pith. sign in

arxiv: 2606.27539 · v1 · pith:NVOCKKUVnew · submitted 2026-06-25 · 💻 cs.SI · cs.AI· cs.LG

Benchmarking Multi-Modal Graph-based Social Media Popularity Prediction

Utkarsh Sahu , Zhisheng Qi , Li Zhu , Yizhao Yang , Jun Li , Ryan Rossi , Yu Wang This is my paper

Pith reviewed 2026-06-29 00:43 UTC · model grok-4.3

classification 💻 cs.SI cs.AIcs.LG
keywords social media popularity predictionmulti-modal learninggraph neural networksbenchmarkcross-platform generalizationmulti-task predictionLLM limitationsBluesky Reddit datasets
0
0 comments X

The pith

MMG-PopNet jointly models multi-modal signals and graph-structured social interactions to outperform baselines on four social media datasets.

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

The paper introduces MMG-Pop, a benchmark that standardizes datasets, modalities, observation windows, and protocols for predicting social media content popularity across Bluesky and Reddit. It proposes MMG-PopNet as a network that combines textual, visual, temporal, and interaction signals within graph structures. Experiments show this joint modeling yields higher accuracy than prior methods while revealing patterns in cross-platform generalization, multi-task learning gains, modality importance, and LLM shortcomings. A reader would care because fragmented prior work made it hard to know which signals actually drive reach and how to compare approaches fairly.

Core claim

MMG-PopNet jointly models multi-modal signals and graph-structured social interactions, demonstrating superior performance on four datasets and yielding new insights into cross-platform training generalization, multi-task prediction benefits, multi-modality contributions, and LLM prediction limitation.

What carries the argument

MMG-PopNet, a unified multi-modal graph-based network that jointly models multi-modal signals and graph-structured social interactions.

If this is right

  • MMG-PopNet achieves higher prediction accuracy than representative baselines on the four unified datasets.
  • Training across platforms improves generalization to new data.
  • Multi-task setups provide measurable prediction benefits over single-task training.
  • Different modalities contribute unequally to final accuracy.
  • Large language models alone show clear limitations compared with the graph-based approach.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The standardized protocol could be applied to additional platforms to test whether the performance edge persists.
  • The graph modeling of interactions suggests that social network structure may be more predictive than content alone in many cases.
  • The benchmark setup enables direct testing of whether adding new modalities or signals further improves results.
  • Insights on cross-platform transfer could guide deployment of one model across multiple sites without full retraining.

Load-bearing premise

The selected datasets, observation windows, prediction targets, and baselines under the standardized protocol are sufficiently representative to support general claims about multi-modal and graph-based popularity prediction across platforms.

What would settle it

Evaluating MMG-PopNet and the baselines on a held-out dataset from an unseen platform under the same protocol and finding no accuracy gain or reversed insights on generalization and modality contributions.

Figures

Figures reproduced from arXiv: 2606.27539 by Jun Li, Li Zhu, Ryan Rossi, Utkarsh Sahu, Yizhao Yang, Yu Wang, Zhisheng Qi.

Figure 1
Figure 1. Figure 1: Overview of MMG-Pop Benchmark. Social cascades from Bluesky and Reddit are repre￾sented as tree-structured graphs, where each node carries multi-modal attributes. Given only an early observed prefix Gt , the task is to predict six complementary popularity dimensions characterizing the future cascade state Gt ′ . The benchmark evaluates baselines alongside our proposed MMG-PopNet across multiple observation… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of MMG-PopNet Model: The model embeds node-level text and temporal signals for bidirectional graph message passing over the cascade, and root visual content and thread metadata are encoded as separate contextual features. The learned root, graph, visual, and metadata representations are fused at the prediction stage to support multi-task popularity forecasting. Multi-Modal Feature Embedding. To en… view at source ↗
Figure 3
Figure 3. Figure 3: MSE-Loss trajectories across datasets, comparing different models for target [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Dataset-Specific vs. Unified Training. Avg MSE of MMG-PopNet under dataset-specific and unified training. Lower is better. Unified training greatly improves performance on Reddit communities and remains competitive on Bluesky. 0.5 1.0 Width Depth Virality Size Users Like Root 0.5 1.0 Width Depth Virality Size Users Like 2m 0.5 1.0 Width Depth Virality Size Users Like 10m 0.5 1.0 Width Depth Virality Size U… view at source ↗
Figure 5
Figure 5. Figure 5: Normalized LLM Performance on Bluesky. Scores are normalized with MMG-PopNet as the reference baseline, fixed at 1.0 on all axes, where smaller areas indicate worse performance. MMG-PopNet outperforms LLM baselines across all settings. Among LLMs, retrieval-augmented few-shot prompting performs better in sparse early windows, while fine-tuning becomes stronger as longer cascade prefixes provide richer temp… view at source ↗
Figure 7
Figure 7. Figure 7: The top example shows a relatively accurate high-popularity prediction, where the predicted [PITH_FULL_IMAGE:figures/full_fig_p040_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The top example shows an over-predicted cascade, where the predicted [PITH_FULL_IMAGE:figures/full_fig_p041_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The highlighted example shows an under-predicted cascade, where the actual [PITH_FULL_IMAGE:figures/full_fig_p042_9.png] view at source ↗
read the original abstract

