PaperlayerView exact revision on arXiv

Source-linked AI summary

Network Analysis in Communication Research: Research Topics, Knowledge Organization, and Research Practices

Pengjia Cui

arXiv:2609.30905v1cs.SIstat.OT

TL;DR

The review asks how network-analysis research connects communication problems, knowledge, and research practices. It combines bibliographic comparison with purposive reading of selected studies, finding task-dependent relevance and differentiation within shared settings, while some media-use influence findings remain unresolved.

  • Problem

    Communication network research spans information access, participation, and public meaning, raising the question of how its connections enter individual studies’ explanations and findings.

    Source: §1 · pp. 1–2

  • Method

    The review combines a co-citation resource comparison with purposive reading of selected studies, comparing communication questions and explanations with observations and findings.

    Source: §2.3 · p. 4, §4.6 · p. 13

  • Results

    Connections’ relevance depends on task and value criterion, shared outlets and issue agendas can accommodate differentiated choices and evaluations, and media-use influence findings remain unresolved despite models of selection and content co-nomination.

    Source: §2 · p. 2

  • Takeaways & Limitations

    The review links organizational, journalistic, audience, and interpersonal research by identifying where evidence complements or qualifies other findings and where explanations remain open.

    Source: §2 · p. 2

    Citation-context functions were examined in two articles, without an intercoder-reliability estimate, and were not coded across abstract-reviewed sets.

    Source: §2.3 · p. 4

Abstract

from arXiv · show

Communication network research explains access to information, patterns of participation, and the organization of public meaning through different relationships and observations. This integrative review connects research topics, knowledge organization, and research practices, using bibliographic analysis of a Web of Science candidate pool to guide selective reading. Three judgments emerge from the comparisons. First, the relevance of a connection depends on the task and the criterion of value: team members anticipate consulting colleagues whose expertise they recognize, while journalists distinguish monitoring sources, using their information, and citing them. Second, commonality at one level can coexist with differentiation at another: audiences share outlets while selecting different articles, and shared issue agendas accommodate different evaluations. Third, some differences remain unresolved: contrasting media-use influence findings cannot be explained simply by whether models include selection and content co-nomination. Reading the uses of homophily, transactive memory, sourcing, agenda-setting, and framing resources clarifies which expectations and observations support these judgments. Selected citation contexts also show how conceptual and measurement resources enter the same argument. The resulting agenda calls for comparisons of relationship types across group stages, source use across reporting tasks, and encountered content with recipients' interpretations. These purposive comparisons establish specific connections among literatures without estimating their prevalence or demonstrating a common causal mechanism.

1 Introduction

Prior reviews connect network relations and literatures, while this review asks how those connections enter individual studies’ explanations and findings. It compares information access and media use across communication settings.

1 Introduction

  • Recognized expertise and reporting tasks give access to relevant information different significance across organizational knowledge seeking and journalistic sourcing.Media-use and audience studies also relate social affiliation to content choice at different levels; the review distinguishes findings that qualify one another from differences in separate outcomes.Source: §1 · p. 1
  • The review asks which communication topics network-analysis studies address and how those topics connect.Earlier reviews distinguish relation types and trace connections across topics and subfields, prompting the question of how connections shape individual studies’ explanations and findings.Source: §1 · pp. 1–2

2. How is this literature organized through its publication context and shared knowledge resources?

The review compares complementary findings and unresolved differences across questions, explanations, observations, and concepts. Its three judgments distinguish task-specific value, differentiation within commonality, and unresolved media-use influence findings.

2. How is this literature organized through its publication context and shared knowledge resources?

  • Bibliometric maps describe relations among literature, not communication among the actors studied; centrality and graph properties depend on represented relations.Figure 1’s outer connections relate questions to explanations, problems to observations, and concepts to measurement; dotted connections inform synthesis, not research stages.Source: §2 · p. 2, Figure 1 · p. 2
  • A connection’s relevance depends on the task and criterion of value; shared outlets or issue agendas can coexist with differentiated choices or evaluations.Media-use influence findings remain unresolved even when both studies model selection and content co-nomination.Source: §2 · p. 2

2 Review Design and Evidence Sources

The review combines a bounded Web of Science candidate pool with bibliometric description, purposive reading, and comparisons of questions, explanations, observations, and findings. Its evidence supports specific connections, not prevalence estimates or a recovered selection protocol.

