the slider to test whether a particular structural feature is robust across resolutions or specific to one resolution choice.

The From / To year filters restrict the network by article publication year. Detecting communities on a 2010–2025 slice yields a substantively different partition than the all-time view: the social-justice TC cluster grows into the largest community, and several pre-2000 communities disappear because their core articles fall outside the date filter.

The Journals picker narrows the corpus to a chosen subset. Restricting to a journal cluster (the technical-communication venues, for instance) detects communities within that scope and produces a finer-grained view of how that sub-area's research fronts are organized.

References

1. Fortunato, S. (2010). Community detection in graphs. Physics Reports, 486(3–5), 75–174.
2. Newman, M. E. J. (2006). Modularity and community structure in networks. Proceedings of the National Academy of Sciences, 103(23), 8577–8582.
3. Blondel, V. D., Guillaume, J.-L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 2008(10), P10008.
4. Reichardt, J., & Bornholdt, S. (2006). Statistical mechanics of community detection. Physical Review E, 74(1), 016110.

Tool orientation

The Article Co-Citation graph maps which articles the citing community treats as part of the same intellectual conversation. Two articles are co-cited when a third article includes both in its reference list; the more often that joint citation happens, the stronger the field's perceived kinship between the pair. Direct citation records who cites whom; co-citation records who is treated as belonging together — a different question.

Run with default parameters — minimum three co-citations, top 400 articles by co-citation strength — the graph contains 3,435 edges. Ranked by total co-citation strength, the article most embedded in the co-citation structure is Agboka's 2013 "Participatory Localization" at 1,083, followed by Jones's 2016 "The Technical Communicator as Advocate" at 1,056, Jones, Moore, and Walton's 2016 "Disrupting the Past" at 1,000, and Haas's 2012 "Race, Rhetoric, and Technology" at 878. The single heaviest co-citation tie pairs Jones, Moore, and Walton (2016) with Jones (2016) at 60 co-citations — an unusually high number that reflects both texts being routinely cited together as the foundational pair of the social-justice TC turn. Agboka and Haas pair at 52, Agboka and Jones at 49, Jones and Haas at 49.

Use the tool to identify the field's perceived intellectual groupings, locate the central texts of recognised research fronts, and find articles that bridge clusters — nodes whose edges reach into multiple visually distinct neighborhoods.

Methodology

Co-citation analysis was introduced by Henry Small as a measure of perceived relationship between documents.¹ The construction is straightforward but the analytic move is not. For each indexed article C with a reference list deposited at CrossRef, the analysis identifies every pair (X, Y) of cited articles in C's reference list and increments the co-citation count for that pair by one. The accumulated counts produce an undirected weighted graph where each node is a cited article and each edge weight is the number of indexed articles that have cited both endpoints together.

Where direct citation records intellectual debt as a directed claim from one author about another, co-citation records perceived kinship as a derived signal from many citing authors collectively. Two articles can be heavily co-cited even if neither has ever cited the other; the edge weight is built up by the citing community, not by either endpoint. Node size in the rendered graph reflects co-citation strength, computed as the sum of edge weights incident to that node, which measures the article's depth of embedding in the perceived-kinship structure rather than its raw citation count. Layout uses a force-directed simulation in which co-cited articles attract each other; the resulting visual clusters correspond to research fronts as the citing community has organised them.

The displayed graph is filtered. An edge must meet the minimum co-citations threshold to be drawn; the eligible node set is then capped at the top 400 articles by total co-citation strength to keep the simulation responsive. Edges are retained only when both endpoints survive the node filter, which means the visible degree of any node is bounded by the cap. Co-citation is retrospective in a way that direct citation is not: very recent articles are penalized because subsequent work has had less time to co-cite them, while landmark articles from prior decades accumulate co-citation strength on a long horizon. The default parameter run shows this asymmetry plainly — the top of the strength ranking is dominated by 2012–2016 social-justice TC pieces because those texts have been intensively co-cited in the last decade, while pre-2000 work appears in the rankings primarily through articles that became canonical reference frames (Russell 1997, Connors 1982) rather than through volume.

Two limitations bear on interpretation. The graph reflects only co-citations from indexed articles whose reference lists have been deposited with CrossRef; journals without reference deposit contribute no co-citation signal. The network is also corpus-bounded: a pair of articles routinely co-cited in books, dissertations, or unindexed journals will appear less central here than their broader perception warrants. For the author-level equivalent — which scholars the field treats as intellectually paired regardless of which specific articles are cited — see the Author Co-Citation tool.

Controls

Four controls scope the graph. Filter changes do not auto-recompute; the Compute button re-runs the aggregation against the current filter state.

The minimum-co-citations slider filters which edges enter the graph. At the default of three co-citations the graph admits a wide neighborhood; raising to ten or fifteen isolates only the most established perceived kinships, which can collapse the visible structure to a small core dominated by social-justice TC and a separate genre/activity cluster.

The From / To year filters restrict the set of citing articles whose reference lists feed the aggregation. Restricting to 2018–2025 reframes the question as "what does the field perceive as related now?" and tends to amplify the social-justice cluster because that's where recent citation traffic concentrates. A 1990–2010 filter surfaces a substantively different set of perceived kinships built on the genre and rhetoric-of-science citation chains active in that era.

The Journals picker narrows the set of citing articles to a chosen subset. Restricting to a single venue or journal cluster asks "what does this sub-community perceive as related?" rather than "what does the field as a whole perceive?" The cluster structure can shift considerably across these scopings.

