Elamur Research⌕
☰
home / papers / 2026 / Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse

Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse

Paper · 2026 · Soorya Ram Shimgekar, Abhay Goyal, Roy Ka-Wei Lee et al.

Summary

Topic models that rely only on semantic similarity often merge conspiratorial and ordinary posts, because both use the same vocabulary about health, politics or money. The authors address this for Singapore-based Telegram groups with a two-step pipeline. First, a RoBERTa-large classifier, fine-tuned on 2,000 balanced expert-labelled messages, tags every post as conspiratorial or not. Next, the posts become nodes in a signed graph. Its edges combine textual similarity with whether two messages share or differ in conspiratorial label. A Signed Belief Graph Neural Network (SiBeGNN), trained with a Sign-Disentanglement Loss, learns embeddings that keep belief stance distinct from general discourse traits. Hierarchical clustering over these embeddings then produces the topics. On 553,648 messages from six groups, the method yields seven clusters and much more compact, better-separated topics than clustering standard embeddings. The substantive finding is that conspiratorial talk appears inside routine themes such as finance, law and everyday chat, not in isolated pockets. This suggests that platform moderation should take topic into account.

Key points

  • Stage 1 classifier: RoBERTa-large reaches F1 = 0.866. It was trained on 1,000 conspiratorial and 1,000 non-conspiratorial messages, labelled using the Diab et al. (2024) definition, with 85% inter-annotator agreement and a third expert resolving disagreements.
  • Corpus: complete histories of six Singapore Telegram groups (e.g., 1M65 Chill Corner, Healing The Divide, SG Corona Freedom Lounge), totalling 553,648 messages and about 24.3M words.
  • Clustering quality: hierarchical clustering on SiBeGNN embeddings scores cDBI = 8.38, compared with 13.60–67.27 for baseline embedding/clustering combinations. Manual cluster interpretation reached 88% inter-rater agreement.
  • Seven topics emerge: Contradictions in Authority, Medical Concerns, Banking and Finance, General Legal Topics, Group Moderation, Media Discussions and General Discussions. Medical Concerns and Contradictions in Authority carry the highest shares of conspiratorial content.
  • The authors plan to release the code and methodology, but access to the raw data stays restricted. Stated limitations include text-only features, heuristic thresholds, and the Singapore-specific context.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 1Research subjects and languages the work is about.Discourse analysis · 466 works

💡 Concepts 1Key ideas and methods explained on our concept pages.Topic modeling · 47 works

Cite

Cite as Soorya Ram Shimgekar, Abhay Goyal, Roy Ka-Wei Lee, Koustuv Saha, Pi Zonooz, Edson C Tandoc Jr and Navin Kumar. 2026. Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse. Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{shimgekar2026belief,
  title = {{Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse}},
  author = {Soorya Ram Shimgekar, Abhay Goyal, Roy Ka-Wei Lee, Koustuv Saha, Pi Zonooz, Edson C Tandoc Jr and Navin Kumar},
  booktitle = {Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science},
  year = {2026},
  url = {https://aclanthology.org/2026.nlpcss-1.20/}
}