2.1 Candidate corpus and retrieval boundary

  • The candidate pool contains 2,114 Web of Science records retrieved within the Communication category, but it is not an individually screened census of eligible studies.The export preserves original document-type strings; substantive inclusion requires an identifiable relation structure, network matrix, network model, or networked text structure addressing a communication question.Source: §2.1 · p. 3

2.2 Literature-level evidence and substantive reading

  • Bibliographic fields describe vocabulary and shared resources, while a conditional co-citation comparison tests selected reference pairs against frequency-preserving randomizations.The comparison tests all 435 possible pairs with Benjamini–Yekutieli adjustment; its four content cases were selected beforehand from the same pool, so the comparison is exploratory.Source: §2.2 · p. 3

2.3 Abstract calibration and selective full-text comparison

  • A purposive 40-record abstract audit calibrated eligibility, while selected full-text cases supplied evidence on questions, explanations, findings, and network construction.The audit retained 29 records and excluded 11, but its within-stratum selection rule and five-record quota rationale are undocumented; the evidence ledger also lacks a complete initial selection order and case-specific rules.Source: §2.3 · pp. 3–4

2.4 Synthesis procedure and interpretive scope

  • The synthesis compares explanations with observations and findings, treating agreements as comparable outcomes and apparent conflicts through differences in populations, relations, or timing.Citation-context functions were examined in two articles, without an intercoder-reliability estimate or coding across abstract-reviewed sets.Source: §2.4 · p. 4, §2.3 · p. 4

3 Research Landscape and Communication Problems

The selected literature organizes overlapping communication problems around participation, influence, meaning, and access. Its findings distinguish outcomes, relationships, sources, and interpretations rather than collapsing them into a single account.

3.8 Synthesis: overlapping communication problems

  • The 2,114-record candidate literature spans 1973–2026, and the selected reading groups studies into overlapping question families rather than corpus proportions.Across the families, common problems intersect through meaning, evaluation, and whose experiences are represented; the groupings do not assign each subfield an exclusive place.Source: §3.1 · p. 4, §3.2 · p. 5, Table 1 · p. 7

3.3 Public discussion, political disagreement, and collective action

  • Research on public discussion distinguishes rumor participation, cross-ideological brokerage, message selection, and the quality of conversation.Shin et al. (2017), Abul-Fottouh (2018), and Song et al. (2020) examine different participation questions; Moy and Gastil (2006) distinguish clarity, reasoning, and comprehension, with mixed findings.Source: §3.3 · p. 5

3.4 Information flows, influence, and social selection

  • Media-use studies follow relationships and preferences over time, while relationship-maintenance and support studies examine continuity and task-specific help.Friemel (2015, 2021) asks whether ties shape use, preferences shape ties, or both; Vriens and van Ingen (2018) follow core discussants, and Lee et al. (2019) consider information and tangible-help ties.Source: §3.4 · p. 5

3.5 Meaning, framing, and contested issues

  • Shared issues and texts accommodate competing frames, commenter evaluations, semantic differences, and identity-specific recognition rather than uniform public meaning.Schultz et al. (2012) report textual correspondence, not a causal effect on journalism; Hsiao and Hindman (2026) find commenters’ emphases differ from videos, while Kwon et al. (2009) and Kim (2023) examine semantic and identity differences.Source: §3.5 · pp. 5–6

3.6 Access to knowledge, sources, and public attention

  • Organizational and journalistic studies distinguish recognized expertise, communication roles, source access, and task-specific source use rather than one measure of importance.Generalist organizations brokered links in Shumate and Lipp (2008); journalists’ source positions and monitoring, use, and citation practices describe different parts of access and representation.Source: Journalism, source access, and news… · p. 6, Organizational knowledge,… · p. 6, §3.6 · p. 6

Audiences, media repertoires, and public attention

  • Shared audience outlets can coexist with differentiated article repertoires, while identity cues and audience definitions operate at distinct levels.Rauchfleisch et al. (2020) compare outlet sharing and article repertoires; Dvir-Gvirsman (2017) examines audience-identity cues, and Rusche (2024) distinguishes politician followers from unique users.Source: Audiences, media repertoires, and… · pp. 6–7

Discussion, norms, and health outcomes

  • Health and support findings vary by outcome and perceived network climate, linking discussion, norms, and relationships to distinct perceptions.Demetriades et al. (2024) report differing associations across knowledge and vaccination; Lee et al. (2022) distinguish future perceptions from social competence and self-perception.Source: Discussion, norms, and health outcomes · p. 7

Climate discourse, sources, and mobilization

  • Climate-science communication spans professional roles but shows uneven incoming attention; diverse discourse can draw on concentrated sources or unequal representation.Walter et al. (2019) do not establish sustained exchange; Veltri and Atanasova (2017) find concentrated professional news sources, and Stoddart et al. (2025) report Indigenous themes visible while Indigenous actors remain peripheral.Source: Scientific knowledge, policy, and… · p. 7, Climate discourse, sources, and… · p. 8

4 Theoretical Explanations and Knowledge Connections

This section examines how communication research connects theories, methods, and observations across relationships, participation, public meaning, organizational knowledge, and audience choice. It combines selected bibliographic associations with citation-context readings to clarify the scope and limits of these task-specific connections.