Social media popularity prediction aims to forecast the future reach or influence of online content from early-stage observations. Accurate prediction enables key downstream applications, such as advertising optimization and strategic content planning by users, creators, and platforms. Despite substantial progress, existing popularity prediction works often fail to jointly consider multimodal content and temporal social interaction signals. Moreover, the literature remains highly fragmented across datasets, modalities, observation windows, prediction targets, and evaluation protocols. This fragmentation prevents fair comparison and obscures a systematic understanding of how textual, visual, temporal, and interaction-based signals jointly shape popularity dynamics. To address these challenges, we introduce MMG-Pop, a Multi-modal Graph-based Popularity Prediction benchmark, which unifies datasets, modalities, temporal interaction signals, and representative baselines under a standardized evaluation protocol. Furthermore, we propose MMG-PopNet, a unified multi-modal graph-based network that jointly models the aforementioned multi-modal signals and graph-structured social interactions. Extensive experiments on MMG-Pop, comprising four datasets across Bluesky and Reddit platforms, demonstrate the superior performance of MMG-PopNet and yield new insights into cross-platform training generalization, multi-task prediction benefits, multi-modality contributions, and LLM prediction limitation. These findings establish a unified foundation for future research on social dynamics modeling and intervention under heterogeneous modalities and socially-aware agentic ecosystem paradigms.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces MMG-Pop, a benchmark that unifies four datasets from Bluesky and Reddit under a standardized protocol for multi-modal graph-based social media popularity prediction, and proposes MMG-PopNet, a model jointly modeling multi-modal content signals and graph-structured social interactions. Experiments on the benchmark claim that MMG-PopNet achieves superior performance and yields insights into cross-platform generalization, multi-task prediction benefits, multi-modality contributions, and LLM limitations.

Significance. If the experimental claims hold under rigorous verification, the standardized benchmark addresses fragmentation in the popularity prediction literature and could enable more systematic comparisons across modalities and interaction graphs. The joint modeling approach in MMG-PopNet represents a concrete step toward integrating heterogeneous signals, and the cross-platform experiments provide a starting point for studying generalization, though the limited platform coverage constrains broader applicability.

major comments (2)
  1. [§5] §5 (Experiments and Results): The claims of 'superior performance' and 'new insights' rest on comparisons that omit error bars, standard deviations across runs, statistical significance tests (e.g., paired t-tests or Wilcoxon), and explicit details on baseline re-implementations, hyperparameter tuning, and data exclusion rules. Without these, it is impossible to assess whether reported gains are robust or attributable to implementation differences.
  2. [§4] §4 (Datasets and Protocol): The central claims about cross-platform training generalization and multi-modality contributions are derived from only two platforms (Bluesky, Reddit) and four datasets under a single observation-window/prediction-target protocol. No quantitative coverage argument, diversity metrics across platforms, or sensitivity analysis to alternative windows/targets is provided, which directly limits the transferability of the reported insights.
minor comments (2)
  1. [§3] Notation for modalities and graph construction should be defined more explicitly in §3 to avoid ambiguity when comparing to prior single-modality baselines.
  2. [Figures/Tables in §5] Figure captions for performance tables could include the exact number of runs and random seeds used, improving reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on experimental rigor and benchmark scope. We address each major comment below and will update the manuscript to strengthen the presentation of results and clarify limitations.

read point-by-point responses

  1. Referee: [§5] §5 (Experiments and Results): The claims of 'superior performance' and 'new insights' rest on comparisons that omit error bars, standard deviations across runs, statistical significance tests (e.g., paired t-tests or Wilcoxon), and explicit details on baseline re-implementations, hyperparameter tuning, and data exclusion rules. Without these, it is impossible to assess whether reported gains are robust or attributable to implementation differences.

