research | Soham De

I'm currently interested in mitigating online harms (such as misinformation, polarisation) by carefully leveraging state-of-the-art language models (LLMs). In parallel, I'm also keen to understand the broader impact of LLMs on online spaces - and what we can do towards bulding a better future.

Much of my past research deals with strategic information operations and political influencing in the Global South. This includes studies on rumours, conspiracies and other forms of online influencing, primarily on X (formerly, Twitter).

ongoing

FeedSpan: A bridging-based prosocial feed reranking algorithm

Luke Thorburn, Soham De, Kayla Duskin, and 4 more authors

🏆 Winning Intervention (Top 3)

In The Prosocial Ranking Challenge 2024

Abstract Web

The algorithm detects civic content that doesn’t attract diverse engagement (not ‘bridging’) and replaces it with similar content that is expected to be more bridging, hopefully reducing divisiveness without decreasing civic content.
Supernotes: Driving consensus in crowd-sourced fact-checking

Soham De, Michiel Bakker, Jay Baxter, and 1 more author

In Submission

Abstract Web Code

X’s Community Notes, a crowd-sourced fact-checking system, allows users to annotate potentially misleading posts. Notes rated as helpful by a diverse set of users are prominently displayed below the original post. While demonstrably effective at reducing misinformation’s impact when notes are displayed, there is an opportunity for notes to appear on many more posts: for 91% of posts where at least one note is proposed, no notes ultimately achieve sufficient support from diverse users to be shown on the platform. This motivates the development of Supernotes: AI-generated notes that synthesize information from several existing community notes and are written to foster consensus among a diverse set of users. Our framework uses an LLM to generate many diverse Supernote candidates from existing proposed notes. These candidates are then evaluated by a novel scoring model, trained on millions of historical Community Notes ratings, selecting candidates that are most likely to be rated helpful by a diverse set of users. To test our framework, we ran a human subjects experiment in which we asked participants to compare the Supernotes generated by our framework to the best existing community notes for 100 sample posts. We found that participants rated the Supernotes as significantly more helpful, and when asked to choose between the two, preferred the Supernotes 75.2% of the time. Participants also rated the Supernotes more favorably than the best existing notes on quality, clarity, coverage, context, and argumentativeness. Finally, in a follow-up experiment, we asked participants to compare the Supernotes against LLMgenerated summaries and found that the participants rated the Supernotes significantly more helpful, demonstrating that both the LLM-based candidate generation and the consensus-driven scoring play crucial roles in creating notes that effectively build consensus among diverse users.

published

Enhancing Crowd-Sourced Fact-Checking Using LLM-Driven Consensus

Soham De, Michiel Bakker, and Martin Saveski

In The 10th International Conference on Computational Social Science 2024

Abstract PDF

One of the main limitations of Community Notes—Twitter’s crowd-sourced fact-checking system—is that it typically takes too long for raters to reach the consensus needed for a note to be attached to a misleading post. We propose speeding up the process by generating super-notes, LLM-generated fact-checking notes that (a) combine key information spread across several existing notes, (b) use language likely to drive consensus among diverse raters, and (c) follow first-order principles of effective fact-checking. Our contributions include a pipeline to generate such supernotes that involves a custom reward-model trained on historical Community Notes data that predicts how a user would rate any note with 70% accuracy. We also outline ongoing and future experiments using human-subjects to evaluate the effectiveness of such an intervention in the wild.
Networks and Influencers in Online Propaganda Events: A Comparative Study of Three Cases in India

Anirban Sen, Soham De, and Joyojeet Pal

In Proceedings of the 27th ACM Conference on Computer-Supported Cooperative Work and Social Computing 2024

Abstract Web PDF

The structure and mechanics of organized outreach around certain issues, such as in propaganda networks, is constantly evolving on social media. We collected tweets on two propaganda events and one non-propaganda event with varying degrees of organized messaging. We then perform a comparative analysis of the user and network characteristics of social media networks around these events and find clearly distinguishable traits across events. We find that influential entities like prominent politicians, digital influencers, and mainstream media prefer to engage more with social media events with lesser degree of propaganda while avoiding events with high degree of propaganda, which are mostly sustained by lesser known but dedicated micro-influencers. We also find that network communities of events with high degree of propaganda are significantly centralized with respect to the influence exercised by their leaders. The methods and findings of this study can pave the way for modeling and early detection of other propaganda events, using their user and community characteristics.
On the Influence and Political Leaning of Overlap between Propaganda Communities

Anirban Sen, Somyasis Gun, Soham De, and 1 more author

In ACM Journal on Computing and Sustainable Societies (JCSS) 2024

Abstract Web PDF

Social media offers increasingly diverse mechanisms for the distribution of motivated information, with multiple propaganda communities exhibiting overlaps with respect to user, content, and network characteristics. This has particularly been an issue in the Global South, where recent work has shown various forms of strife related to polarizing speech online. It has also emerged that propagandist information, including fringe positions on issues, can find its way into the mainstream when sufficiently reinforced in tone and frequency, some of which often requires sophisticated organizing and information manipulation. In this study, we analyze the overlap between three events with varying degrees of propagandist messaging by analyzing the content and network characteristics of users leading to overlap between their users and discourse. We find that a significant fraction of users leading to overlap between the three event communities are influential in information spread across the three event networks, and political leaning is one of the factors that helps explain what brings the communities together. Our work sheds light on the importance of network characteristics of users, which can prove to be instrumental in establishing the role of political leaning on overlap between multiple propaganda communities.
Social Media and Society in India