References

1. Small, H. (1973). Co-citation in the scientific literature: A new measure of the relationship between two documents. Journal of the American Society for Information Science, 24(4), 265–269.
2. Marshakova-Shaikevich, I. (1973). Co-citation in scientific literature: A new measure of the relationship between publications. Scientific and Technical Information Processing, 6(2), 3–8.

Tool orientation

Bibliographic Coupling links articles that cite the same prior work. Two articles are bibliographically coupled when they both include a reference to a third indexed article; the coupling strength of the pair is the number of references they share. A high coupling strength means the two articles are reading the same foundational literature — engaged with the same conversation, working from the same theoretical sources, or addressing the same body of empirical work — even if neither cites the other directly. Where Co-Citation measures perceived kinship from the field's reception of prior work, bibliographic coupling measures intellectual input as fixed at the moment of publication.

Run with default parameters — minimum three shared references, top 400 articles by coupling strength — the graph contains 5,065 edges. Ranked by total coupling strength, the most heavily-coupled article is Jones, Moore, and Walton's 2016 "Disrupting the Past" at 797, followed by Moeggenberg, Edenfield, and Holmes's 2022 "Trans Oppression Through Technical Rhetorics" at 762, Itchuaqiyaq, Edenfield, and Grant-Davie's 2022 "Sex Work and Professional Risk Communication" at 674, Edenfield, Colton, and Holmes's 2019 "Always Already Geopolitical" at 646, and Petersen and Walton's 2018 "Bridging Analysis and Action" at 620. The single heaviest coupling tie pairs Moran and Tebeaux's 2012 history-of-technical-communication article with Edward Malone's 2007 piece on the same topic at 46 shared references — a coupling that signals two articles drawing on essentially the same source corpus, which is the structural fingerprint of historical-survey work.

Use the tool to identify recent articles that share intellectual foundations even before the field has co-cited them, locate methodological clusters where authors are reading the same theoretical sources, and surface coupling patterns that mark survey or review articles as distinct from primary research articles.

Methodology

Bibliographic coupling was introduced by Kessler as one of the first quantitative methods for measuring intellectual relationship between scientific papers, predating Small's co-citation by a decade.¹ The construction is symmetric to co-citation but inverted in direction: where co-citation builds an edge weight from the count of articles that cite both endpoints, coupling builds an edge weight from the count of articles that both endpoints cite. Co-citation strength accumulates as the field reads two articles together; coupling strength is fixed the moment the article's reference list is deposited.

The construction proceeds article by article. For each indexed article whose reference list has been fetched, the analysis extracts the set of indexed-article DOIs in that reference list and pairs the article with every other indexed article whose reference list overlaps. The edge weight equals the cardinality of the intersection — the number of references the two articles share. The resulting undirected weighted graph carries one node per article and one edge per article pair with at least one shared reference. References to non-indexed work (books, dissertations, journals outside the corpus) are not visible to the matcher and contribute no coupling weight, which means an article whose reference list is dominated by external sources will appear less coupled than its actual intellectual position warrants.

Node size in the rendered graph reflects total coupling strength — the sum of edge weights incident to that node — rather than raw citation count. This means a heavily-cited landmark article from 1990 may not appear among the top coupling nodes if its own reference list is short or weighted toward external sources, while a recent literature-review article with a long indexed-only reference list may appear at the top of the strength ranking. The Moran-Tebeaux / Malone tie at 46 shared references illustrates this signature: both are historical-survey articles whose long indexed-source reference lists overlap heavily.

Two limitations bear on interpretation. Coupling strength is influenced by reference-list length, which biases the ranking toward articles whose references are long and stay inside the indexed corpus — literature reviews, methodological surveys, and articles with extensive citation engagement. Articles that engage seriously with prior work but cite mostly external sources will register as less central than they should. The second limitation is bibliometric scope: bibliographic coupling produces visible clusters that correspond to "what's being read" rather than "what's been written about," which can diverge from a field's perceived structure when the citing community has not yet reached consensus on what counts as canonical.²

Controls

Four controls scope the graph. Filter changes do not auto-recompute; the Compute button re-runs the aggregation against the current filter state.

The minimum-shared-refs slider filters which edges enter the graph. At the default of three shared references the graph admits a wide neighborhood; raising to ten produces a sparse graph dominated by literature-review and survey articles, which is the analytic move when the question is "which articles are reading nearly the same source corpus?"

The From / To year filters restrict the article set by publication year. Coupling is well suited to mapping recent work because the metric is fixed at publication; restricting to 2020–2025 produces a coupling map of the post-pandemic publication burst that would be substantially harder to build from co-citation, which has not had time to accumulate.

The Journals picker narrows the corpus to a chosen subset. Restricting to a single venue or journal cluster maps how that sub-area's articles read each other's foundations.

References

1. Kessler, M. M. (1963). Bibliographic coupling between scientific papers. American Documentation, 14(1), 10–25.
2. Boyack, K. W., & Klavans, R. (2010). Co-citation analysis, bibliographic coupling, and direct citation: Which citation approach represents the research front most accurately? Journal of the American Society for Information Science and Technology, 61(12), 2389–2404.

Tool orientation

The Sleeping Beauties tool surfaces articles whose citation history is sharply non-linear: years of dormancy after publication, then a late surge of recognition. The metaphor — that an article "sleeps" until a "prince" paper or a disciplinary turn makes it newly relevant — is bibliometric rather than literary, and the algorithm operationalizes it via the Beauty Coefficient, a scalar that scores how far an article's actual citation curve departs from a steady linear rise toward its peak year.