    Authors: We agree that these statistical details are necessary to substantiate the performance claims. In the revised version we will report mean performance and standard deviations over multiple random seeds, include paired t-tests (and Wilcoxon where appropriate) for significance, and add explicit sections detailing baseline re-implementations, hyperparameter search ranges and selection criteria, and data exclusion rules. revision: yes

  2. Referee: [§4] §4 (Datasets and Protocol): The central claims about cross-platform training generalization and multi-modality contributions are derived from only two platforms (Bluesky, Reddit) and four datasets under a single observation-window/prediction-target protocol. No quantitative coverage argument, diversity metrics across platforms, or sensitivity analysis to alternative windows/targets is provided, which directly limits the transferability of the reported insights.

    Authors: We acknowledge the limited platform coverage. Bluesky and Reddit were chosen because they supply aligned multi-modal and interaction data under comparable collection conditions; we will add quantitative platform descriptors (e.g., activity distributions, content-type statistics) and dataset-diversity metrics. We will also run sensitivity experiments with alternative observation windows and prediction horizons. Broader platform coverage remains future work due to data-access constraints. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical benchmarking without derivation chain

full rationale

The paper is an empirical benchmarking study introducing MMG-Pop and MMG-PopNet, evaluated via experiments on four datasets from Bluesky and Reddit. The abstract and description contain no equations, derivations, fitted parameters renamed as predictions, or self-citation chains that reduce claims to inputs by construction. All central claims rest on comparative performance metrics under a standardized protocol, which are externally falsifiable via replication on the datasets. This matches the default expectation for non-circular empirical work; the reader's score of 1.0 is consistent with minor framing but no load-bearing circular steps exist.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available; no specific free parameters, axioms, or invented entities are detailed in the provided text.

axioms (1)
  • domain assumption Standard assumptions of graph neural networks and multi-modal fusion techniques apply to social media data.
    The proposed model relies on typical GNN and fusion methods without stating deviations.

pith-pipeline@v0.9.1-grok · 5787 in / 1144 out tokens · 27240 ms · 2026-06-29T00:43:36.652523+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

77 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    Social dynamics

    UNESCO. Social dynamics. https://www.unesco.org/en/tags/social-dynamics-0

  2. [2]

    Social dynamics management: What is it and why is it important for intervention?Journal of Emotional and Behavioral Disorders, 2018

    Thomas W Farmer, Betsy Talbott, Molly Dawes, Heartley B Huber, Debbie S Brooks, and Emily E Powers. Social dynamics management: What is it and why is it important for intervention?Journal of Emotional and Behavioral Disorders, 2018

    2018

  3. [3]

    Discrete choice with social interactions.The Review of Economic Studies, 2001

    William A Brock and Steven N Durlauf. Discrete choice with social interactions.The Review of Economic Studies, 2001

    2001

  4. [4]

    The spread of behavior in an online social network experiment.science, 2010

    Damon Centola. The spread of behavior in an online social network experiment.science, 2010

    2010

  5. [5]

    The structure of scientific collaboration networks.Proceedings of the national academy of sciences, 2001

    Mark EJ Newman. The structure of scientific collaboration networks.Proceedings of the national academy of sciences, 2001

    2001

  6. [6]

    A survey of information cascade analysis: Models, predictions, and recent advances.ACM Computing Surveys (CSUR), 2021

    Fan Zhou, Xovee Xu, Goce Trajcevski, and Kunpeng Zhang. A survey of information cascade analysis: Models, predictions, and recent advances.ACM Computing Surveys (CSUR), 2021

    2021

  7. [7]

    Exposure to opposing views on social media can increase political polarization.Proceedings of the National Academy of Sciences, 2018

    Christopher A Bail, Lisa P Argyle, Taylor W Brown, John P Bumpus, Haohan Chen, MB Fallin Hunzaker, Jaemin Lee, Marcus Mann, Friedolin Merhout, and Alexander V olfovsky. Exposure to opposing views on social media can increase political polarization.Proceedings of the National Academy of Sciences, 2018