4.1 Influence, opinion leadership, homophily, and selection

  • Evidence about media influence and isolation depends on the relationships, information environments, and outcomes studied, rather than resolving influence versus selection alone.Barberá et al. (2015) report greater ideological segregation for political than several nonpolitical Twitter issues, while Bakshy et al. (2015) distinguish friend-shared, News Feed-displayed, and clicked content; neither dissemination nor clicks establish persuasion. Terren and Borge’s (2021) review associates digital-trace studies with clear or mixed echo-chamber evidence and self-report studies with mixed or no evidence, without explaining the disagreement causally. Health studies link relationships and norm interpretation to vaccination and coping, while Vriens and van Ingen (2018) describe reorganization rather than uniform withdrawal; Lee et al. (2019) distinguish information from tangible support, with help effectiveness unmeasured.Source: §4.1 · pp. 9–10

4.2 Bridging, collective action, and organizational contributions

  • Connections can extend access, organize participation, or contribute to collective reach, but their importance depends on how participants use and assess them.Weak ties need not be bridges or produce mobilization, and reported communication need not match attributed campaign importance.Source: §4.2 · p. 10

4.3 Agenda, framing, and contested public meaning

  • Agenda-setting and framing both study associations, but distinguish media–public correspondence from the interpretation and reception of represented meaning.Entman (1993) cautions that textual frames do not guarantee audience effects; network agenda setting proposes joint retrieval without passive copying of every association. Similar issue attention can coexist with divergent evaluations, as in genetic-testing testimony.Source: §4.3 · pp. 10–11

4.4 Organizational relationships and journalistic knowledge

  • Transactive memory makes recognized expertise relevant to anticipated consultation, while journalistic sourcing connects institutional access and routines to public representation.Palazzolo (2005) separates self-rated expertise, others’ recognition, and anticipated consultation; that comparison concerns consultation, not team performance. Tuchman (1978) and Zelizer (1993) foreground organizational routines and negotiated interpretations that following or co-mention networks alone cannot demonstrate.Source: §4.4 · p. 12

4.5 Audience affiliation, repertoires, and public attention

  • Audiences may share media outlets while choosing different articles, and shared use does not establish shared interpretations or attitudes.Repertoire accounts distinguish outlet combinations from selected news accounts; the selected studies examine audience affiliation and content choice separately, so their joint prevalence is unknown.Source: §4.5 · p. 12

4.6 Shared resources and co-citation

  • The primary co-citation view contains 30 reference-string identities and 329 edges, but citation co-occurrence does not demonstrate adoption of cited resources.Among 1,101 records citing a retained identity, 563 cite at least two and contribute edges; 213 records combine method and theoretical resources, producing 87 distinct pair edges.Source: §4.6 · p. 13
  • Method–theory pairs appear less often than expected, while method–method pairs appear more often, under the specified co-citation baseline.Method–theory combinations contribute 380 record–reference-pairs against 465.78 expected (observed/expected = 0.82); method–method combinations contribute 324 against 237.76 (1.36). Both have adjusted q ≈0.00073, and four alternative specifications retain these directions; citation patterns do not establish adoption or theoretical exclusion.Source: §4.6 · p. 13

4.7 Relational resources across communication questions

  • The four selected co-citation pairs recur 6.55–7.64 times as often as expected, but this bibliographic excess does not establish a common theoretical function.The within-year baseline preserves selected-string frequencies and records’ selected-reference counts; it cannot separate intellectual affinity from publication settings.Source: §4.7 · p. 15
  • Five records across two sets show uneven links: some explicitly integrate framing, topic modeling, and network methods, while others indicate proximity rather than a unified lineage.One record proposes discursive polarization, but its wording does not establish framing theory.Source: §4.7 · p. 15
  • Four co-citing sets span 18, 13, 12, and nine publication sources, and none disappears when one source is removed.Under joint year-and-source conditioning, excess co-citation remains for Entman–Gamson, Blei–Maier, and Barberá–Colleoni, but not Fraser–Habermas or the additional Entman–Maier pair at the 0.05 level. The auxiliary model fixes different margins from Curveball, so the change cannot be attributed solely to source composition.Source: §4.7 · p. 15

4.8 Shared resources across different research practices

  • Page-level reading shows the BP crisis article uses Entman’s 1993 work to define framing, Gamson and Modigliani’s 1989 work as background, and Schultz et al. (2012) for associative framing.The health-journalist study uses Bakshy and colleagues’ 2015 work to motivate homophily versus cross-cutting diversity and co-mention precedents, including the news-net, to represent source combinations (Zhang and Zhu, 2024, pp. 1661, 1664–1665).Source: §4.8 · p. 16

