Pith. sign in

REVIEW 4 major objections 5 minor 55 references

Bridging the Narrative Divide: Cross-Platform Discourse Networks in Fragmented Ecosystems

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A content-based network built from shared narratives shows that 2,864 bridge users, just 0.33 percent of all users, were the first cross-platform carriers for roughly 70 percent of narratives that migrated between Truth Social and X…

desk verdict A genuinely useful method and a striking bridge-user finding, but the headline 70% is not yet shown to be stable under the clustering threshold that defines the narratives. read the letter →

arxiv 2505.21729 v1 pith:6CCRZYOC submitted 2025-05-22 cs.SI cs.CY

Patrick Gerard , Hans W. A. Hanley , Luca Luceri , Emilio Ferrara This is my paper
classification cs.SIcs.CY
keywords bridgeuserscross-platformnarrativemigrationdiscoursenetworksclusteraffiliationnetworkembeddingTruthSocialX(Twitter)DP-Meansclusteringtransferentropy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that cross-platform political narratives move along identifiable structural channels rather than diffusing at random. It introduces CANE, a content-based method that builds a user network from shared participation in latent narrative clusters instead of reposts, mentions, or follower links. Applied to Truth Social and X during the 2024 U.S. presidential election, the method finds a bridge zone of 2,864 mixed-platform users — 0.33 percent of all users and 2.14 percent of posts — who served as the first cross-platform carriers for 68 percent of simple migrating narratives and 69 percent of significant migrating narratives. If correct, narrative migration is concentrated in a small, structurally placed subgroup, giving platform governance and content moderation a concrete target set. The paper also reports that this network matches or beats interaction-based baselines on information-operation detection, ideological stance prediction, and cross-platform engagement prediction while requiring far less data.

What carries the argument

The central object is CANE (Cluster Affiliation Network Embedding), a three-step pipeline: posts are embedded with MPNet, DP-Means clusters the embeddings into latent narratives using a cosine-distance threshold near 0.65–0.70, and each user becomes a TF-IDF-weighted vector over cluster affiliations whose cosine similarities (computed with the FAISS-HNSW approximate nearest neighbor index) define a weighted user graph. t-CANE adds a temporal update rule that strengthens edges when users co-engage with the same clusters across time windows and decays inactive edges. Running Louvain community detection on the unified X-plus-Truth-Social graph then isolates the high-entropy bridge community, and its members are the bridge users; the load-bearing step is the assumption that cluster boundaries correspond to narrative boundaries.

What would settle it

Recompute the bridge-zone and migration statistics while sweeping the DP-Means cosine threshold from about 0.55 to 0.80, or while defining narratives by alternative signals such as shared URLs or verbatim quotes, and check whether bridge users remain the first cross-platform carriers in 68 to 69 percent of cases. If the share falls toward the 0.33 percent baseline user share, or if the high-entropy bridge community dissolves, the concentration claim would be an artifact of the clustering choice.

Watch

Extended reading notes

Core claim

The paper claims that during the 2024 U.S. presidential election, narratives migrated between Truth Social and X along repeatable structural pathways, and that the key carriers were a small set of bridge users embedded at the intersection of otherwise siloed discourse communities. These users are not high-volume or highly visible: their engagement metrics sit near platform medians. Yet they appeared among the first posters on the receiving platform in 68 to 69 percent of migrating narratives, an overrepresentation of more than 200 times their population share, and they seeded 26 percent of all narratives. The paper further reports that Truth Social, which contributed only about 1 to 2 percent of post volume within migrating narratives, originated 15.6 to 18.9 percent of them, making it an outsized incubator of cross-platform discourse. These patterns, the paper argues, only become visible when users are linked through shared narrative engagement rather than through platform-specific behaviors.

Load-bearing premise

The load-bearing premise is that DP-Means clusters at a cosine-similarity threshold around 0.65 to 0.70 really isolate coherent narratives, since both the user network and the definition of narrative migration are built from those clusters; if the threshold is too coarse or too fine, the bridge community and the roughly 70 percent introduction rate could be artifacts of cluster granularity.