Run with default parameters — minimum five total citations, all years and journals — the tool returns the top 50 articles ranked by Beauty Coefficient. The leading sleeper is Kelli Cargile Cook's 2002 "Layered Literacies: A Theoretical Frame for Technical Communication Pedagogy" with B = 116.0: published in Technical Communication Quarterly with near-zero citations through the mid-2010s, then awakened in 2024 to a peak of 16 citations after sleeping for 22 years. Robert J. Connors's 1984 "Journals in Composition Studies" follows at B = 101.0 with a 35-year sleep that ended in 2019, and the 1974 "Students' Right to Their Own Language" introduction sits at B = 63.0 with a 44-year sleep that ended in 2018. The pattern is interpretable: each leading sleeper exemplifies a disciplinary turn that made earlier work newly citable — pedagogy as a frontier in TC scholarship for Cargile Cook, the historical-self-study turn for Connors, the social-justice turn for "Students' Right."

Use the tool to identify articles whose late uptake marks a disciplinary turn, locate texts that may now be in their "prince" phase — articles attracting fresh citations because the field's interests have shifted to where the article was already pointing, and read citation timelines for the underlying shape rather than the headline number.

Methodology

The metric is the Beauty Coefficient B introduced by Ke, Ferrara, Radicchi, and Flammini.¹ For an article published in year t₀ with peak citation year t_m, B is computed as the cumulative gap between a hypothetical linear citation trajectory connecting (t₀, 0) to (t_m, c_tm) and the article's actual year-by-year citation curve, summed across the years from publication to peak. An article that accumulates citations steadily produces B near zero; an article that stays near zero through most of the interval and then jumps sharply to its peak produces a large positive B. The metric was introduced as a parameter-free alternative to earlier sleeping-beauty definitions that required ad-hoc thresholds.²

The supporting trajectory is built from the publication years of citing articles. For each indexed article whose reference list contains the target, the citing article's publication year increments the target's per-year count. The result is a year-by-year citation timeline from publication to the present. Peak year t_m is the year with the highest per-year count; awakening year is the first year in which the per-year count rises above 20% of the way from the article's pre-awakening average to its peak and then stays above that threshold for at least two consecutive years. Sleep duration is the year gap between publication and awakening.

A naive application of the Beauty Coefficient to internal-only citation data produces a large class of false positives: articles that are heavily cited in books, dissertations, and unindexed journals look like they are sleeping when measured against the indexed corpus alone, but are not actually sleeping in the wider scholarly conversation.³ Flower and Hayes's 1981 "A Cognitive Process Theory of Writing" is the canonical example: it has accumulated more than 6,000 global citations across its lifetime but only 13 internal citations inside the indexed journals, and a naive B computation would flag it among the top sleepers. To prevent this, the implementation scales B by the ratio of internal to total citations (internal plus the CrossRef-deposited global count) when the global count substantially exceeds the internal count, which effectively removes from the ranking any article whose apparent dormancy is a coverage artefact rather than a genuine delay.

Two limitations bear on interpretation. The metric requires citation history to compute, which means articles published in the last five to ten years rarely appear — they have not had time to either sleep or wake. The false-positive correction depends on CrossRef's global citation count, which is itself incomplete for older articles, especially work published before reference deposit became standard practice. A high B score here therefore means delayed recognition within the indexed corpus; the per-row display of both internal and global citation counts is the user's hook for assessing whether any given case is a genuine sleeper or a borderline coverage artifact.

Controls

Four controls scope the ranking. Filter changes do not auto-recompute; the Compute button is required because the per-article timeline reconstruction is the slow operation in the pipeline.

The minimum total citations slider filters which articles are eligible for ranking. At the default of five citations, an article must have accumulated at least five lifetime citations (internal plus global) to be considered. Lowering the threshold admits articles with very thin citation histories whose Beauty Coefficient is statistically unstable; raising to ten or fifteen tightens the ranking to articles whose late surge is well-documented enough to interpret confidently.

The Published from / to year filters restrict the candidate set by publication year. Restricting to 1990–2010 surfaces sleepers from the era when the field was establishing its current research fronts — articles whose late awakening tracks specific disciplinary turns — while a 2010–2020 filter mostly returns empty results because those articles have not had time to either sleep or wake.

The Journals picker narrows the candidate set to a chosen subset. Restricting to a single venue answers "what work in this journal had a delayed reception?", which often surfaces articles that pre-figured the journal's later editorial direction.

Each row's expandable detail panel renders the year-by-year citation timeline as a small bar chart. The shape is part of the analysis: a flat-then-jump curve confirms a Beauty pattern, while a curve with intermittent activity throughout the sleep period suggests an article that was always trickling but never canonical.

References

1. Ke, Q., Ferrara, E., Radicchi, F., & Flammini, A. (2015). Defining and identifying Sleeping Beauties in science. Proceedings of the National Academy of Sciences, 112(24), 7426–7431.
2. van Raan, A. F. J. (2004). Sleeping Beauties in science. Scientometrics, 59(3), 467–472.
3. Glanzel, W., & Garfield, E. (2004). The myth of delayed recognition: Citation analysis demonstrates that papers recognized as "important" are rapidly cited. The Scientist, 18(11), 8–9.

Tool orientation

The Journal Citation Flow chord diagram aggregates the article-level citation graph up to the journal level. Each cell in the underlying 36×36 matrix counts the number of times articles published in journal A cite articles published in journal B. The chord diagram renders that matrix visually: each arc on the ring is one journal, and each chord connecting two arcs has two ends whose widths show the citation flow in each direction. Where a chord is markedly wider at one end, the citation relationship is asymmetric — one journal cites the other more than it is cited in return.