4.9 Synthesis: shared resources, differentiated uses

  • Explanatory resources specify distinct comparisons: recognized versus self-assessed knowledge, content selection versus shared attention, issue correspondence versus framing evaluations, and tie selection versus common preferences or tie-linked change.These distinctions identify which similarities matter and which differences need explanation.Source: §4.9 · pp. 16–17
  • Localized resource associations coexist with less aggregate method–theory pairing than the conditional baseline predicts; selected pairs do not explain that aggregate difference or its cause.Citation-context reading shows how framing definitions and associative-measurement resources, or information-diversity expectations and a co-mention precedent, can be assembled for particular research practices.Source: §4.9 · p. 17

5 Research Practices and Empirical Findings

This section compares communication-network findings and research practices, emphasizing how conclusions depend on the relationships, communications, populations, measures, and tasks examined. It synthesizes specific differences and associations without presenting a pooled estimate or establishing a jointly tested causal mechanism.

Message viewing and relational co-evolution

  • Relationship selection, continuity, and frequency reveal different aspects of communication, and more frequent discussion need not mean better deliberation.In Song et al. (2020), participants increasingly selected messages from same-candidate supporters over the campaign, while also selecting one another based on shared candidate qualities that did not necessarily imply support for the same candidate. Frequent social-media users had larger core discussion networks with both more turnover and more stable discussants, but changes in media use were not significantly associated with changes in talking frequency; Moy and Gastil (2006) report mixed associations between discussion frequency and talk quality, including negative relationships among adult-literacy respondents. These findings rely on bounded or self-reported measures and do not resolve causal direction or deliberative understanding.Source: Message viewing and relational… · p. 18, Everyday relationship continuity and… · pp. 18–19, Discussion networks and deliberative… · p. 19

A closer comparison within media-use research

  • Friemel’s contrasting media-use findings cannot be explained simply by whether models include selection and content co-nomination.The earlier study’s full models report no influence term significant at 5%, while the later study reports positive influence estimates for television and YouTube use. Group stage, relationship type, content opportunities, sample composition, and specification remain possible explanations; the studies do not establish which causes the contrast.Source: A closer comparison within media-use… · p. 18

Rumor diffusion, correction, and relational participation

  • Rumor participation, activist brokerage, and televised debate reveal distinct forms of public engagement, but their network ties do not establish interpersonal contact or audience effects.Shin et al. (2017) find that 65.48% of rumor-endorsing tweets were retweets, compared with 44.43% in a random political-tweet sample; among fact-checking-notice references initially coded as rejecting, 557 repeated the false claim and 225 (28.77%) explicitly said it was false. Abul-Fottouh (2018) distinguishes communication position from institutional affiliation, while Fuhse (2023) finds debate patterns broadly corresponding to governing, established-opposition, and outsider roles; close reading distinguishes supportive from adversarial interruptions.Source: Rumor diffusion, correction, and… · p. 19, Movement brokerage and organizational… · p. 19, Interactional alignment, conflict, and… · p. 20

Discussion networks, normative climates, and resilience

  • Health and resilience findings vary by communication topic and outcome, while cross-sectional reports cannot establish causal order or what partners exchanged.In Demetriades et al. (2024), closure was weakly associated with lower knowledge but did not distinguish vaccination status; the indirect pathway through injunctive norms was unsupported. In Lee et al. (2022), communication variables added explanatory information for some resilience perceptions, and partners for career goals or successes were associated with self and future perceptions and social competence, but no examined topic predicted structured style. The studies do not support a general prescription for a healthier network.Source: Discussion networks, normative… · p. 20

Everyday help, organizational roles, and family relationships

  • In Lee et al. (2019), deacons had higher betweenness but not significantly higher degree in information-support ties, and demographic mixing depended on including family ties.The symmetrized, cross-sectional reports identify whom members approach, not who provided useful help, whether ties were reciprocal, or how brokerage was enacted.Source: Everyday help, organizational roles,… · p. 20

Concept associations and actor–frame configurations

  • Text-based studies distinguish concept association, represented interpretation, and audience response rather than treating them as equivalent.Schultz et al. (2012) find that news partly echoes BP releases’ associations while adding political, legal, and protest actors, whose roles depend on surrounding associations. Shin (2020) shows that two newspapers attach different frame combinations to the same politician’s position; Kwon et al. (2009) compare co-occurrence across seven declaration versions, with shared-term restrictions and translation choices limiting interpretation. Kim (2023) analyzes creators’ practices and viewers’ public interpretations, while climate studies connect public meanings to linked sources or coded images; these observations do not establish audience acceptance or causal effects.Source: Concept associations and actor–frame… · p. 21