Editorial extensions

If this is right

  • Cross-platform narrative flow can be reconstructed without any platform-specific interaction signals, so the method transfers to platforms with restricted APIs or sparse metadata.
  • A tiny structural subgroup, not high-engagement accounts, carries most migrating narratives, so early-diffusion monitoring and governance can concentrate on bridge-zone positions rather than on volume leaders.
  • Truth Social's disproportionate origin share, 11 to 14 times its post volume, suggests small alt-tech platforms can incubate narratives that later reach much larger audiences.
  • Early participation by bridge users is associated with significantly higher downstream likes, reposts, and replies, even for low-virality narratives, consistent with an early-validator role.
  • The network reaches near-peak predictive performance with only 5 to 10 percent of user content, making large-scale cross-platform monitoring computationally feasible.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The 70 percent introduction rate is defined relative to DP-Means clusters at one cosine threshold; sweeping that threshold, which the paper validates on only 50 post pairs, would show how much of the concentration is structural and how much depends on cluster granularity.
  • If the same small set of users reappears as bridges in other election cycles or crises, the method could serve as an early-warning signal for narrative laundering from Truth Social to X; the paper does not test the temporal stability of bridge membership.
  • The same cluster-affiliation representation could be applied to organizations or media outlets instead of individuals, tracing which outlets act as cross-platform bridges rather than which individual users do.
  • Because the method needs only post content, it could be applied retroactively to archived data where interaction graphs are unavailable, potentially revealing bridge structures in earlier fragmented ecosystems.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes CANE and t-CANE, a content-based, platform-agnostic framework for constructing user-user networks from latent narrative clusters. The method is evaluated on information-operation detection, ideological stance prediction, and a new cross-platform engagement prediction benchmark, where it reports state-of-the-art results. The paper then applies the framework to Truth Social and X discourse during the 2024 U.S. election, identifying a set of 2,864 'bridge users' who comprise 0.33% of users and 2.14% of posts, and claims these users are the first cross-platform carriers for 68% of simple migrating narratives and 69% of significant migrating narratives. The paper also reports that Truth Social disproportionately seeds migrating narratives and that bridge users are associated with higher downstream engagement even for low-prominence narratives.

Significance. If the central empirical claim holds, the paper makes a meaningful contribution: it offers a scalable, platform-agnostic network construction method that performs competitively on multiple benchmarks, and it identifies a small, structurally identifiable group of users who appear to facilitate cross-platform narrative diffusion. The paper includes several strengths: code and data are promised, the benchmark tasks are diverse, the temporal extension and ablations are documented, and the bridge-user analysis includes a de-duplication audit and comparisons against alternative graph constructions. However, the headline 68-69% introduction rate depends on an inconsistently reported clustering threshold and lacks a null model, so the empirical contribution is not yet fully established.