Soham De, Arshia Arya, Divya Ramesh, and 2 more authors

In Proceedings of SMSI 2023, University of Michigan 2023

Abstract Web PDF

The dramatic increase in access to social media in the last decade in India has brought significant changes to the ways that a range of economic, cultural, and political aspects of social life are conducted. From using YouTube to plan vacations and nightly meals to using Instagram for brand management and Twitter for running political campaigns or social justice outreach, a broad range of engagements and economic activities alike are now either mediated through, or primarily driven by, social media activity. This collection of papers and talks from the proceedings of the Social Media and Society in India conference held at the University of Michigan, Ann Arbor, in April 2023. The talk summaries cover the conversations with practitioners on various domains from the impact of social media on journalism and activism to mental health, culture, and industry. The research papers here include peer reviewed original work presented at the conference on a range of topics including gender, urbanity, financial behavior, family vlogs, misinformation, and hate speech.
Sustaining Discourse in a Rumour Network on Twitter

Soham De, Anirban Sen, and Joyojeet Pal

In The 9th International Conference on Computational Social Science 2023

Abstract PDF

Social media has evolved to be an effective means for individuals to stay relevant and garner a profitable followership online. We study an online conspiracy community formed on Twitter around the death of popular Bollywood star Sushant Singh Rajput in 2020. Much past and continuing research has shown this community as circulating misinformation, conspiracy theories, and propagandist messaging, based on the foundational claim that the actor’s death was not by suicide as claimed by the police. By extension, the messages in the community make accusations against various key figures in the film industry as well as in mainstream politics, attributable to the overarching concept that systemic nepotism in the Hindi film industry aims to undermine, or even murder, outsiders such as Rajput. We identify the core members of this community (N=200) using a snowball-sampling approach, starting from 10 manually identified seed users, collecting over 6 million tweets from 14 June 2020 to 20 April 2022, and find that despite falling off the news cycle, the community remains active on Twitter, driven by three strategies – managing a hierarchical outreach structure, navigating platform algorithms, and managing the discourse around the topic carefully.
Poster: Experiences of Journalists on Twitter in India

Soham De*, Sukhnidh Kaur*, and Joyojeet Pal

🏆 Best Poster Award

In ACM SIGCHI Symposum for HCI and Friends 2022

Web
Closed Ranks: The Discursive Value of Military Support for Indian Politicians

Agrima Seth*, Soham De*, Arshia Arya, and 3 more authors

In Proceedings of the 12th International Conference on Information & Communication Technologies and Development (ICTD) 2022

Abstract Web PDF

Influencers play a crucial role in shaping public narratives through information creation and diffusion in the Global South. While public figures from various walks of life and their impact on public discourse have been studied, defence veterans as influencers of the political discourse have been largely overlooked. Veterans matter in the public spehere as a normatively important political lobby. They are also interesting because, unlike active-duty military officers, they are not restricted from taking public sides on politics, so their posts may provide a window into the views of those still in the service. In this work, we systematically analyze the engagement on Twitter of self-described defence-related accounts and politician accounts that post on defence-related issues. We find that self-described defence-related accounts disproportionately engage with the current ruling party in India. We find that politicians promote their closeness to the defence services and nationalist credentials through engagements with defence-related influencers. We briefly consider the institutional implications of these patterns and connections
Note: Picking Sides: The influencer-driven #HijabBan discourse on Twitter

Soham De, Anmol Panda, and Joyojeet Pal

In Proceedings of the 5th ACM SIGCAS/SIGCHI Conference on Computing and Sustainable Societies (COMPASS) 2022

Abstract Web PDF

Social Media is increasingly central to culture wars, and in recent years has been at the center of debates around the weaponization of peoples’ opinions in polarized situations. We examine one such issue, an attempt to allow schools and junior colleges to ban girls from wearing Hijab in India in the southern state of Karnataka, India, by studying Twitter messaging related to the issue in early 2022, when it was in the news. We find that the narrative supporting the ban of the Hijab on Twitter is primarily driven by a minority of highly-influential individuals who are predominantly male and polarised in favour of the ruling party, while the discourse against the ban relies largely on influencers from within the Muslim community. Our findings show that social media can be a useful tool to craft the contours of politically-motivated escalation in the Global South.
DISMISS: Database of Indian Social Media Influencers on Twitter

Arshia Arya*, Soham De*, Dibyendu Mishra*, and 12 more authors

In Proceedings of the 16th International AAAI Conference on Web and Social Media (ICWSM) 2022

Abstract Web PDF Dataset

Databases of highly networked individuals have been indispensable in studying narratives and influence on social media. To support studies on Twitter in India, we present a systematically categorized database of accounts of influence on Twitter in India, identified and annotated through an iterative process of friends, networks, and self-described profile information, verified manually. We built an initial set of accounts based on the friend network of a seed set of accounts based on real-world renown in various fields, and then snowballed “friends of friends" multiple times, and rank ordered individuals based on the number of in-group connections, and overall followers. We then manually classified identified accounts under the categories of entertainment, sports, business, government, institutions, journalism, civil society accounts that have independent standing outside of social media, as well as a category of “digital first" referring to accounts that derive their primary influence from online activity. Overall, we annotated 11580 unique accounts across all categories. The database is useful studying various questions related to the role of influencers in polarization, misinformation, extreme speech, political discourse etc.

media

Who are BJP’s favourite journalists? What about Congress and AAP? A Michigan study reveals

Soham De, and Joyojeet Pal

Newslaundry 2022

Web
Coordinated campaign? India #1 source for pro-Russia tweets amid Ukraine war, says Ashoka study

Regina Mihindukulasuriya, and Soham De

The Print 2022

Web
Exploring the Intersection between Computer Science and Social Development

Yukti Arora

Ashoka University 2023

Web