Media agendas and users’ associative reinterpretation

  • Media–public issue correspondence coexists with differentiated media alignment and evaluation, without identifying a unique process behind the pattern.Vu et al. (2014) report positive media–public network correlations in all five years (r = .65– .87), but the public network measures aggregate co-prominence. Vargo et al. (2014) find supporter networks aligned with general-news and partisan-media networks in most of 17 weeks, alongside differing best-fitting media alignments. Hsiao and Hindman (2026) find shared issues but stronger partisan and negative evaluations in comments, including a shift in the target of a shared evaluative term. Their comparison cannot establish where comment associations originated or whether exposure changed attitudes.Source: Media agendas and users’ associative… · p. 22

Knowledge seeking and interorganizational communication

  • Expected knowledge seeking follows recognized expertise, while communication contact, attributed importance, and source monitoring capture different kinds of access.Palazzolo (2005) finds perceived expertise closely associated with consultation nominations, but self-rated expertise receives little support; nominations indicate likely rather than observed consultation. Taylor and Doerfel (2003) report a positive association between channel index and relationship-importance ratings (r = .78, p < .001), although the most connected partners need not receive the most campaign-importance nominations. Verweij (2012) finds that seeking access and attracting attention need not coincide among journalists; following records do not establish exchanges or source credibility.Source: Knowledge seeking and… · pp. 22–23, Mediated sources, organizational… · p. 23

Mediated sources, organizational links, and audience repertoires

  • Journalists’ following lists and reported sourcing overlap in major source groups but differ in prominence, with media peers supporting monitoring and other sources helping develop information or distinctive angles.Interviews also distinguish background use from explicit citation; their retrospective accounts, paired with following data collected about a year earlier, do not track a continuous production sequence or verify sources in published news. Reader contact is not necessarily sourcing: complaints sometimes supply leads but often do not develop into stories, and this does not establish uniform exclusion of citizens.Source: Mediated sources, organizational… · p. 23
  • German-speaking Swiss and German far-right communities both shared NZZ articles, but their selections emphasized different subjects; shared outlet use therefore did not imply the same article repertoire.Cross-media duplication likewise describes overlapping outlet use, not interpersonal contact; sharing is not verified reading, endorsement, or a complete record of news use.Source: Mediated sources, organizational… · p. 24, Audience composition, selection, and… · p. 24

Cross-role scientific address and differentiated expression

  • Scientists’ tweets to journalists, civil-society actors, and political actors contained more negative-emotion words than tweets to peers, while certainty words were significantly more frequent only toward political actors.Outgoing connections exceeded incoming connections in journalist and political subnetworks, but not clearly in civil-society comparisons; these directed ties and addressee-based text units do not establish reciprocity, dialogue, understanding, trust, or why expression differs.Source: Cross-role scientific address and… · p. 24

Audience composition, selection, and attitude change

  • Audience cues changed content choice, but did not significantly change ideological self-placement extremity; congruent content-and-audience cues increased perceived extremity relative to content information alone.Separately, strongly committed AfD followers were about 7% of unique party followers but 55–75% of most politicians’ followers, a contrast that does not directly evidence attitude change. These are distinct observations, not estimates for one audience population.Source: Audience composition, selection, and… · pp. 24–25

5.5 What the practice comparisons add

  • The review’s core comparisons distinguish original-study findings from its synthesis, which is neither a pooled estimate nor a jointly tested mechanism; broader cases add conditions involving health, support, and representation (Table 4).Those cases distinguish discussion content and perceived norms for health outcomes, continuity of support from its usefulness, and public representation from participation in producing it.Source: §5.5 · p. 25, Table 4 · p. 26

6 Cross-Problem Synthesis and Research Agenda

The core comparisons show that connection value depends on task and criterion, while shared attention can coexist with different choices and evaluations. Some influence findings remain unresolved; the review uses these contrasts to identify conditions for further study.

6.1 Three judgments from the core comparisons

  • Resource value is task- and criterion-dependent; common outlets or issue agendas can coexist with differentiated choices and evaluations, while influence findings remain unresolved despite controls for selection and co-nomination.The studies identify complementary or compatible observations, but the contrasting influence findings do not establish a difference between underlying effects.Source: §6.1 · p. 27

6.2 Connecting problems, knowledge resources, and findings

  • The comparisons distinguish compatible findings from unresolved contrasts, and show that theoretical resources shape how evidence is assessed rather than proving a shared mechanism.The co-citation context connects conceptual definitions, explanatory expectations, and measurement precedents, but intellectual proximity and substantive comparability require distinct evidence.Source: §6.1 · p. 27, §6.2 · p. 29