major comments (4)
  1. [Cross-Platform Narrative Migration; Content Clustering; Appendix A; Appendix G] The narrative clusters that define migration are computed with an inconsistently reported DP-Means threshold. The main text first states a cosine-distance cutoff of 0.30 'i.e., minimum similarity of 0.70', then says 'we adopt a 0.65 threshold'; Appendix A says 'We adopt a threshold of 0.70'; and Appendix G uses 0.35 for the same style of clustering. Because the 68%/69% bridge-user introduction rate is counted over these clusters, and because the same clusters are used to build the CANE graph from which bridge users are extracted, the headline number is not well-defined until the threshold is pinned down. Table 11 validates pairwise semantic equivalence on only 50 post pairs per threshold; it does not test how the bridge-zone size, the 200x overrepresentation, or the 68-69% introduction rate change with lambda. Please add a lambda sweep over the plausible range (e.g., 0.60-0.80) reporting the headline statistics, and state one adopted threshold in all places.
  2. [Bridge Users; Appendix G, Table 22] The paper states that 2,864 bridge users comprise 0.33% of all users, but Table 22 reports 261,398 total users in the X/Truth Social engagement dataset; 2,864/261,398 is approximately 1.1%, not 0.33%. The denominator for the 0.33% figure must be stated explicitly (e.g., the full user set of the discourse network after filtering), and the user/post percentages must be recomputed consistently. This is a headline quantity, so the mismatch is load-bearing.
  3. [Discourse as Structures: Bridge Zones; Bridge Users] The bridge community is selected as the high-entropy community in a Louvain partition of a graph whose edges are cluster-affiliation overlaps, and migration is defined using those same clusters. The statement that bridge users 'serve as the first cross-platform carriers for 68% of simple migrating narratives and 69% of significant migrating narratives' needs a null model to rule out the possibility that any cross-platform-mixed community of similar size would show comparable rates. Table 27 compares against alternative graph constructions, but those baselines share the same narrative clusters and none is a random or matched control. I suggest a permutation test (e.g., random communities of equal size and platform mix matched on activity, or label-permuted Louvain partitions) and reporting the introduction-rate distribution under the null.
  4. [Bridge Users; Appendix I, Table 28] The headline introduction rates are point estimates with no uncertainty, and they are based on a single Louvain run. Louvain is stochastic, and the 'first carrier' rule depends on the resulting partition. Please report the introduction rates across Louvain seeds (or a consensus partition) and provide confidence intervals for the 68%/69% and 26% seeding figures. A single run is not sufficient for a central quantitative claim.
minor comments (5)
  1. [Information Operations; Tables 2 and 3] The text says t-CANE reaches 0.85 Macro-F1 and 0.92 AUC on the China IO dataset, but Table 2 reports 0.83 ± 0.01 and 0.98 ± 0.01. Please reconcile the text with the table.
  2. [Appendix G] Appendix G uses a 0.35 cosine-distance cutoff for the engagement prediction clustering, while the narrative migration analysis uses 0.65 or 0.70. Clarify whether this is a deliberate design choice or an inconsistency.
  3. [Appendix B; Appendix I; Paper Checklist] There are several typos: 'illutrating' in Appendix B, 'Cross-Platfrom' in the Appendix I heading, and 'we we apply' in the checklist answer. Please proofread.
  4. [Appendix I, Table 25] In Table 25, the Random GCN baseline shows AUC 0.56 at t=7 but AUC 0.00 at t=3, 5, and 14. If this is due to threshold sweeping or class weighting, please state that explicitly; otherwise it looks like an error.
  5. [Bridge Users and Narrative Diffusion] The sentence about two users pinning an identical inflammatory image would benefit from a precise account of how the image match was determined, since exact image matching across platforms can be confounded by re-uploads and platform-specific compression.

Circularity Check

2 steps flagged · score 6.0 of 10

Bridge-zone definition embeds the migration outcome, and the engagement benchmark uses the same clusters for graph edges and labels; the headline 68–69% statistic is therefore partly a re-description of the shared clustering instrument rather than an independent structural discovery.

  1. self definitional [Bridge Users section (Connective Narratives: X and Truth Social)]
    "Building on the concept of bridge zones (regions of structural overlap in the discourse network where narratives migrate across platforms), we define bridge users as members of these high-entropy, cross-platform communities. ... They serve as the first cross-platform carriers for 68% of simple migrating narratives and 69% of significant migrating narratives."

    The paper defines bridge zones, in the paper's own words, as regions where narratives migrate across platforms; bridge users are then defined as members of these zones. The headline result that bridge users are the first carriers of 68–69% of migrating narratives is thus partly an unpacking of the construct rather than an independent empirical test. The operational selection uses Louvain entropy, but both sides of the statistic are generated by the same DP-Means cluster instrument: CANE graph edges encode shared cluster participation, and a migrating narrative is a DP-Means cluster that appears on both platforms.

  2. self definitional [Cross-Platform Engagement Prediction section and Appendix G]
    "We apply this to a dataset of U.S. election-related posts from X and Truth Social (May–Nov 2024), clustered into 321 cross-platform narrative themes using multilingual MP-Net embeddings and DP-means clustering. ... we train GCNs on each graph, using binary user-topic features to predict future engagement. ... To prevent circularity, narrative text was excluded from graph features."