References

64 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Mohammad Abdou. 2021. Good governance and covid-19: The digital bureaucracy to response the pandemic (singapore as a model). Journal of Public Affairs, 21(4):e2656. find
  2. AI Singapore. 2024. SEA-LION-v1-7B: A Southeast Asian Multilingual Language Model. https://huggingface.co/aisingapore/SEA-LION-v1-7B. Accessed: 2025-10-07. ↗
  3. Val Alvern Cueco Ligo, Lam Yin Cheung, Roy Ka-Wei Lee, Koustuv Saha, Edson C Tandoc Jr, and Navin Kumar. 2025. User archetypes and information dynamics on telegram: Covid-19 and climate change discourse in singapore. In Companion Proceedings of the ACM on Web Conference 2025, pages 2685– 2688. find
  4. Mathias Angermaier, João Pinheiro-Neto, Elisabeth Hoeldrich, and Jana Lasser. 2025. The schwurbelarchiv: a german language telegram dataset for the study of conspiracy theories. arXiv preprint arXiv:2504.06318. — [arXiv:2504.06318](https://arxiv.org/abs/2504.06318) find
  5. Oscar Araque, Irene Corcuera-Platas, Juan Fernando Sánchez-Rada, and Carlos A. Iglesias. 2017. Enhancing deep learning sentiment analysis with ensemble techniques in social applications. Expert Systems with Applications, 77:236–246. find
  6. Oscar Araque, Lorenzo Gatti, Jacopo Staiano, and Marco Guerini. 2019. Depechemood++: a bilingual emotion lexicon built through simple yet powerful techniques. volume 13, pages 496–507. IEEE. find
  7. Md Rabiul Awal, Minh Dang Nguyen, Roy Ka-Wei Lee, and Kenny Tsu Wei Choo. 2022. Muscat: Multilingual rumor detection in social media conversations. In 2022 IEEE International Conference on Big Data (Big Data), pages 455–464. IEEE. find
  8. V. V. Baligodugula and F. Amsaad. 2025. Unsupervised learning: Comparative analysis of clustering techniques on high-dimensional data. arXiv preprint arXiv:2503.23215. — [arXiv:2503.23215](https://arxiv.org/abs/2503.23215) find
  9. S. Baranwal and 1 others. 2025. Optimizing clustering of cdr3 sequences using nlp embeddings, pca, and k-means. Frontiers in Bioinformatics. find
  10. Ryan L Boyd, Ashwini Ashokkumar, Sarah Seraj, and James W Pennebaker. 2022. The development and psychometric properties of liwc-22. Austin, TX: University of Texas at Austin, pages 1–47. find
  11. Keyu Chen, Marzieh Babaeianjelodar, Yiwen Shi, Kamila Janmohamed, Rupak Sarkar, Ingmar Weber, Thomas Davidson, Munmun De Choudhury, Jonathan Huang, Shweta Yadav, and 1 others. 2023. Partisan us news media representations of syrian refugees. In Proceedings of the International AAAI Conference on Web… find
  12. Susmita Gangopadhyay, Danilo Dessi, Dimitar Dimitrov, and Stefan Dietze. 2025. Telescope: A longitudinal dataset for investigating online discourse and information interaction on telegram. arXiv preprint arXiv:2504.19536. — [arXiv:2504.19536](https://arxiv.org/abs/2504.19536) find
  13. Keyu Chen, Ashley Feng, Rohan Aanegola, Koustuv Saha, Allie Wong, Zach Schwitzky, Roy Ka-Wei Lee, Robin O’Hanlon, Munmun De Choudhury, Frederick L Altice, and 1 others. 2022. Categorizing memes about the ukraine conflict. International Conference on Computational Data and Social Networks, pages 27–3… find
  14. Google DeepMind. 2024. Gemini 2.0 flash: Scaling multimodal foundation models for efficiency and responsiveness. https://deepmind.google/ technologies/gemini/. Accessed: 2025-10-16. ↗
  15. Gang Cheng, Qinliang You, Lei Shi, Zhenxue Wang, Jia Luo, and Tianbin Li. 2023. A survey of topic models: From a whole-cycle perspective. Journal of Intelligent & Fuzzy Systems, 45(6):9929–9953. find
  16. Gene H Golub and Charles F Van Loan. 2013. Matrix computations. JHU press. find
  17. Abhay Goyal, Val Alvern Cueco Ligo, Lam Yin Cheung, Roy Ka-Wei Lee, Koustuv Saha, Edson C Tandoc Jr, and Navin Kumar. 2025. Analyzing conspiratorial content across singapore-based telegram groups. medRxiv, pages 2025–07. find
  18. Abhay Goyal, Muhammad Siddique, Nimay Parekh, Zach Schwitzky, Clara Broekaert, Connor Miche-lotti, Allie Wong, Lam Yin Cheung, Robin O Hanlon, Munmun De Choudhury, and 1 others. 2023. Chatgpt and bard responses to polarizing questions. arXiv preprint arXiv:2307.12402. — [arXiv:2307.12402](https://arxiv.org/abs/2307.12402) find