Run with default parameters across the 36 indexed journals, the matrix carries 44,640 inter-journal citations of which 20,530 (46.0%) are self-citations — articles citing other articles in the same journal. Among cross-journal flows the heaviest single edge runs from Journal of Technical Writing and Communication to Technical Communication Quarterly at 1,157 citations, followed by TCQ to Journal of Business and Technical Communication at 1,065, JBTC to TCQ at 881, and Written Communication to Research in the Teaching of English at 788. Ranked by total outbound flow the top three citing journals are Computers and Composition (5,672 outbound citations), TCQ (5,119), and IEEE Transactions on Professional Communication (4,227); ranked by inbound flow the top three are TCQ (6,341 inbound), Computers and Composition (4,632), and JBTC (4,337). TCQ is the field's structural attractor: a citing source for many venues and the most-cited target overall.

Use the tool to identify dominant citing-cited dyads, locate hubs that pull citations from across the field, and read off how concentrated or dispersed each venue's citation footprint is. The 46% self-citation rate is itself the most important number on the diagram: it indicates that nearly half of all internal citation traffic stays within the same venue rather than crossing journal boundaries.

Methodology

The matrix is a journal-level aggregation of the article-level citation graph. For each indexed article whose reference list contains another indexed article, the citing article's journal is rolled up to the matrix row and the cited article's journal to the matrix column. Self-citations appear on the diagonal as chord loops on the visualization. The full matrix preserves directionality — the flow from journal A to journal B is recorded separately from the flow from B to A — which lets the chord diagram surface asymmetric relationships rather than averaging them out.

Three structural features of the diagram bear interpretation. First, the high self-citation rate is partly editorial and partly topical: authors publishing in a venue tend to engage with that venue's prior conversation, and editors often value continuity with the journal's existing scholarship. Second, the strongest cross-journal flows almost all run within the technical-communication sub-cluster (JTWC↔TCQ↔JBTC↔IEEE TPC), which reflects a tightly integrated research community in which articles routinely cite across the four flagship TC venues but rarely reach into composition journals or rhetoric journals. Third, asymmetric chords often signal a status or directional borrowing relationship; a chord that is much wider at journal A's end than at B's end means A is reading B more than B is reading A.

Two limitations bear on interpretation. The raw counts are not normalized for journal output volume.¹ A large journal that publishes 200 articles per year generates more citing flows simply because its articles have more reference list entries; comparing raw flows across journals of different sizes can misrepresent the relative intensity of inter-journal engagement. Normalization by article count or by total reference-list length would correct for this but is not currently exposed in the UI. The second limitation is bibliometric scope: journals that do not deposit reference lists with CrossRef appear as invisible citing sources, though they may still appear as cited targets when other journals cite their articles. The Coverage Notes page at /coverage documents which journals contribute citing signal and which do not.

Controls

Four controls scope the matrix. Filter changes do not auto-recompute; the Compute button re-runs the aggregation against the current filter state.

The minimum-citations slider filters which chords are drawn. At the default of one, every nonzero flow is rendered; raising to 25 or 50 strips out the long tail of light flows and produces a visualization dominated by the major axes — useful when the question is "where does citation traffic concentrate?" rather than "what is the full topology?"

The From / To year filters restrict the citation set by the citing article's publication year. Restricting to 2015–2025 reframes the question from "what is the all-time flow pattern?" to "how is the field citing across journals now?" and tends to amplify the technical-communication sub-cluster relative to the all-time view, because the recent decade has tightened internal TC citation traffic.

The Journals picker narrows the matrix to a chosen subset of journals on both axes. Restricting to the four TC venues isolates that sub-cluster and shows its internal flow pattern with cross-cluster citations dropped; restricting to the rhetoric venues exposes a substantively different sub-graph in which self-citation and cross-cluster citation patterns shift.

References

1. Leydesdorff, L., & Rafols, I. (2011). Indicators of the interdisciplinarity of journals: Diversity, centrality, and citations. Journal of Informetrics, 5(1), 87–100.
2. Borner, K., Klavans, R., Patek, M., Zoss, A. M., Biberstine, J. R., Light, R. P., Lariviere, V., & Boyack, K. W. (2012). Design and update of a classification system: The UCSD map of science. PLoS ONE, 7(7), e39464.

Tool orientation

Citation Half-Life reports two complementary medians for each indexed journal: how far back in time the journal's authors typically reach when citing earlier work (citing half-life), and how long the journal's own articles continue to attract citations from later work (cited half-life). A short citing half-life signals a fast-moving research front where recent work dominates the reference lists; a long cited half-life signals a journal whose articles have lasting influence on the field's conversation.

Run with default parameters across the indexed corpus, the citing half-life ranges from 5 to 10 years across journals with sufficient citation history. College Composition and Communication and Argumentation have the longest citing half-lives at 8 years, indicating that their authors regularly reach back nearly a decade for sources. Rhetorica, Rhetoric & Public Affairs, and Research in the Teaching of English sit at 5 years, reading recent work more heavily. Cited half-lives spread further, from 2 years (Literacy in Composition Studies) to 10 years (College Composition and Communication): CCC is the longest-tail venue in the index, a journal whose articles continue to be cited by other journals roughly a decade after publication, which is the bibliometric signature of a venue with deep canonical authority.

Use the tool to identify which journals serve as the field's long-memory venues (long cited half-life), which journals operate as fast-moving research fronts (short citing half-life), and where the two metrics disagree — a venue with short citing but long cited half-life is reading new but being read for a long time, which is the structural fingerprint of a flagship journal.