    The prediction labels are future engagement with one of the 321 DP-Means narrative themes, while the CANE graph edges are cosine similarities of users' cluster-affiliation vectors over those same 321 themes, and the node features are binary indicators of past engagement with those same themes. For a given theme k, the CANE graph connects users whose history already contains theme k or similar themes, so the GCN can propagate near-label information from the features and edges; the large AUC gap over hashtag/URL baselines is an artifact of this shared cluster instrument. Excluding narrative text does not remove the circularity, because the cluster labels themselves are the target variable and the graph-construction input.

full rationale

The central bridge-user claim is not wholly forced: temporal firstness, the 24-hour migration window, the engagement floor, and Louvain community assignment are not logically determined by the clustering alone, and 31–32% of migrating narratives are first carried by non-bridge users. However, the paper's own definition of bridge zones as regions where narratives migrate, combined with the fact that both the CANE graph and the migration labels derive from the same DP-Means clusters, makes the 68–69% statistic partially a re-description of the measurement instrument rather than an independent finding. The engagement-prediction benchmark has a similar same-cluster coupling: graph edges, node features, and labels all reference the same 321 narrative themes, so the strong performance advantage is expected by construction. The inconsistent threshold reporting (0.65 vs. 0.70 vs. 0.30/0.35) is a robustness concern but not itself a circularity. The information-operation and ideological-stance benchmarks use external labels and are not circular, and the self-citations for clustering thresholds are supplemented by human evaluation, so they are not load-bearing. Overall, the paper has partial, not total, circularity, concentrated in the headline bridge-user result and the engagement benchmark.

Assumptions & free parameters 6 free parameters · 4 assumptions · 1 invented entities

The central claim depends on several calibrated thresholds and the modeling assumption that cluster participation equals discourse alignment. No new physical entities are proposed; 'bridge users' are a data-derived label.

free parameters (6)
  • Clustering similarity threshold lambda = 0.65 or 0.70 (inconsistent in text)
    Controls DP-Means cluster granularity, defining what counts as a narrative; calibrated via human evaluation on 50 post pairs (Appendix A, Table 11).
  • t-CANE memory weight alpha = 0.8
    Controls how much new similarities influence edge weights; chosen via grid search on China IO task (Appendix C, Table 12).
  • t-CANE decay weight beta = 0.2
    Controls decay of inactive edges; chosen via grid search (Appendix C, Table 12).
  • Minimum engagement threshold for migration = 10 posts (35th percentile)
    Defines 'substantive diffusion' in narrative migration; set post hoc, with sensitivity analysis mentioned (Table 11).
  • Early seeding threshold = first 5% of participants
    Defines 'early' bridge user participation; robustness checks vary 5-30% (Appendix I).
  • Fear speech score threshold = 0.75
    Threshold for classifying fear-laden posts; chosen to prioritize precision, with stricter checks at 0.9.
assumptions (4)
  • domain assumption MPNet embeddings capture semantic similarity such that cosine distance is a valid measure of narrative overlap.
    Invoked in the clustering step (Methods: Content Aggregation and Embedding).
  • domain assumption Users' cluster participation profiles are a faithful representation of their discourse position.
    This is the core modeling assumption behind CANE (Eq. 1-2).
  • domain assumption Transfer entropy of post-count time series measures directional narrative migration.
    Used to define 'significant migration' (Cross-Platform Narrative Migration).
  • domain assumption The 2024 U.S. election discourse on X and Truth Social is representative of fragmented ecosystems.
    The empirical conclusions are drawn from this single case study.
invented entities (1)
  • Bridge users / bridge zone
    purpose: To explain and predict cross-platform narrative diffusion as a structural role.
    The bridge zone is a Louvain community in the same graph used to define narratives; the 'predictions' about engagement are tested within the same dataset, not against an external benchmark.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bridging the Narrative Divide: Cross-Platform Discourse Networks in Fragmented Ecosystems." pith.science (2026). https://pith.science/paper/6CCRZYOC