6.1 Three judgments from the core comparisons

  • Research on public voice must specify whose contribution and which outcome are assessed, while temporal claims require following relationships beyond a single nomination or event.The broader cases do not test a shared process linking participation, visibility, and recognition; continuity alongside turnover and changing brokerage concern persistence over time.Source: §6.1 · p. 27

6.3 From contact to communication consequences

  • Future studies should connect relationships with the content and understanding that pass through them, including common observation of friendship, media conversation, group development, and health discussion.The proposed designs compare perceived normative climate with partners’ standpoints and actual discussion, and follow requests, responses, and recipients’ assessments across settings.Source: §6.3 · p. 29

6.4 From shared content to agreement or disagreement

  • Shared issues or sources can sustain agreement and interpretive conflict; testing this requires linking encountered content to prior associations and later expressions while retaining issue-level overlap.Semantic associations, represented standpoints, interactional reading, and audience evidence address different parts of how content is produced, circulated, and interpreted; existing aggregate and pooled data do not provide the linked sequence.Source: §6.4 · pp. 29–30

6.5 From platform records to social organization

  • Platform prominence alone cannot establish lasting social roles or expertise use; comparisons should follow actors across channels and occasions and connect public records with reported exchanges and decisions.Journalistic sourcing distinguishes monitoring, information gathering, and public citation, while the movement case leaves offline coalition networks unobserved.Source: §6.5 · p. 30

7 Limitations

The review’s evidence is bounded by its retrieval, metadata, and purposive case-selection choices. These constraints limit claims about coverage, prevalence, and generality across communication-network research.

Corpus scope

  • The corpus is bounded by Web of Science coverage, the Communication category, the query, and available metadata, and it is not a screened population of all eligible studies.Coverage through 31 August 2026 gives newer records shorter citation and observation windows; final query-execution history and complete export settings were not recovered.Source: §7 · p. 30

Measurement and citation limits

  • Metadata and reference-string limits mean bibliographic diagnostics describe retrieved materials rather than the prevalence of substantive practices or fully consolidated works.Reference counts cannot recover absent strings, keywords have different origins, and one identity check changes a resource’s frequency position.Source: §7 · p. 30

Case and subfield coverage

  • Purposive comparisons and uneven subfield coverage constrain generalization, including conclusions about citation functions, network agenda setting, and journalistic sourcing.The review’s selected cases, settings, and observations leave distributions across the candidate literature open; several relevant areas receive limited treatment.Source: §7 · p. 30

8 Conclusion

The review connects research topics, knowledge organization, and practices through comparisons that clarify what findings explain together. It identifies unresolved conditions for further study while keeping its claims within purposively selected evidence.

Conclusion

  • Connection value depends on task and criterion, shared media or issues can accommodate differentiated choices and evaluations, and some influence findings remain unresolved.These judgments extend a map of related literatures into an account of what their findings explain together.Source: §8 · p. 31
  • Compatible findings and unresolved contrasts require different treatment: shared attention can sustain different meanings, while influence comparisons need comparable relational and temporal conditions.Knowledge-resource analysis connects the judgments to individual studies’ expectations and measurements; bibliographic proximity alone cannot establish them.Source: §8 · p. 31
  • The agenda calls for comparing friendship and content-related conversation across group stages, following sources through reporting tasks, and linking encountered content with recipients’ interpretations.These purposive comparisons identify connections without establishing their prevalence across the field or resolving them causally.Source: §8 · p. 31

Declaration of AI Assistance

The author reports using LLMs for literature coding, code development and checking, and language revision, while retaining responsibility for the manuscript. The appendices document search provenance, a bounded supplementary case search, and differences between keyword fields.

Declaration of AI Assistance

  • LLMs assisted with literature coding, analysis and visualization code development and checking, and manuscript language revision; the author verified the final content and interpretations.The author takes responsibility for the manuscript.Source: Declaration of AI Assistance · p. 31

A.1 Search expression and provenance

  • The final export contains 2,114 records, but the documented confirmation of the query and Article-or-Review filter is retrospective and the final execution history is unavailable.The archived query records pilot searches; the document-type filter is reported separately from the query string.Source: §A.1 · p. 33

A.2 Bounded search for qualifying cases

  • A supplementary search screened 171 unique records and retained comparisons through a capped full-text queue, but its synthesis-informed queries and lexical coverage cannot establish saturation or independent validation.Screening yielded 2/32/17/0, 20/52/25/1, and 21/5/0/1 records across the four categories for the three queries; one original remained unavailable.Source: §A.2 · pp. 33–34

A.3 DE and ID as distinct vocabulary measurements

  • Author keywords and Keywords Plus are retained as distinct vocabulary measurements because their coverage, origins, terms, and co-occurrence patterns differ.Their union covers 2,042 records (96.6%); this is metadata coverage, not eligibility, and neither field measures the prevalence of communication-network practices.Source: §A.3 · p. 34

A.4 Full-pool title-and-abstract structure

Title-and-abstract vocabulary describes broad methods and objects, while the synthesis organizes communication questions into four interpretive families. Neither vocabulary profile nor family grouping provides an exhaustive or mutually exclusive classification.