    2018

  8. [8]

    Combining interventions to reduce the spread of viral misinformation.Nature Human Behaviour, 2022

    Joseph B Bak-Coleman, Ian Kennedy, Morgan Wack, Andrew Beers, Joseph S Schafer, Emma S Spiro, Kate Starbird, and Jevin D West. Combining interventions to reduce the spread of viral misinformation.Nature Human Behaviour, 2022

    2022

  9. [9]

    Enquiring minds: Early detection of rumors in social media from enquiry posts

    Zhe Zhao, Paul Resnick, and Qiaozhu Mei. Enquiring minds: Early detection of rumors in social media from enquiry posts. InProceedings of the 24th international conference on world wide web, 2015

    2015

  10. [10]

    The spread of low-credibility content by social bots.Nature communications, 2018

    Chengcheng Shao, Giovanni Luca Ciampaglia, Onur Varol, Kai-Cheng Yang, Alessandro Flammini, and Filippo Menczer. The spread of low-credibility content by social bots.Nature communications, 2018

    2018

  11. [11]

    Any- one can become a troll: Causes of trolling behavior in online discussions

    Justin Cheng, Michael Bernstein, Cristian Danescu-Niculescu-Mizil, and Jure Leskovec. Any- one can become a troll: Causes of trolling behavior in online discussions. InProceedings of the 2017 ACM conference on computer supported cooperative work and social computing, 2017

    2017

  12. [12]

    Early prediction of hate speech propagation

    Ken-Yu Lin, Roy Ka-Wei Lee, Wei Gao, and Wen-Chih Peng. Early prediction of hate speech propagation. In2021 International Conference on Data Mining Workshops (ICDMW). IEEE, 2021

    2021

  13. [13]

    Using a model of social dynamics to predict popularity of news

    Kristina Lerman and Tad Hogg. Using a model of social dynamics to predict popularity of news. InProceedings of the 19th international conference on World wide web, 2010

    2010

  14. [14]

    A multimodal approach to predict social media popularity

    Mayank Meghawat, Satyendra Yadav, Debanjan Mahata, Yifang Yin, Rajiv Ratn Shah, and Roger Zimmermann. A multimodal approach to predict social media popularity. In2018 IEEE conference on multimedia information processing and retrieval (MIPR), 2018

    2018

  15. [15]

    Predicting the popularity of online content.Commu- nications of the ACM, 2010

    Gabor Szabo and Bernardo A Huberman. Predicting the popularity of online content.Commu- nications of the ACM, 2010

    2010

  16. [16]

    Using early view patterns to predict the popularity of youtube videos

    Henrique Pinto, Jussara M Almeida, and Marcos A Gonçalves. Using early view patterns to predict the popularity of youtube videos. InProceedings of the sixth ACM international conference on Web search and data mining, 2013

    2013

  17. [17]

    Algorithmic censorship by social platforms: Power and resistance

    Jennifer Cobbe. Algorithmic censorship by social platforms: Power and resistance. 2021

    2021

  18. [18]

    Popularity prediction of facebook videos for higher quality streaming

    Linpeng Tang, Qi Huang, Amit Puntambekar, Ymir Vigfusson, Wyatt Lloyd, and Kai Li. Popularity prediction of facebook videos for higher quality streaming. In2017 USENIX Annual Technical Conference (USENIX ATC 17), 2017. 10

    2017

  19. [19]

    Accurate and novel recommendations: an algorithm based on popularity forecasting.ACM Transactions on Intelligent Systems and Technology (TIST), 2014

    Amin Javari and Mahdi Jalili. Accurate and novel recommendations: an algorithm based on popularity forecasting.ACM Transactions on Intelligent Systems and Technology (TIST), 2014

    2014

  20. [20]

    Multimodal popularity prediction of brand-related social media posts

    Masoud Mazloom, Robert Rietveld, Stevan Rudinac, Marcel Worring, and Willemijn Van Dolen. Multimodal popularity prediction of brand-related social media posts. InProceedings of the 24th ACM international conference on Multimedia, 2016

    2016

  21. [21]