@misc{pith2026250521729,
  author       = {Pith},
  title        = {Pith review of: Bridging the Narrative Divide: Cross-Platform Discourse Networks in Fragmented Ecosystems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6CCRZYOC}},
  note         = {Machine review of arXiv:2505.21729}
}
abstract

Political discourse has grown increasingly fragmented across different social platforms, making it challenging to trace how narratives spread and evolve within such a fragmented information ecosystem. Reconstructing social graphs and information diffusion networks is challenging, and available strategies typically depend on platform-specific features and behavioral signals which are often incompatible across systems and increasingly restricted. To address these challenges, we present a platform-agnostic framework that allows to accurately and efficiently reconstruct the underlying social graph of users' cross-platform interactions, based on discovering latent narratives and users' participation therein. Our method achieves state-of-the-art performance in key network-based tasks: information operation detection, ideological stance prediction, and cross-platform engagement prediction$\unicode{x2013}$$\unicode{x2013}$while requiring significantly less data than existing alternatives and capturing a broader set of users. When applied to cross-platform information dynamics between Truth Social and X (formerly Twitter), our framework reveals a small, mixed-platform group of $\textit{bridge users}$, comprising just 0.33% of users and 2.14% of posts, who introduce nearly 70% of $\textit{migrating narratives}$ to the receiving platform. These findings offer a structural lens for anticipating how narratives traverse fragmented information ecosystems, with implications for cross-platform governance, content moderation, and policy interventions.

Figures

Figures reproduced from arXiv: 2505.21729 by the authors.

Figure 1
Figure 1. Overview of our cross-platform user network inference framework. Content is embedded and clustered into se [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. A conceptual illustration of cross-platform nar [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Comparison of computational complexity across network [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Proportion of peak AUC achieved as a function of [PITH_FULL_IMAGE:figures/full_fig_p017_4.png]
Figure 5
Figure 5. Figure 5: Visualization of the discourse network colored [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

55 extracted references · 42 canonical work pages

  1. [1]

    Aizawa, A. 2003. An information-theoretic perspective of tf--idf measures. Information Processing & Management, 39(1): 45--65

    work page 2003

  2. [2]

    Alieva, I.; Ng, L. H. X.; and Carley, K. M. 2022. Investigating the spread of Russian disinformation about biolabs in Ukraine on Twitter using social network analysis. In 2022 IEEE international conference on big data (big data), 1770--1775. IEEE

    work page 2022

  3. [3]

    Balasubramanian, A.; Zou, V.; Narayana, H.; You, C.; Luceri, L.; and Ferrara, E. 2024. A public dataset tracking social media discourse about the 2024 us presidential election on twitter/x. arXiv preprint arXiv:2411.00376

    arXiv 2024

  4. [4]

    Bollen, J.; Gon c alves, B.; Ruan, G.; and Mao, H. 2011. Happiness is assortative in online social networks. Artificial life, 17(3): 237--251

    work page 2011

  5. [5]

    Burghardt, K.; Rao, A.; Chochlakis, G.; Sabyasachee, B.; Guo, S.; He, Z.; Rojecki, A.; Narayanan, S.; and Lerman, K. 2024. Socio-linguistic characteristics of coordinated inauthentic accounts. In Proceedings of icwsm, volume 18, 164--176

    work page 2024

  6. [6]

    Cinelli, M.; De Francisci Morales, G.; Galeazzi, A.; Quattrociocchi, W.; and Starnini, M. 2021. The echo chamber effect on social media. Proceedings of the National Academy of Sciences, 118(9): e2023301118

    work page 2021

  7. [7]

    Cinus, F.; Minici, M.; Luceri, L.; and Ferrara, E. 2025. Exposing cross-platform coordinated inauthentic activity in the run-up to the 2024 us election. In Proceedings of the ACM on Web Conference 2025, 541--559

    work page 2025

  8. [8]