Methodology

Citation half-life was introduced by Garfield as a standard bibliometric indicator and is reported annually for journals indexed in Journal Citation Reports.¹ The construction here follows the standard definition. For each citation edge in the indexed citation graph where both the citing and cited articles carry publication years, the age of the citation is computed as the citing year minus the cited year. The ages are grouped by journal — by the citing journal for the citing half-life, by the cited journal for the cited half-life — and the median age in each group is reported as the half-life. The interquartile range Q25–Q75 is reported alongside the median as a shaded band, since the median alone can hide a long-tailed distribution.

The citing-cited distinction matters analytically. Citing half-life describes the journal's citing behavior: it answers the question "when this journal's authors cite something, how old is the cited work, on average?" It is largely under the editorial and methodological control of the journal — a venue that publishes articles drawing heavily on classical rhetoric will have a longer citing half-life than a venue that publishes work on emerging technologies. Cited half-life describes the journal's longevity in the field's reference lists: it answers "when other journals cite this journal's articles, how old are those articles?" A long cited half-life is the bibliometric signature of canonical work that continues to be referenced; a short cited half-life can indicate either a young journal whose articles haven't had time to age, or a venue whose articles attract attention briefly and are then displaced.

Three structural observations bear on interpretation. Humanities journals run longer half-lives than STEM journals on average because theoretical and historical scholarship retains relevance over decades.² Within this corpus, rhetoric venues that engage with classical and historical material tend to extend toward the long end of both metrics, while technology-focused venues such as Communication Design Quarterly and IEEE Transactions on Professional Communication sit on the shorter end as their topical landscape shifts. The 46% self-citation share documented in the Journal Citation Flow tool is also relevant: half of the citations contributing to each cited half-life come from articles in the same journal, which means cited half-lives index internal canonical-status as much as cross-journal recognition.

Two limitations bear on interpretation. The metric counts only internal citations within the indexed corpus, which biases the citing half-life shorter than its true value — references to older foundational books and to journals outside the index are not visible, and those references are precisely the older end of the typical reference list. Journals with few resolved citations produce unstable medians, which is why the rendering shows the per-journal sample size and dims the bar for venues below the reliability threshold. Journals indexed via RSS or web scraping deposit no reference lists, so they contribute no citing-side data and appear only on the cited side.

Controls

Three controls scope the analysis. Filter changes do not auto-recompute; the Compute button re-runs the aggregation against the current filter state.

The From / To year filters restrict the citation set by the citing article's publication year. Restricting to 2015–2025 reads the half-lives off recent citation behavior only; this often shortens both metrics across the board because recent citation traffic skews toward recent work, and shortens the citing half-life of journals whose authors have shifted toward newer source material.

The Journals picker narrows the analysis to a chosen subset. Restricting to a small cluster (the four flagship TC venues, for example) computes both metrics on the within-cluster citation graph, which removes cross-cluster citations from the age distribution and produces sub-cluster-internal half-lives that often differ from the all-journals values.

The citing/cited toggle in the chart toolbar switches between the two metrics on the same set of journals; switching is instant and requires no recomputation.

References

1. Garfield, E. (1972). Citation analysis as a tool in journal evaluation. Science, 178(4060), 471–479.
2. Moed, H. F. (2005). Citation analysis in research evaluation. Springer.

Tool orientation

Main Path Analysis traces the single most important chain of articles connecting today's writing-studies work to its foundational sources. Think of the citation network as a river system: every article cites earlier ones, every newer article gets cited by someone, and most citations are tributaries. The main path is the trunk, the route along which the most intellectual traffic flows.

Run with default parameters against the full Pinakes index, the tool returns a 25-article chain that begins with Agboka and Dorpenyo's 2022 "Curricular Efforts in Technical Communication After the Social Justice Turn" at the frontier and ends with Carol Berkenkotter's 1981 "Understanding a Writer's Awareness of Audience" at the foundation. Between those endpoints sit Russell's 1997 "Rethinking Genre in School and Society" (the most-cited article on the entire path, with 129 citations inside the index), Berkenkotter and Huckin's 1993 sociocognitive-genre piece, and Spivey's 1990 work on text transformation. The path threads through four decades of genre theory, activity systems, and professional communication research; reading it from bottom to top is one compressed way to follow how the field arrived at its current concerns.

Use the tool to ask: what work is the writing-studies citation graph most actively channelling? Which articles, if removed from the index, would disrupt the field's transmission of ideas most?

Methodology

Main path analysis was introduced by Hummon and Doreian to study the development of DNA theory and has since been extended into a standard bibliometric instrument for tracing knowledge flow through citation networks.¹ Where centrality measures score individual nodes, main path analysis scores edges and traces the single highest-traffic route through the graph. The tool here implements the Search Path Count (SPC) variant introduced by Liu and Lu, which computes path counts more efficiently than the original Search Path Link Count and corrects an asymmetry in the original procedure.²

Construction proceeds in four steps. First, the citation graph is filtered to articles that meet the minimum-citations threshold and any active year-range or journal scope; the result is loaded as a directed graph with edges flowing from citing (newer) to cited (older) work. Second, any cycles produced by publication-date inversions or imprecise pub dates are removed by deleting the cycle edge whose source-target year gap is largest in the wrong direction. Third, articles with in-degree zero are designated as sources (the field's frontier) and articles with out-degree zero as sinks (foundational works that cite nothing else inside the index). Fourth, for each edge (u, v) the tool computes SPC(u, v) as the product of (number of source-to-u paths) and (number of v-to-sink paths) using two passes over the topologically sorted graph.