    How to become instagram famous: Post popularity prediction with dual-attention

    Zhongping Zhang, Tianlang Chen, Zheng Zhou, Jiaxin Li, and Jiebo Luo. How to become instagram famous: Post popularity prediction with dual-attention. In2018 IEEE international conference on big data (big data). IEEE, 2018

    2018

  22. [22]

    Toward predicting popularity of social marketing messages

    Bei Yu, Miao Chen, and Linchi Kwok. Toward predicting popularity of social marketing messages. InInternational conference on social computing, behavioral-cultural modeling, and prediction. Springer, 2011

    2011

  23. [23]

    Who is leading the campaign charts? comparing individual popularity on old and new media.Information, communication & society, 2017

    Peter Van Aelst, Patrick Van Erkel, Evelien D’heer, and Raymond A Harder. Who is leading the campaign charts? comparing individual popularity on old and new media.Information, communication & society, 2017

    2017

  24. [24]

    Predicting the speed, scale, and range of information diffusion in twitter

    Jiang Yang and Scott Counts. Predicting the speed, scale, and range of information diffusion in twitter. InProceedings of the International AAAI Conference on Web and Social Media, 2010

    2010

  25. [25]

    What’s in a name? understanding the interplay between titles, content, and communities in social media

    Himabindu Lakkaraju, Julian McAuley, and Jure Leskovec. What’s in a name? understanding the interplay between titles, content, and communities in social media. InProceedings of the international AAAI conference on web and social media, 2013

    2013

  26. [26]

    What makes an image popular? In Proceedings of the 23rd international conference on World wide web, pages 867–876, 2014

    Aditya Khosla, Atish Das Sarma, and Raffay Hamid. What makes an image popular? In Proceedings of the 23rd international conference on World wide web, pages 867–876, 2014

    2014

  27. [27]

    Seismic: A self-exciting point process model for predicting tweet popularity

    Qingyuan Zhao, Murat A Erdogdu, Hera Y He, Anand Rajaraman, and Jure Leskovec. Seismic: A self-exciting point process model for predicting tweet popularity. InProceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining, 2015

    2015

  28. [28]

    Modelling structure and predicting dynamics of discussion threads in online boards.Journal of Complex Networks, 2019

    Alexey N Medvedev, Jean-Charles Delvenne, and Renaud Lambiotte. Modelling structure and predicting dynamics of discussion threads in online boards.Journal of Complex Networks, 2019

    2019

  29. [29]

    Deephawkes: Bridging the gap between prediction and understanding of information cascades

    Qi Cao, Huawei Shen, Keting Cen, Wentao Ouyang, and Xueqi Cheng. Deephawkes: Bridging the gap between prediction and understanding of information cascades. InProceedings of the 2017 ACM on Conference on Information and Knowledge Management, 2017

    2017

  30. [30]

    Deepcas: An end-to-end predictor of information cascades

    Cheng Li, Jiaqi Ma, Xiaoxiao Guo, and Qiaozhu Mei. Deepcas: An end-to-end predictor of information cascades. InProceedings of the 26th international conference on World Wide Web, pages 577–586, 2017

    2017

  31. [31]

    Casseqgcn: Combining network structure and temporal sequence to predict information cascades.Expert Systems with Applications, 2022

    Yansong Wang, Xiaomeng Wang, Yijun Ran, Radosław Michalski, and Tao Jia. Casseqgcn: Combining network structure and temporal sequence to predict information cascades.Expert Systems with Applications, 2022

    2022

  32. [32]

    Content matters: A gnn-based model combined with text semantics for social network cascade prediction

    Yujia Liu, Kang Zeng, Haiyang Wang, Xin Song, and Bin Zhou. Content matters: A gnn-based model combined with text semantics for social network cascade prediction. InPacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, 2021

    2021

  33. [33]

    Conversation modeling on reddit using a graph-structured lstm.Transactions of the Association for Computational Linguistics, 6:121–132, 2018

    Victoria Zayats and Mari Ostendorf. Conversation modeling on reddit using a graph-structured lstm.Transactions of the Association for Computational Linguistics, 6:121–132, 2018

    2018

  34. [34]

    User-guided hierarchical attention network for multi-modal social image popularity prediction

    Wei Zhang, Wen Wang, Jun Wang, and Hongyuan Zha. User-guided hierarchical attention network for multi-modal social image popularity prediction. InProceedings of the 2018 world wide web conference, 2018

    2018

  35. [35]