    Cross, R.; and Prusak, L. 2002. The people who make organizations go--or stop. Networks in the Knowledge Economy, 80(6): 248--260

    work page 2002

Show all 55 references
  1. [9]

    Gerard, P.; Botzer, N.; and Weninger, T. 2023. Truth social dataset. In Proceedings of icwsm, volume 17, 1034--1040

    2023

  2. [10]

    Gerard, P.; Theisen, W.; Weninger, T.; and Lerman, K. 2024. Fear and Loathing on the Frontline: Decoding the Language of Othering by Russia-Ukraine War Bloggers. arXiv:2409.13064

    2024 arXiv

  3. [11]

    V.; and Fernandez, R

    Gould, R. V.; and Fernandez, R. M. 1989. Structures of mediation: A formal approach to brokerage in transaction networks. Sociological methodology, 89--126

    1989

  4. [12]

    Granovetter, M. 1983. The strength of weak ties: A network theory revisited. Sociological theory, 201--233

    1983

  5. [13]

    Grover, A.; and Leskovec, J. 2016. node2vec: Scalable feature learning for networks. In Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining, 855--864

    2016

  6. [14]

    W.; and Durumeric, Z

    Hanley, H. W.; and Durumeric, Z. 2024. Partial mobilization: Tracking multilingual information flows amongst russian media outlets and telegram. In Proceedings of icwsm, volume 18, 528--541

    2024

  7. [15]

    W.; Kumar, D.; and Durumeric, Z

    Hanley, H. W.; Kumar, D.; and Durumeric, Z. 2023. Happenstance: utilizing semantic search to track Russian state media narratives about the Russo-Ukrainian war on Reddit. In Proceedings of icwsm, volume 17, 327--338

    2023

  8. [16]

    W.; Kumar, D.; and Durumeric, Z

    Hanley, H. W.; Kumar, D.; and Durumeric, Z. 2024. Specious sites: Tracking the spread and sway of spurious news stories at scale. In 2024 IEEE Symposium on Security and Privacy (SP), 1609--1627. IEEE

    2024

  9. [17]

    H.; and Padang, S

    Hasugian, A. H.; and Padang, S. 2023. Image Similarity Detection Application Using The Ratcliff/Obershelp Method And Levenshtein Distance. Instal: Jurnal Komputer, 15(01): 80--88

    2023

  10. [18]

    Iofciu, T.; Fankhauser, P.; Abel, F.; and Bischoff, K. 2011. Identifying users across social tagging systems. In Proceedings of icwsm, volume 5, 522--525

    2011

  11. [19]

    Jiang, J.; Ren, X.; and Ferrara, E. 2023. Retweet-bert: political leaning detection using language features and information diffusion on social networks. In Proceedings of icwsm, volume 17, 459--469

    2023

  12. [20]

    J.; Taimre, T.; and Pollett, P

    Laub, P. J.; Taimre, T.; and Pollett, P. K. 2015. Hawkes processes. arXiv preprint arXiv:1507.02822

    2015 arXiv

  13. [21]

    Luceri, L.; Cresci, S.; and Giordano, S. 2021. Social media against society. The Internet and the 2020 Campaign, 1

    2021

  14. [22]

    Luceri, L.; Pant \`e , V.; Burghardt, K.; and Ferrara, E. 2024 a . Unmasking the web of deceit: Uncovering coordinated activity to expose information operations on twitter. In Proceedings of the ACM on Web Conference 2024, 2530--2541

    2024

  15. [23]

    Luceri, L.; Ye, J.; Jiang, J.; and Ferrara, E. 2024 b . The Susceptibility Paradox in Online Social Influence. arXiv preprint arXiv:2406.11553

    2024 arXiv

  16. [24]

    Magelinski, T.; Ng, L.; and Carley, K. 2022. A synchronized action framework for detection of coordination on social media. Journal of Online Trust and Safety, 1(2)

    2022

  17. [25]