  19. Daniel Wai Kit Chin, Kwan Hui Lim, and Roy Ka-Wei Lee. 2024. Rumorgraphxplainer: Do structures really matter in rumor detection. IEEE Transactions on Computational Social Systems, 11(5):6038–6055. find
  20. Eunsol Choi, Hannah Rashkin, Luke Zettlemoyer, and Yejin Choi. 2016. Document-level sentiment inference with social, faction, and discourse context. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 333–343. find
  21. Avihay Chriqui and Inbal Yahav. 2022. Hebert and hebemo: A hebrew bert model and a tool for polarity analysis and emotion recognition. INFORMS Journal on Data Science, 1(1):81–95. find
  22. Tyler Derr, Yao Ma, and Jiliang Tang. 2018. Signed graph convolutional network. arXiv preprint arXiv:1808.06354. — [arXiv:1808.06354](https://arxiv.org/abs/1808.06354) find
  23. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. NAACL-HLT.
  24. Ahmad Diab, Rr Nefriana, and Yu-Ru Lin. 2024. Classifying conspiratorial narratives at scale: False alarms and erroneous connections. In Proceedings of the International AAAI Conference on Web and Social Media, volume 18, pages 340–353. find
  25. Justin Grimmer and Brandon M Stewart. 2013. Text as data: The promise and pitfalls of automatic content analysis methods for political texts. Political analysis, 21(3):267–297. find
  26. Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2021. Deberta: Decoding-enhanced bert with disentangled attention. arXiv preprint arXiv:2006.03654.
  27. Mohamad Hoseini, Philipe Melo, Fabrício Benevenuto, Anja Feldmann, and Savvas Zannettou. 2021. On the globalization of the qanon conspiracy theory through telegram. arXiv preprint arXiv:2105.13020. — [arXiv:2105.13020](https://arxiv.org/abs/2105.13020) find
  28. X. Hu, Jiliang Tang, Yao Ma, and Yu Wang. 2024. Disentangled generative graph representation learning (diggr). arXiv preprint arXiv:2408.13471. — [arXiv:2408.13471](https://arxiv.org/abs/2408.13471) find
  29. I. K. O. Jabari and 1 others. 2024. Learning-augmented k-means clustering using dimensional reduction. arXiv preprint arXiv:2401.03198. — [arXiv:2401.03198](https://arxiv.org/abs/2401.03198) find
  30. Fedor Krasnov and Anastasiia Sen. 2019. The number of topics optimization: Clustering approach. Machine Learning and Knowledge Extraction, 1(1):25. find
  31. Paul DiMaggio, Manish Nag, and David Blei. 2013. Exploiting affinities between topic modeling and the sociological perspective on culture: Application to newspaper coverage of us government arts funding. Poetics, 41(6):570–606. find
  32. Srijan Kumar, Francesca Spezzano, VS Subrahmanian, and Christos Faloutsos. 2016. Edge weight prediction in weighted signed networks. In 2016 IEEE 16th international conference on data mining (ICDM), pages 221–230. IEEE. find
  33. Karen M Douglas, Robbie M Sutton, and Aleksandra Cichocka. 2017. The psychology of conspiracy theories. Current directions in psychological science, 26(6):538–542. find
  34. H. Li, H. Peng, Z. Liu, L. He, J. Wu, and S. Yu Philip. 2021. Disentangled graph contrastive learning. Advances in Neural Information Processing Systems (NeurIPS). find
  35. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692.
  36. Ketan Rajshekhar Shahapure and Charles Nicholas. 2020. Cluster quality analysis using silhouette score. In 2020 IEEE 7th international conference on data science and advanced analytics (DSAA), pages 747– 748. IEEE. find
  37. Jörn Lötsch. 2024. Comparative assessment of projection and clustering combinations. ScienceDirect. find
  38. Soorya Ram Shimgekar, Abhay Goyal, Shayan Vassef, Koustuv Saha, Christian Poellabauer, Xavier Vautier, Pi Zonooz, and Navin Kumar. 2025a. Nimblelabs: Accelerating healthcare ai development through agentic ai. preprints.org preprints:202508.1713.v1. find
  39. Jian Ma, Jiliang Tang, Qiaozhi Zhu, and Qiaozhu Mei. 2019a. Disentangled graph convolutional networks (disengcn). In Proceedings of Machine Learning Research, Vol. 97, pages 5–19. find
  40. Jianxin Ma, Peng Cui, Kun Kuang, Xin Wang, and Wenwu Zhu. 2019b. Disentangled graph convolutional networks. In International conference on machine learning, pages 4212–4221. PMLR. find
  41. Raija Markkanen and Hartmut Schröder. 1997. Hedging: A challenge for pragmatics and discourse analysis. In Hedging and Discourse: Approaches to the Analysis of a Pragmatic Phenomenon in Academic Texts, pages 3–18. Walter de Gruyter. find