    Micro-video popularity prediction via multi- modal variational information bottleneck.IEEE Transactions on Multimedia, 2021

    Jiayi Xie, Yaochen Zhu, and Zhenzhong Chen. Micro-video popularity prediction via multi- modal variational information bottleneck.IEEE Transactions on Multimedia, 2021. 11

    2021

  36. [36]

    Predict- ing micro-video popularity via multi-modal retrieval augmentation

    Ting Zhong, Jian Lang, Yifan Zhang, Zhangtao Cheng, Kunpeng Zhang, and Fan Zhou. Predict- ing micro-video popularity via multi-modal retrieval augmentation. InProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2024

    2024

  37. [37]

    Multi-modal variational auto-encoder model for micro-video popularity prediction

    Zhuoran Zhang, Shibiao Xu, Li Guo, and Wenke Lian. Multi-modal variational auto-encoder model for micro-video popularity prediction. InProceedings of the 8th International Conference on Communication and Information Processing, 2022

    2022

  38. [38]

    The pulse of news in social media: Forecasting popularity

    Roja Bandari, Sitaram Asur, and Bernardo Huberman. The pulse of news in social media: Forecasting popularity. InProceedings of the International AAAI Conference on Web and Social Media, 2012

    2012

  39. [39]

    What’s in a hashtag? content based prediction of the spread of ideas in microblogging communities

    Oren Tsur and Ari Rappoport. What’s in a hashtag? content based prediction of the spread of ideas in microblogging communities. InProceedings of the fifth ACM international conference on Web search and data mining, 2012

    2012

  40. [40]

    Image popularity prediction in social media using sentiment and context features

    Francesco Gelli, Tiberio Uricchio, Marco Bertini, Alberto Del Bimbo, and Shih-Fu Chang. Image popularity prediction in social media using sentiment and context features. InProceedings of the 23rd ACM international conference on Multimedia, 2015

    2015

  41. [41]

    Social media popularity prediction: A mul- tiple feature fusion approach with deep neural networks

    Keyan Ding, Ronggang Wang, and Shiqi Wang. Social media popularity prediction: A mul- tiple feature fusion approach with deep neural networks. InProceedings of the 27th ACM International Conference on Multimedia, 2019

    2019

  42. [42]

    Understanding popu- larity, reputation, and social influence in the twitter society.Policy & Internet, 9(3):343–364, 2017

    David Garcia, Pavlin Mavrodiev, Daniele Casati, and Frank Schweitzer. Understanding popu- larity, reputation, and social influence in the twitter society.Policy & Internet, 9(3):343–364, 2017

    2017

  43. [43]

    Generative models of online discussion threads: state of the art and research challenges.Journal of Internet Services and Applications, 2017

    Pablo Aragón, Vicenç Gómez, David García, and Andreas Kaltenbrunner. Generative models of online discussion threads: state of the art and research challenges.Journal of Internet Services and Applications, 2017

    2017

  44. [44]

    Conversations gone awry: Detecting early signs of conversational failure

    Justine Zhang, Jonathan Chang, Cristian Danescu-Niculescu-Mizil, Lucas Dixon, Yiqing Hua, Dario Taraborelli, and Nithum Thain. Conversations gone awry: Detecting early signs of conversational failure. InProceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2018

    2018

  45. [45]

    Analysing how people orient to and spread rumours in social media by looking at conversational threads.PloS one, 2016

    Arkaitz Zubiaga, Maria Liakata, Rob Procter, Geraldine Wong Sak Hoi, and Peter Tolmie. Analysing how people orient to and spread rumours in social media by looking at conversational threads.PloS one, 2016

    2016

  46. [46]

    Popsim: So- cial network simulation for social media popularity prediction.arXiv preprint arXiv:2512.02533, 2025

    Yijun Liu, Wu Liu, Xiaoyan Gu, Allen He, Weiping Wang, and Yongdong Zhang. Popsim: So- cial network simulation for social media popularity prediction.arXiv preprint arXiv:2512.02533, 2025

    work page arXiv 2025

  47. [47]

    arXiv preprint arXiv:2411.11581 , year=

    Ziyi Yang, Zaibin Zhang, Zirui Zheng, Yuxian Jiang, Ziyue Gan, Zhiyu Wang, Zijian Ling, Jinsong Chen, Martz Ma, Bowen Dong, et al. Oasis: Open agent social interaction simulations with one million agents.arXiv preprint arXiv:2411.11581, 2024

    work page arXiv 2024

  48. [48]