    A.; and Yashunin, D

    Malkov, Y. A.; and Yashunin, D. A. 2018. Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs. IEEE transactions on pattern analysis and machine intelligence, 42(4): 824--836

    2018

  18. [26]

    Mendelsohn, J.; Ghosh, S.; Jurgens, D.; and Budak, C. 2023. Bridging nations: quantifying the role of multilinguals in communication on social media. In Proceedings of icwsm, volume 17, 626--637

    2023

  19. [27]

    Minici, M.; Luceri, L.; Cinus, F.; and Ferrara, E. 2024. Uncovering Coordinated Cross-Platform Information Operations Threatening the Integrity of the 2024 US Presidential Election Online Discussion. arXiv preprint arXiv:2409.15402

    2024 arXiv

  20. [28]

    Ng, L. H. X.; and Carley, K. M. 2022. A Combined Synchronization Index for Grassroots Activism on Social Media. arXiv preprint arXiv:2212.13221

    2022 arXiv

  21. [29]

    Ng, L. H. X.; Cruickshank, I. J.; and Carley, K. M. 2022. Cross-platform information spread during the January 6th capitol riots. Social Network Analysis and Mining, 12(1): 133

    2022

  22. [30]

    Ng, L. H. X.; Cruickshank, I. J.; and Carley, K. M. 2023. Coordinating Narratives Framework for cross-platform analysis in the 2021 US Capitol riots. Computational and Mathematical Organization Theory, 29(3): 470--486

    2023

  23. [31]

    Nizzoli, L.; Tardelli, S.; Avvenuti, M.; Cresci, S.; and Tesconi, M. 2021. Coordinated behavior on social media in 2019 UK general election. In Proceedings of icwsm, volume 15, 443--454

    2021

  24. [32]

    T.; Flammini, A.; and Menczer, F

    Pacheco, D.; Hui, P.-M.; Torres-Lugo, C.; Truong, B. T.; Flammini, A.; and Menczer, F. 2021. Uncovering coordinated networks on social media: methods and case studies. In Proceedings of ICWSM, volume 15, 455--466

    2021

  25. [33]

    Pinto, G.; Bickham, C.; Salkar, T.; Luceri, L.; and Ferrara, E. 2024. Tracking the 2024 US Presidential Election Chatter on Tiktok: A Public Multimodal Dataset. arXiv:2407.01471

    2024 arXiv

  26. [34]

    H.; Blackburn, J.; Bradlyn, B.; De Cristofaro, E.; Stringhini, G.; Long, S.; Greenberg, S.; and Zannettou, S

    Ribeiro, M. H.; Blackburn, J.; Bradlyn, B.; De Cristofaro, E.; Stringhini, G.; Long, S.; Greenberg, S.; and Zannettou, S. 2021. The evolution of the manosphere across the web. In Proceedings of icwsm, volume 15, 196--207

    2021

  27. [35]

    H.; and Casiraghi, G

    Russo, G.; Verginer, L.; Ribeiro, M. H.; and Casiraghi, G. 2023. Spillover of antisocial behavior from fringe platforms: The unintended consequences of community banning. In Proceedings of icwsm, volume 17, 742--753

    2023

  28. [36]

    K.; Pandey, S

    Saha, P.; Garimella, K.; Kalyan, N. K.; Pandey, S. K.; Meher, P. M.; Mathew, B.; and Mukherjee, A. 2023. On the rise of fear speech in online social media. Proceedings of the National Academy of Sciences, 120(11): e2212270120

    2023

  29. [37]

    Schreiber, T. 2000. Measuring information transfer. Physical review letters, 85(2): 461

    2000

  30. [38]

    C.; Pote, M.; Nwala, A.; Yin, L.; Luceri, L.; Flammini, A.; and Menczer, F

    Seckin, O. C.; Pote, M.; Nwala, A.; Yin, L.; Luceri, L.; Flammini, A.; and Menczer, F. 2024. Labeled Datasets for Research on Information Operations. arXiv preprint arXiv:2411.10609

    2024 arXiv

  31. [39]