A.4 Full-pool title-and-abstract structure

  • Title-and-abstract vocabulary counts measure literal lexical incidence in the retrieved records, without stemming, synonym merging, or inferred topic assignments.Preprocessing retained all 2,114 candidates for lexical analysis; the full pool includes 1,920 English and 194 non-English records.Source: §A.4 · p. 34
  • Network appeared in 1,556 English records (81.0%), analysis in 1,425 (74.2%), social in 1,117 (58.2%), and media in 903 (47.0%); leading bigrams were also broad method–object terms.Network analysis appeared in 854 records (44.5%), social network in 473 (24.6%), and content analysis in 195 (10.2%). These frequencies indicate common terminology, not uniform network objects or practices.Source: §A.4 · p. 34
  • The family-probe counts are transparent term counts, not family membership counts, because records may match multiple probes and terms may cross problem boundaries.For example, counts include 80 records for polarization, 91 for activism, 83 for diffusion, and 69 for exposure.Source: §A.4 · pp. 34–35
  • Four interpretive families organize questions identified through reading, whereas lexical profiles describe procedures and broad objects; neither is exhaustive or mutually exclusive.Across the pool, family probes cover public discussion and collective action, information flow and social selection, meaning and framing, and organization, news, and audience connections.Source: §A.4 · pp. 34–35

A.5 Period-level lexical context

Period-level lexical hits, citation ages, and co-citation checks describe different aspects of the retrieved literature. Their patterns support descriptive comparisons, not estimates of topic prevalence or theoretical integration.

A.5 Period-level lexical context

  • Literal object vocabulary rose from 14/86 records (16.28%) in the earliest period to 663/962 (68.92%) in 2020–2025, while meaning/framing and public-discussion vocabulary also increased.Information-flow/social-selection vocabulary remained comparatively flat after 2005–2014, while organization/news/audience-connection vocabulary peaked in 2015–2019 before reaching 30.87% in 2020–2025.Source: §A.5 · p. 35
  • These period-level patterns are descriptive lexical-hit counts, not topic-prevalence estimates or evidence of theoretical replacement; changing wording and indexing may partly contribute.The 2026 records are right-censored and treated as a monitoring endpoint, not a complete period.Source: §A.5 · p. 35

A.6 Cited-reference age profile

  • Among 85,084 computable citation ages, the median is 8.0 years, the mean is 12.23 years, and the range is 0–504 years.Citation age is the citing year minus the cited year for strings with one parseable year; interpretation is limited by citation conventions, year discrepancies, and research-area differences.Source: §A.6 · p. 35

A.7 Descriptive reference-view and illustrative-pair checks

  • In the primary co-citation view, raw contributions favor method–theory pairs, but fractional weighting puts method–method pairs slightly higher; these descriptive comparisons establish coexistence, not dominance.The selected McPherson–Bakshy and Granovetter–Bennett string-level links each remain jointly cited by nine records under the reported checks, not work-level prevalence estimates.Source: §A.7 · p. 35

B Conditional Co-Citation Comparison

The conditional co-citation comparison tests selected reference-string pairs against a year- and margin-preserving randomized baseline. Saved-state diagnostics and alternative specifications support stability of retained states and directional patterns, while leaving exploratory choices and baseline uncertainty in view.

B Conditional Co-Citation Comparison

  • The primary comparison tests all 435 possible pairs among 30 reference-string identities, including unobserved pairs; the four content cases were selected contrastively rather than preregistered.The 69 record–pair memberships correspond to 64 unique records, and the statistical comparison does not verify specific citation functions.Source: B Conditional Co-Citation Comparison · p. 35
  • The Curveball randomization trades selected-reference incidences within citing years while preserving row and column totals; four chains save 10,000 states each for the primary analysis.Benjamini–Yekutieli adjustment covers all possible pairs within each specification, while the ten resource-type combinations form a separate testing family.Source: B Conditional Co-Citation Comparison · pp. 35–36
  • Saved-state diagnostics show maximum split bR = 1.00028 and minimum estimated effective sample size of approximately 36,747 of 40,000 states; they do not validate the baseline or exact tail-error guarantees.Figures 7 and 8 display traces, cumulative means, and autocorrelations for selected pairs and diagnostic extremes; warmup was not saved and simulation settings were not varied.Source: B Conditional Co-Citation Comparison · pp. 36, 38, Figure 7 · p. 37, Figure 8 · p. 38
  • The approximate Monte Carlo upper-bound adjustment is about 0.0445 for the selected pairs, but its binomial approximation for dependent states is a sensitivity check, not an exact confidence bound.Equal adjusted values for the four pairs reflect finite simulation resolution rather than equal effects.Source: B Conditional Co-Citation Comparison · p. 36
  • Selected-pair excesses and the directions of method–theory and method–method contrasts persist across four alternative specifications, though approximate Monte Carlo bounds can exceed 0.05.The alternatives lower the reference threshold, exclude 2026 or a flagged string, or retain records with matching reported and exported reference counts; this supports directional stability, not identical rejection decisions.Source: B Conditional Co-Citation Comparison · p. 36