    Autocas: Autoregressive cascade predictor in social networks via large language models.arXiv preprint arXiv:2502.18040, 2025

    Yuhao Zheng, Chenghua Gong, Rui Sun, Juyuan Zhang, Liming Pan, and Linyuan Lv. Autocas: Autoregressive cascade predictor in social networks via large language models.arXiv preprint arXiv:2502.18040, 2025

    work page arXiv 2025

  49. [49]

    Forecasting the buzz: Enriching hashtag popularity prediction with llm reasoning

    Yifei Xu, Jiaying Wu, Herun Wan, Yang Li, Zhen Hou, and Min-Yen Kan. Forecasting the buzz: Enriching hashtag popularity prediction with llm reasoning. InProceedings of the 34th ACM International Conference on Information and Knowledge Management, 2025

    2025

  50. [50]

    Smtpd: A new benchmark for temporal prediction of social media popularity

    Yijie Xu, Bolun Zheng, Wei Zhu, Hangjia Pan, Yuchen Yao, Ning Xu, Anan Liu, Quan Zhang, and Chenggang Yan. Smtpd: A new benchmark for temporal prediction of social media popularity. InProceedings of the Computer Vision and Pattern Recognition Conference, 2025. 12

    2025

  51. [51]

    Smp challenge: An overview and analysis of social media prediction challenge

    Bo Wu, Peiye Liu, Wen-Huang Cheng, Bei Liu, Zhaoyang Zeng, Jia Wang, Qiushi Huang, and Jiebo Luo. Smp challenge: An overview and analysis of social media prediction challenge. In Proceedings of the 31st ACM International Conference on Multimedia, 2023

    2023

  52. [52]

    Predicting the popularity of news articles

    Yaser Keneshloo, Shuguang Wang, Eui-Hong Han, and Naren Ramakrishnan. Predicting the popularity of news articles. InProceedings of the 2016 SIAM international conference on data mining. SIAM, 2016

    2016

  53. [53]

    A comparison of methods for cascade prediction

    Ruocheng Guo and Paulo Shakarian. A comparison of methods for cascade prediction. In2016 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM). IEEE, 2016

    2016

  54. [54]

    The anatomy of a large-scale hypertextual web search engine

    Sergey Brin and Lawrence Page. The anatomy of a large-scale hypertextual web search engine. Computer networks and ISDN systems, 1998

    1998

  55. [55]

    The spread of true and false news online.science, 2018

    Soroush V osoughi, Deb Roy, and Sinan Aral. The spread of true and false news online.science, 2018

    2018

  56. [56]

    The structural virality of online diffusion.Management science, 2016

    Sharad Goel, Ashton Anderson, Jake Hofman, and Duncan J Watts. The structural virality of online diffusion.Management science, 2016

    2016

  57. [57]

    Conspiracy vs science: A large-scale analysis of online discussion cascades.World wide web, 2021

    Yafei Zhang, Lin Wang, Jonathan JH Zhu, and Xiaofan Wang. Conspiracy vs science: A large-scale analysis of online discussion cascades.World wide web, 2021

    2021

  58. [58]

    A measurement-driven analysis of information propagation in the flickr social network

    Meeyoung Cha, Alan Mislove, and Krishna P Gummadi. A measurement-driven analysis of information propagation in the flickr social network. InProceedings of the 18th international conference on World wide web, 2009

    2009

  59. [59]

    A survey on predicting the popularity of web content.Journal of Internet Services and Applications, 2014

    Alexandru Tatar, Marcelo Dias De Amorim, Serge Fdida, and Panayotis Antoniadis. A survey on predicting the popularity of web content.Journal of Internet Services and Applications, 2014

    2014

  60. [60]

    i’m in the bluesky tonight

    Andrea Failla and Giulio Rossetti. “i’m in the bluesky tonight”: insights from a year worth of social data.PloS one, 2024

    2024

  61. [61]

    The pushshift reddit dataset

    Jason Baumgartner, Savvas Zannettou, Brian Keegan, Megan Squire, and Jeremy Blackburn. The pushshift reddit dataset. InProceedings of the international AAAI conference on web and social media, volume 14, pages 830–839, 2020