    Shah, K.; Gerard, P.; Luceri, L.; and Ferrara, E. 2024. Unfiltered Conversations: A Dataset of 2024 US Presidential Election Discourse on Truth Social. arXiv preprint arXiv:2411.01330

    2024 arXiv

  32. [40]

    Sipka, A.; Hannak, A.; and Urman, A. 2022. Comparing the Language of QAnon-related content on Parler, Gab, and Twitter. In Proceedings of the 14th ACM Web Science Conference 2022, 411--421

    2022

  33. [41]

    Song, K.; Tan, X.; Qin, T.; Lu, J.; and Liu, T.-Y. 2020. Mpnet: Masked and permuted pre-training for language understanding. Advances in neural information processing systems, 33: 16857--16867

    2020

  34. [42]

    Tardelli, S.; Nizzoli, L.; Avvenuti, M.; Cresci, S.; and Tesconi, M. 2024 a . Multifaceted online coordinated behavior in the 2020 US presidential election. EPJ Data Science, 13(1): 33

    2024

  35. [43]

    Tardelli, S.; Nizzoli, L.; Tesconi, M.; Conti, M.; Nakov, P.; Da San Martino, G.; and Cresci, S. 2024 b . Temporal dynamics of coordinated online behavior: Stability, archetypes, and influence. Proceedings of the National Academy of Sciences, 121(20): e2307038121

    2024

  36. [44]

    A.; Waltman, L.; and Van Eck, N

    Traag, V. A.; Waltman, L.; and Van Eck, N. J. 2019. From Louvain to Leiden: guaranteeing well-connected communities. Scientific reports, 9(1): 1--12

    2019

  37. [45]

    Tromble, R. 2021. Where have all the data gone? A critical reflection on academic digital research in the post-API age. Social Media+ Society, 7(1): 2056305121988929

    2021

  38. [46]

    Veli c kovi \'c , P.; Cucurull, G.; Casanova, A.; Romero, A.; Lio, P.; and Bengio, Y. 2017. Graph attention networks. arXiv preprint arXiv:1710.10903

    2017 arXiv

  39. [47]

    S.; Nogara, G.; Cardoso, F.; Cresci, S.; Giordano, S.; and Luceri, L

    Vishnuprasad, P. S.; Nogara, G.; Cardoso, F.; Cresci, S.; Giordano, S.; and Luceri, L. 2024. Tracking fringe and coordinated activity on Twitter leading up to the US Capitol attack. In Proceedings of ICWSM, volume 18, 1557--1570

    2024

  40. [48]

    Wang, L.; Yang, N.; Huang, X.; Yang, L.; Majumder, R.; and Wei, F. 2024. Multilingual e5 text embeddings: A technical report. arXiv preprint arXiv:2402.05672

    2024 arXiv

  41. [49]

    Wilson, T.; and Starbird, K. 2020. Cross-platform disinformation campaigns: lessons learned and next steps. Harvard Kennedy School Misinformation Review, 1(1)

    2020

  42. [50]

    Xiao, Z.; Song, W.; Xu, H.; Ren, Z.; and Sun, Y. 2020. TIMME: Twitter ideology-detection via multi-task multi-relational embedding. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, 2258--2268

    2020

  43. [51]

    Zhang, S.; Tong, H.; Xu, J.; and Maciejewski, R. 2019. Graph convolutional networks: a comprehensive review. Computational Social Networks, 6(1): 1--23

    2019

  44. [52]

    Zhang, Y.; Lukito, J.; Suk, J.; and McGrady, R. 2025. Trump, Twitter, and Truth Social: how Trump used both mainstream and alt-tech social media to drive news media attention. Journal of Information Technology & Politics, 22(2): 229--242

    2025

  45. [53]

    Zulli, D.; Coe, K.; and Isaacs, Z. 2023. News framing in the aftermath of the January 6 attacks on the US Capitol: An analysis of labels, definitional uncertainty, and contextualization. American behavioral scientist, 67(6): 702--720

    2023

  46. [54]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  47. [55]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.