C Case–Resource Linkage and Publication-Source Checks

The case–resource analysis uses selected citation strings to guide purposive reading, while distinguishing bibliographic linkage from evidence about communication practices. Coverage, conditional comparisons, and identity checks delimit what these bibliographic patterns can support.

C Case–Resource Linkage and Publication-Source Checks

  • The linkage graph contains 2,114 candidate records and 3,031 citation incidences, but 869 records lack matches to the 60 selected reference strings.Among 24 topic-reading cases, 19 connect to a selected string; the five without matches remain in the comparison, which does not estimate prevalence.Source: C Case–Resource Linkage and… · p. 39
  • Abstract coding draws on 101 verified excerpts, retains disagreements, and identifies broad relation families without estimating intercoder reliability or establishing exact edge definitions.The five selected resource pairs were also checked after deleting each candidate publication source; this addition does not enlarge the original content review.Source: C Case–Resource Linkage and… · p. 39
  • Publication-source sensitivity deletes each of 278 candidate publication sources in turn; a surviving co-citing set shows no single source accounts for all its records.The checks concern citing-paper publication sources, not sources of information studied in those papers.Source: C Case–Resource Linkage and… · p. 39, Table 6 · p. 39
  • The auxiliary comparisons condition on each reference string’s citing-record frequency within year, source, or joint strata, supplementing rather than extending the Curveball null.The tests fix endpoint margins, not each paper’s complete selected-resource degree.Source: C Case–Resource Linkage and… · p. 39
  • All five selected pairs pass the year-only and source-only adjusted thresholds, while three pass joint year-by-source conditioning; retained co-citation after source deletion does not retest significance.Fine stratification leaves limited variation, and failure to pass the adjusted threshold does not establish absence of a substantive connection.Source: C Case–Resource Linkage and… · p. 39

C.1 Coverage and reference identity

  • Among 869 records outside the selected-string view, 858 have at least two distinct exported references, and one abstract reports network analysis of 225 farmers.That case is an abstract-level counterexample to treating noncoverage as theoretical absence, not an additional full-text case; the diagnostic does not estimate omitted theories or practices.Source: §C.1 · pp. 39–40
  • Merging two verified book-reference strings produces a citing-record union of 179, compared with separate frequencies of 110 and 69, showing that splitting can alter apparent frequency rank.A further DOI-confirmed variant raises the union to 183 and coverage by three records; this limited correction does not assess other identity errors.Source: §C.1 · p. 40

D Observation Guide for Practice Comparisons

The observation guide compares what network studies record with the boundaries of the interpretations they support. Across practice types, observed ties, choices, and textual signals are not interchangeable with interaction, exposure, attitudes, or knowledge uptake.

D Observation Guide for Practice Comparisons

  • Table 7 records the objects and observations behind the practice comparisons, allowing closer operational comparison for full-text cases than for abstract-level cases.Unreported observation windows, weights, and model details remain unreported rather than inferred.Source: D Observation Guide for Practice… · p. 40
  • Knowledge-seeking studies concern anticipated retrieval choices or reported channels tied to recognized expertise, whereas co-mention and URL-sharing studies characterize source combinations or media repertoires.Neither observation directly establishes retrieval events, trust, performance, interaction, or population-wide consumption.Source: D Observation Guide for Practice… · p. 41
  • Audience overlap captures shared outlets or attendance, not identical content, interpretation, attitudes, or general attitude change.The observations concern outlet pairs over one month or composition and choice in the reported setting.Source: D Observation Guide for Practice… · p. 41
  • Rumor co-membership projects user–rumor posting affiliation, while tweet-level indicators track stance before and after a first fact-check.These observations concern selective participation and diffusion or correction trends, not direct user interaction, exposure, endorsement, or attitude change.Source: D Observation Guide for Practice… · p. 41
  • Climate-network comparisons use directed retweets and mentions plus addressee-based text comparison to distinguish outgoing address, incoming attention, and expression.They do not measure repeated-message counts, sustained dialogue, or knowledge uptake.Source: D Observation Guide for Practice… · p. 41
Loading 2609.30905v1…