    2020

  62. [62]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021

    2021

  63. [63]

    Statistical physics of social dynamics

    Claudio Castellano, Santo Fortunato, and Vittorio Loreto. Statistical physics of social dynamics. Reviews of modern physics, 2009

    2009

  64. [64]

    Models of social influence: Towards the next frontiers.Journal of Artificial Societies and Social Simulation, 2017

    Andreas Flache, Michael Mäs, Thomas Feliciani, Edmund Chattoe-Brown, Guillaume Deffuant, Sylvie Huet, and Jan Lorenz. Models of social influence: Towards the next frontiers.Journal of Artificial Societies and Social Simulation, 2017

    2017

  65. [65]

    Dynamic models of segregation.Journal of mathematical sociology, 1971

    Thomas C Schelling. Dynamic models of segregation.Journal of mathematical sociology, 1971

    1971

  66. [66]

    Threshold models of collective behavior.American journal of sociology, 1978

    Mark Granovetter. Threshold models of collective behavior.American journal of sociology, 1978

    1978

  67. [67]

    A simple model of global cascades on random networks.Proceedings of the National Academy of Sciences, 99, 2002

    Duncan J Watts. A simple model of global cascades on random networks.Proceedings of the National Academy of Sciences, 99, 2002

    2002

  68. [68]

    Reaching a consensus.Journal of the American Statistical association, 69(345):118–121, 1974

    Morris H DeGroot. Reaching a consensus.Journal of the American Statistical association, 69(345):118–121, 1974

    1974

  69. [69]

    Social influence and opinions.Journal of mathematical sociology, 15(3-4):193–206, 1990

    Noah E Friedkin and Eugene C Johnsen. Social influence and opinions.Journal of mathematical sociology, 15(3-4):193–206, 1990. 13

    1990

  70. [70]

    Information diffusion in online social networks: A survey.ACM Sigmod Record, 42(2):17–28, 2013

    Adrien Guille, Hakim Hacid, Cecile Favre, and Djamel A Zighed. Information diffusion in online social networks: A survey.ACM Sigmod Record, 42(2):17–28, 2013

    2013

  71. [71]

    Large Language Model based Multi-Agents: A Survey of Progress and Challenges

    Taicheng Guo, Xiuying Chen, Yaqi Wang, Ruidi Chang, Shichao Pei, Nitesh V Chawla, Olaf Wiest, and Xiangliang Zhang. Large language model based multi-agents: A survey of progress and challenges.arXiv preprint arXiv:2402.01680, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  72. [72]

    Exploring the limits of weakly supervised pretraining

    Dhruv Mahajan, Ross Girshick, Vignesh Ramanathan, Kaiming He, Manohar Paluri, Yixuan Li, Ashwin Bharambe, and Laurens Van Der Maaten. Exploring the limits of weakly supervised pretraining. InProceedings of the European conference on computer vision (ECCV), 2018

    2018

  73. [73]

    Inductive representation learning on large graphs.Advances in neural information processing systems, 2017

    Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs.Advances in neural information processing systems, 2017

    2017

  74. [74]

    Chapman and Hall/CRC, 1994

    Bradley Efron and Robert J Tibshirani.An introduction to the bootstrap. Chapman and Hall/CRC, 1994

    1994

  75. [75]

    Grad-sam: Explaining transformers via gradient self-attention maps

    Oren Barkan, Edan Hauon, Avi Caciularu, Ori Katz, Itzik Malkiel, Omri Armstrong, and Noam Koenigstein. Grad-sam: Explaining transformers via gradient self-attention maps. InProceedings of the 30th ACM International Conference on Information & Knowledge Management, pages 2882–2887, 2021

    2021

  76. [76]

    Grad-cam: visual explanations from deep networks via gradient-based localization.International journal of computer vision, 128, 2020

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: visual explanations from deep networks via gradient-based localization.International journal of computer vision, 128, 2020

    2020

  77. [77]

    Root Only

    Jacob Gildenblat and contributors. Pytorch library for cam methods.https://github.com/ jacobgil/pytorch-grad-cam, 2021. 14 Appendix Table of Contents A Dataset Details 16 A.1 Curation Details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 A.1.1 Cascade construction. . . . . . . . . . . . . . . . . . . . . . . . . . . 16 A.1.2 Node at...

    work page arXiv 2021