The main path is then extracted greedily: starting from the source incident to the highest-SPC outgoing edge, the algorithm follows the highest-SPC edge at each step until it terminates at a sink. Running the tool against the Pinakes index with default parameters produces a DAG of 726 articles and 2,279 edges, with 118 sources, 218 sinks, and three cycles removed; the resulting main path has 25 articles, with the heaviest single edge carrying an SPC of roughly 1.69 million source-to-sink paths.

Two limitations bear on interpretation. The path is corpus-bounded: an article that is foundational in the wider scholarly world but rarely cited inside the indexed journals will not appear, and the foundational end of the path tilts toward whichever sub-area happens to have deeper internal citation chains in the corpus. The greedy extraction, second, returns a single backbone rather than the family of high-traffic routes; the Datastories First Spark tool addresses this gap by extracting the top-K SPC routes and reporting their convergence points.

Controls

Four controls scope the DAG that the tool computes against. They sit above the chart area and require an explicit Compute click to run.

The minimum-citations slider filters which articles enter the graph at all. At the default of two internal citations, an article must be cited by at least two other articles inside the indexed journals to be considered. A lower setting (one) admits more peripheral work and tends to lengthen the path. A higher setting (five or above) carves the network down to the heavily-cited core, producing a shorter and steeper path through the field's most central nodes.

The year-range filters restrict the DAG by publication year. A tight range produces a sparser graph with fewer cross-decade traversals, which often shortens the path but can surface era-specific backbones; the 1990s genre-theory cluster becomes legible when the range is set to 1985–2005. A wide range threads paths across the full historical span.

The journals picker narrows the corpus to a chosen subset. Restricting to a cluster (the technical-communication venues, for instance, or the rhetoric-journal cluster) reframes the question from "what is the field's backbone?" to "what is this sub-area's backbone, computed from citations that stay inside the cluster?" Cross-cluster citations are dropped along with their endpoints; the resulting path may be shorter and less interconnected than the full-corpus version.

Filter changes do not auto-recompute. The Compute click is required because main path analysis on the full DAG takes several seconds and would otherwise re-fire on every checkbox toggle.

References

1. Hummon, N. P., & Doreian, P. (1989). Connectivity in a citation network: The development of DNA theory. Social Networks, 11(1), 39–63.
2. Liu, J. S., & Lu, L. Y. Y. (2012). An integrated approach for main path analysis: Development of the Hirsch index as an example. Journal of the American Society for Information Science and Technology, 63(3), 528–542.

Tool orientation

The Temporal Network Evolution tool slices the indexed citation graph into time windows and reports a panel of structural metrics for each slice. The visualization renders those metrics as time series, which lets the analyst read off the field's growth, consolidation, and fragmentation phases at a glance — whether the network is becoming more cohesive over time, whether modular structure is intensifying or weakening, and whether the field's intellectual chains are becoming longer or shorter.

Run with default parameters — minimum one citation, five-year windows — the tool returns 18 windows from 1939–1943 to 2024–2028. The cumulative graph at the most recent window contains 6,673 articles and 22,930 citations. Three structural movements are legible. First, the giant component fraction rises from roughly 0.10 in the 1984–1988 window to 0.58 in the 2019–2023 window: across forty years the field has knitted itself into a cohesive citation network where the largest connected component now contains the majority of indexed articles. Second, average degree rises from 0.64 in 1984–1988 to 1.94 in 2019–2023, reflecting a citation graph that is densifying. Third, modularity Q falls from 0.907 in 2009–2013 to 0.820 in 2019–2023; the field remains modular, but cross-cluster citation traffic has grown enough to soften the boundaries between research fronts.

Use the tool to identify structural turning points, distinguish growth phases (rising node and edge counts) from consolidation phases (rising giant-component fraction or falling modularity), and locate the windows in which today's research fronts crystallized.

Methodology

For each time window the tool builds a directed subgraph from articles published within that window and the citation links between them. An edge enters the window when the citing article's publication year falls inside the window. A cumulative-view alternative aggregates the subgraphs into a growing graph that tracks total network size over time. The metrics are then computed on each window's graph using standard network-science definitions.¹

Density is the fraction of possible directed edges that actually exist; in citation networks density typically falls as the network grows because the number of possible edges grows quadratically while the number of actual edges grows roughly linearly. Average degree reports the mean number of citations incident to each node, which avoids the density confound and is more interpretable for growing networks. Transitivity (the network-wide clustering coefficient) measures the tendency of citation triangles to close. Giant component fraction is the share of nodes in the largest connected component, a direct indicator of network cohesion.² Modularity Q from a Louvain partition measures how strongly the network decomposes into communities and is computed only when the window contains at least ten nodes. Average path length in the giant component reports the mean number of citation hops between articles, which indicates how many steps of intellectual transmission typically separate any two pieces of work.

The trajectory of these metrics across windows is the analytic move. Rising node and edge counts mark a growth phase; the typical signature is dropping density and rising giant-component fraction simultaneously, since the network is adding articles faster than it adds edges but is also stitching itself together topologically. Falling modularity in the presence of stable or rising average degree is the signature of consolidation: cross-community citation traffic is growing relative to within-community traffic, which softens the partition. The 2019–2023 window in this corpus shows exactly that pattern.

Two limitations bear on interpretation. The metrics are sensitive to window size: annual windows produce maximum temporal resolution but very noisy metrics for years with few articles, while five-year windows smooth the signal at the cost of hiding short-lived structural changes. The most recent window in any computation is partial — the 2024–2028 window currently contains only 127 articles because most of its years have not yet been indexed — and its metrics should be read as in-progress rather than as a stable terminal value. The metrics are also corpus-bounded: a structural phenomenon visible in citation networks that include external scholarship may not be visible here.