  42. Meta AI. 2024. The llama 3.2 model family. https://ai.meta.com/llama/. Accessed: 2025-10-16. ↗
  43. Fionn Murtagh. 1983. A Survey of Recent Advances in Hierarchical Clustering Algorithms, volume 26. Oxford University Press. find
  44. Yashna Nainani, Kaveh Khoshnood, Ashley Feng, Muhammad Siddique, Clara Broekaert, Allie Wong, Koustuv Saha, Roy Ka-Wei Lee, Zachary M Schwitzky, Lam Yin Cheung, and 1 others. 2022. Categorizing memes about abortion. In The International Conference on Weblogs and Social Media. find
  45. Lynnette Hui Xian Ng and Jia Yuan Loke. 2020. Analyzing public opinion and misinformation in a covid-19 telegram group chat. IEEE Internet Computing, 25(2):84–91. find
  46. Milena Pustet, Elisabeth Steffen, and Helena Mihaljević. 2024. Detection of conspiracy theories beyond keyword bias in german-language telegram using large language models. arXiv preprint arXiv:2404.17985. find
  47. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 3982–3992.
  48. Markus Reiter-Haas, Beate Klösch, Markus Hadler, and Elisabeth Lex. 2024. Framing analysis of health-related narratives: Conspiracy versus mainstream media. arXiv preprint arXiv:2401.10030. — [arXiv:2401.10030](https://arxiv.org/abs/2401.10030) find
  49. Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108. find
  50. Soorya Ram Shimgekar, Shayan Vassef, Abhay Goyal, Navin Kumar, and Koustuv Saha. 2025b. Agentic ai framework for end-to-end medical data inference. arXiv preprint arXiv:2507.18115. — [arXiv:2507.18115](https://arxiv.org/abs/2507.18115) find
  51. Aelita Skarzauskiene, Monika Maciuliene, Aiste Dirzyte, and Gintare Guleviciute. 2025. Profiling antivaccination channels in telegram: early efforts in detecting misinformation. Frontiers in Communication, 10:1525899. find
  52. Swapna Somasundaran and Janyce Wiebe. 2010. Recognizing stances in ideological on-line debates. In Proceedings of the NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, pages 116–124.
  53. Edson C. Jr. Tandoc, Kaidi Jenkins, and Rachel Neo. 2021. Developing a perceived social media literacy scale: Evidence from singapore. International Journal of Communication, 15:16118. find
  54. Simone Teufel. 2006. Argumentative zoning for improved citation indexing. find
  55. Aleksandra Urman and Stefan Katz. 2020. What they do in the shadows: examining the far-right networks on telegram. Information, Communication & Society, 25(7):904–923. find
  56. Joseph E Uscinski and Joseph M Parent. 2014. American conspiracy theories. Oxford University Press. find
  57. Tran Hien Van, Abhay Goyal, Muhammad Siddique, Lam Yin Cheung, Nimay Parekh, Jonathan Y Huang, Keri McCrickerd, Edson C Tandoc Jr, Gerard Chung, and Navin Kumar. 2023. How is fatherhood framed online in singapore? arXiv preprint arXiv:2307.04053. — [arXiv:2307.04053](https://arxiv.org/abs/2307.04053) find
  58. Teun A. van Dijk. 1998. Ideology: A Multidisciplinary Approach. Sage Publications, London. find
  59. Samantha Walther and Andrew McCoy. 2021. Us extremism on telegram: Fueling disinformation, conspiracy theories, and accelerationism. Perspectives on Terrorism, 15(2):41–60. find
  60. Hanlin Xiao, Mauricio A Álvarez, and Rainer Breitling. 2026. Disentangling similarity and relatedness in topic models. arXiv preprint arXiv:2603.10619. — [arXiv:2603.10619](https://arxiv.org/abs/2603.10619) find
  61. Jing Zeng, Mike S Schäfer, and Thaiane M Oliveira. 2022. Conspiracy theories in digital environments: Moving the research field forward. Convergence, 28(4):929–939. find
  62. Ziwei Zhang, Chuan Shi, Jieyu Yang, and Yao Ma. 2024. Signed graph representation learning: A survey. arXiv preprint arXiv:2402.15980. — [arXiv:2402.15980](https://arxiv.org/abs/2402.15980) find
  63. Ziwei Zhang, Jieyu Yang, and Chuan Shi. 2021. Signed graph neural network with latent groups (gs-gnn). In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (KDD). find
  64. Fabiana Zollo, Alessandro Bessi, Michela Del Vicario, Antonio Scala, Guido Caldarelli, Louis Shekhtman, Shlomo Havlin, and Walter Quattrociocchi. 2017. Debunking in a world of tribes. PloS one, 12(7):e0181821. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shimgekar-2026-belief-is-all-you-need/