@inproceedings{shimgekar-etal-2026-belief,
title = "Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse",
author = "Shimgekar, Soorya Ram and
Goyal, Abhay and
Lee, Roy Ka-Wei and
Saha, Koustuv and
Zonooz, Pi and
Tandoc Jr, Edson C and
Kumar, Navin",
editor = "Card, Dallas and
Field, Anjalie and
Keith, Katherine and
Mendelsohn, Julia",
booktitle = "Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science",
month = jul,
year = "2026",
address = "San Diego",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.nlpcss-1.20/",
pages = "341--355",
ISBN = "979-8-89176-426-2",
abstract = "Conspiratorial discourse is increasingly present in online communication, yet how it is organized across discussion topics remains unclear. We analyze Singapore-based Telegram groups to examine how conspiratorial content appears within everyday conversations rather than isolated echo chambers. To better capture the structure of such discussions, we propose a two-stage framework for topic modeling tailored to conspiratorial posts. First, a RoBERTa-large classifier identifies conspiratorial messages (F1 = 0.866) using 2,000 expert-annotated examples. We then construct a graph where connections reflect textual similarity and conspiratorial stance. This graph is modeled using a Signed Belief Graph Neural Network (SiBeGNN), which learns message embeddings that distinguish conspiratorial from non-conspiratorial content. We apply hierarchical clustering on these embeddings to perform topic modeling over 553,648 Telegram messages, producing seven topic clusters: General Legal Topics, Medical Concerns, Media Discussions, Banking and Finance, Contradictions in Authority, Group Moderation, and General Discussions. Our method substantially outperforms standard embedding-based clustering approaches (cDBI = 8.38 vs. 13.60{--}67.27), with manual evaluation showing 88{\%} inter-rater agreement in cluster interpretation. The results show that conspiratorial content appears across multiple everyday topics, including finance, law, and daily life, rather than forming isolated thematic communities."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="shimgekar-etal-2026-belief">
<titleInfo>
<title>Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse</title>
</titleInfo>
<name type="personal">
<namePart type="given">Soorya</namePart>
<namePart type="given">Ram</namePart>
<namePart type="family">Shimgekar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Abhay</namePart>
<namePart type="family">Goyal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Roy</namePart>
<namePart type="given">Ka-Wei</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Koustuv</namePart>
<namePart type="family">Saha</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pi</namePart>
<namePart type="family">Zonooz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Edson</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Tandoc Jr</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Navin</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dallas</namePart>
<namePart type="family">Card</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anjalie</namePart>
<namePart type="family">Field</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Katherine</namePart>
<namePart type="family">Keith</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Julia</namePart>
<namePart type="family">Mendelsohn</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-426-2</identifier>
</relatedItem>
<abstract>Conspiratorial discourse is increasingly present in online communication, yet how it is organized across discussion topics remains unclear. We analyze Singapore-based Telegram groups to examine how conspiratorial content appears within everyday conversations rather than isolated echo chambers. To better capture the structure of such discussions, we propose a two-stage framework for topic modeling tailored to conspiratorial posts. First, a RoBERTa-large classifier identifies conspiratorial messages (F1 = 0.866) using 2,000 expert-annotated examples. We then construct a graph where connections reflect textual similarity and conspiratorial stance. This graph is modeled using a Signed Belief Graph Neural Network (SiBeGNN), which learns message embeddings that distinguish conspiratorial from non-conspiratorial content. We apply hierarchical clustering on these embeddings to perform topic modeling over 553,648 Telegram messages, producing seven topic clusters: General Legal Topics, Medical Concerns, Media Discussions, Banking and Finance, Contradictions in Authority, Group Moderation, and General Discussions. Our method substantially outperforms standard embedding-based clustering approaches (cDBI = 8.38 vs. 13.60–67.27), with manual evaluation showing 88% inter-rater agreement in cluster interpretation. The results show that conspiratorial content appears across multiple everyday topics, including finance, law, and daily life, rather than forming isolated thematic communities.</abstract>
<identifier type="citekey">shimgekar-etal-2026-belief</identifier>
<location>
<url>https://aclanthology.org/2026.nlpcss-1.20/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>341</start>
<end>355</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings %T Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse %A Shimgekar, Soorya Ram %A Goyal, Abhay %A Lee, Roy Ka-Wei %A Saha, Koustuv %A Zonooz, Pi %A Tandoc Jr, Edson C. %A Kumar, Navin %Y Card, Dallas %Y Field, Anjalie %Y Keith, Katherine %Y Mendelsohn, Julia %S Proceedings of the Seventh Workshop on Natural Language Processing and Computational Social Science %D 2026 %8 July %I Association for Computational Linguistics %C San Diego %@ 979-8-89176-426-2 %F shimgekar-etal-2026-belief %X Conspiratorial discourse is increasingly present in online communication, yet how it is organized across discussion topics remains unclear. We analyze Singapore-based Telegram groups to examine how conspiratorial content appears within everyday conversations rather than isolated echo chambers. To better capture the structure of such discussions, we propose a two-stage framework for topic modeling tailored to conspiratorial posts. First, a RoBERTa-large classifier identifies conspiratorial messages (F1 = 0.866) using 2,000 expert-annotated examples. We then construct a graph where connections reflect textual similarity and conspiratorial stance. This graph is modeled using a Signed Belief Graph Neural Network (SiBeGNN), which learns message embeddings that distinguish conspiratorial from non-conspiratorial content. We apply hierarchical clustering on these embeddings to perform topic modeling over 553,648 Telegram messages, producing seven topic clusters: General Legal Topics, Medical Concerns, Media Discussions, Banking and Finance, Contradictions in Authority, Group Moderation, and General Discussions. Our method substantially outperforms standard embedding-based clustering approaches (cDBI = 8.38 vs. 13.60–67.27), with manual evaluation showing 88% inter-rater agreement in cluster interpretation. The results show that conspiratorial content appears across multiple everyday topics, including finance, law, and daily life, rather than forming isolated thematic communities. %U https://aclanthology.org/2026.nlpcss-1.20/ %P 341-355
Markdown (Informal)
[Belief Is All You Need: Signed Belief Graph Neural Networks for Topic Modeling in Conspiratorial Discourse](https://aclanthology.org/2026.nlpcss-1.20/) (Shimgekar et al., NLP+CSS 2026)