Controls

Four controls scope the analysis. Filter changes do not auto-recompute; the Compute button re-runs the per-window metric computation against the current filter state.

The minimum-citations slider filters which articles are eligible to enter the per-window graph. At the default of one citation the tool admits any article that has at least one incident citation; raising the threshold tightens each window's graph to its more-cited core, which usually shortens average path length and inflates giant-component fraction.

The Window slider selects the temporal granularity from 1 year to 10 years. One-year windows are the right choice when the question is about specific year-on-year transitions; five-year windows smooth the noise and make multi-decade trajectories legible; ten-year windows produce a coarse-grained reading of long-run structural change.

The From / To year filters restrict the analysis to a chosen historical span. Restricting to 1995–2025 strips away the very early years where the indexed citation graph is too sparse to produce stable metrics; restricting to 2010–2025 isolates the recent consolidation phase.

The Network Snapshot mode replaces the time-series view with a force-directed graph at any single year, accessed through the year slider in the snapshot view. Use snapshot mode to inspect the network's actual topology in a window of interest after time-series analysis has identified that window as structurally distinctive.

References

1. Newman, M. E. J. (2010). Networks: An introduction. Oxford University Press.
2. Barabási, A.-L. (2016). Network science. Cambridge University Press.
3. Price, D. J. de S. (1965). Networks of scientific papers. Science, 149(3683), 510–515.

Tool orientation

The Institutions tool reports article output by author affiliation across the indexed corpus. Each row in the bar chart credits one institution with the count of indexed articles on which any of its affiliated scholars appears as an author; the line chart tracks the same counts year over year for the top ten institutions, which makes individual programs' rises and plateaus legible against the field-wide baseline.

Run with default parameters across the indexed corpus, the bar chart returns Pennsylvania State University at the top with 249 indexed articles, followed by the University of Texas at Austin (239), Texas Tech University (219), Arizona State University (210), Iowa State University (182), Purdue University (178), and Carnegie Mellon University (178). The top 25 is dominated by large public research universities, reflecting the long-standing concentration of writing-studies and technical-communication PhD programs at land-grant and flagship public institutions; the only private universities in the top 25 are Carnegie Mellon, Rensselaer Polytechnic Institute, and Texas Christian. The line-chart view of the top ten over time shows that this distribution is not static: Penn State, Arizona State, and Texas Tech have each gone through multi-year publication peaks tied to specific program-building moments, while older programs at the University of Iowa and Carnegie Mellon show flatter, longer-tail trajectories.

Use the tool to identify the institutional concentration of the field's published output, locate program-building peaks tied to particular years, and read off how institutional output complements rather than tracks scholarly individual output.

Methodology

Institutional affiliations are retrieved from OpenAlex, an open bibliographic database that aggregates author affiliations from publisher metadata and disambiguates them against the Research Organization Registry (ROR) where possible.¹ For each indexed article that carries a CrossRef DOI, the fetcher queries OpenAlex's per-article record and pulls out every author-level affiliation listed at the time of publication. Institutions are stored by their canonical OpenAlex names with ROR identifiers retained where present.

Attribution uses a whole-count method: an institution is credited with one full article every time any of its affiliated authors appears on the byline.² A multi-authored article whose authors hold three different affiliations therefore contributes one full count to each of the three institutions, not a fractional count. The alternative is fractional counting, in which the credit for an article is divided among its co-authors' institutions, but whole counting is the standard approach in institutional research-output analysis and is more directly interpretable. The trade-off is that whole counting favors institutions that collaborate widely — an article co-authored with someone at another institution counts fully for both institutions, which tends to inflate the rankings of universities with many active collaborators.

Three structural observations bear on interpretation. The OpenAlex affiliation graph is incomplete and uneven: some authors carry no affiliation in the publisher metadata, especially in older articles, and OpenAlex's disambiguation can split one institution across multiple ROR entries (the index currently lists "University of Minnesota" and "University of Minnesota System" as separate institutions, both attributable to the same actual program). Institutional mergers and name changes can cause similar splits. The second observation is that articles indexed via RSS or web scraping rather than CrossRef carry no OpenAlex record at all, which means institutions associated with those journals are systematically undercounted; the Coverage Notes page at /coverage documents which venues fall in this category. The third observation is that authors who change institutions mid-career are credited to whichever institution OpenAlex records for each specific article, which means an author's full institutional career is distributed across the chart by date, which is almost always the desired behavior but occasionally produces small artifacts when the affiliation data is imprecise.

Two limitations bear on interpretation. The chart reflects affiliation as recorded in publisher metadata, which is precisely the source whose imprecision the methodology relies on. And the chart is silent about scholars who publish primarily in books, edited collections, or unindexed journals, which means an institution with a strong book-publishing tradition will appear smaller here than its actual research output warrants.

Controls

The Institutions tab does not currently expose interactive filters — the bar chart and the line chart render against the full indexed corpus and update only when the underlying affiliation data is refreshed.

Both charts use the same whole-count attribution. The bar chart caps display at the top 25 institutions to keep the chart legible; the line chart caps at the top 10 because more series than that produce a visually unreadable tangle. Hover any bar to read the institution's full count; hover any line in the time-series chart to read its year-by-year trajectory.

The Coverage Notes link in the tab intro is the natural follow-up: it documents which journals contribute affiliation data and which do not, and it is the right place to assess whether a particular institution's apparent ranking is plausible given the indexed venues' affiliation deposit rates.

References

1. Priem, J., Piwowar, H., & Orr, R. (2022). OpenAlex: A fully-open index of scholarly works, authors, venues, institutions, and concepts. arXiv preprint arXiv:2205.01833.
2. Waltman, L. (2016). A review of the literature on citation impact indicators. Journal of Informetrics, 10(2), 365–391.

Tool orientation

The Reading Path tool builds a structured reading list around a seed article. The user picks any indexed article as the seed; the tool then assembles four citation-neighborhood sets — what the seed cites, what cites the seed, what is co-cited with the seed, and what shares references with the seed — and ranks every article in the union of those sets by a composite relevance score. The output is both a tabular reading list and a force-directed graph centered on the seed, which lets the user traverse from a known starting point into the surrounding citation neighborhood.

Run with Jones, Moore, and Walton's 2016 "Disrupting the Past to Disrupt the Future" as the seed, the tool returns 222 unique articles spanning the full neighborhood: the seed cites 57 indexed articles in its own reference list, 146 indexed articles cite the seed in return, 20 articles are co-cited with the seed by other indexed work, and 20 articles share references with the seed. The composite-ranked reading list surfaces Cecilia Shelton's 2020 "Shifting Out of Neutral," Walton's 2016 "Supporting Human Dignity and Human Rights," Jones's solo 2016 "The Technical Communicator as Advocate," Frost's 2016 "Apparent Feminism as a Methodology for Technical Communication," and Cox's 2019 "Working Closets" at the top — each appears in multiple relationship sets and accumulates score across them. The companion graph renders 223 nodes and 243 edges with the seed at center, color-coded by relationship type.

Use the tool to assemble a structured first reading list when entering a new sub-area, locate articles that bridge a known seed to its surrounding citation neighborhood, and surface articles that connect to the seed through multiple distinct relationship channels rather than one.

Methodology

The four relationship sets are computed independently against the indexed citation graph. Backward citations are articles the seed cites — the resolved DOIs in the seed's CrossRef-deposited reference list that match other indexed articles. Forward citations are articles that cite the seed — the inverse query, returning every indexed article whose reference list contains the seed's DOI. Co-citation neighbors are articles X for which at least one indexed article C cites both X and the seed; the strength of the co-citation tie is the count of citing articles that pair them.¹ Bibliographic coupling neighbors are articles X whose reference list shares at least one indexed article with the seed's reference list; the coupling strength is the count of shared references.² The co-citation and coupling sets are each capped at the top 20 neighbors by tie strength to keep the reading list focused on substantively related work rather than long-tail noise.

The composite relevance score combines the four signals into a single ranking. Each related article receives +2 for a direct citation link (either backward or forward), +weight for any co-citation tie (the count of citing articles, capped at 5 to prevent any single dominant pair from saturating the score), +weight for any bibliographic coupling tie (the count of shared references, capped at 5), +1 for a shared topic tag with the seed, and +1 for publication in the same journal as the seed. Articles that appear in multiple relationship sets accumulate score across them, which is the analytic move: an article that cites the seed, is co-cited with the seed in other articles, and shares topic tags with the seed scores higher than an article connected through any single channel, even if that channel is heavier.

This is a hybrid recommendation strategy. Pure citation-based recommendation favors directly-linked articles but misses peripheral work that the field treats as related; pure co-citation favors perceived kinship but misses an article's own intellectual debts; pure coupling favors shared inputs but misses the field's reception of the seed.³ The composite score is constructed to give each channel a chance to surface its strongest neighbors and then rewards articles that appear across channels. The +1 topic-tag and same-journal weights are deliberately small — topical similarity is a useful tie-breaker between citation-equivalent articles but should not override the citation evidence.

Two limitations bear on interpretation. The reading list is bounded by the indexed corpus: an article that the seed cites in its reference list but that is not itself indexed will not appear in the backward-citations set, and the same is true for any external work that should plausibly join the reading list. Articles from journals that do not deposit reference lists with CrossRef have no backward-citation data and limited co-citation and coupling data, which means using such an article as a seed produces a reading list weighted heavily toward forward citations only. The Coverage Notes page at /coverage documents which venues fall in this category.

Controls

The tool exposes one primary input and several display controls.

The Search for a seed article input drives everything else. The user types a title fragment, an author name, or a keyword; the autocomplete dropdown returns matching indexed articles ranked by recency and citation count. Selecting an article triggers the full neighborhood query and renders the seed card, the reading list table, and the relationship graph.

The relationship-set toggles in the table header let the user filter the reading list to a single relationship type or any combination of the four. Toggling off "Backward citations" restricts the list to articles that cite the seed or share intellectual neighborhoods with it, which is the right move when the question is "what came after this article and is related to it?" rather than "what is this article reading?"

The graph view renders the relationship graph with color-coded edges (blue for backward, green for forward, orange for co-citation, purple for coupling). Nodes that appear in multiple relationship sets carry multi-color borders. Node size scales with composite relevance score; click any node to switch the seed to that article and re-run the neighborhood query against the new seed. This recursive traversal is the primary use of the graph view: starting from a known anchor, the user can navigate outward through whichever relationship channel they want to follow.

References

1. Small, H. (1973). Co-citation in the scientific literature: A new measure of the relationship between two documents. Journal of the American Society for Information Science, 24(4), 265–269.
2. Kessler, M. M. (1963). Bibliographic coupling between scientific papers. American Documentation, 14(1), 10–25.
3. Boyack, K. W., & Klavans, R. (2010). Co-citation analysis, bibliographic coupling, and direct citation: Which citation approach represents the research front most accurately? Journal of the American Society for Information Science and Technology, 61(12), 2389–2404.