publications | Mehtaverse

For details see Google Scholar

2025

Market Design Interventions for Safer Agentic AI

Swapneel Mehta, Aaron Nichols, Nina Mazar, and Marshall van Alstyne

International Conference on Information Systems (ICIS), 2025

Abs HTML PDF

Our research investigates safety considerations necessary to govern the deployment of Generative AI agents on e-commerce platforms against abuse by fraudulent sellers. Fraudulent advertisements that overpromise product quality lead to the purchase of subpar goods while platforms lack effective recourse mechanisms for consumers. Online reputation systems such as buyer-provided ratings for sellers attempt to provide a mechanism for verified buyers of a product to register their feedback in a structured way that may inform the future purchase of goods. However reviews and reputation are often gamed using bots and fake accounts, constituting “black hat” seller practices. We design an experiment (n = 250 human participants, 5000 rounds of advertised product sales) in order to demonstrate the occurrence of agentic deception in a two-sided e-commerce marketplace and show how a novel economic intervention involving staked claims limits deceptive strategies to promote honesty in the marketplace.
Identifying the Causal Effects of Twitter’s Interventions on Trump’s Tweets

Swapneel Mehta, James Bisbee, Zeve Sanderson, Richard Bonneau, Joshua Tucker, and Jonathan Nagler

Working Paper, Target: PNAS, May, 2025

Abs PDF

Misinformation has emerged as a key challenge for democratic politics, particularly during election periods. In response, social media platforms have developed and deployed interventions aimed at mitigating the diffusion and harmful effects of misinformation. However, there is limited causal evidence of intervention efficacy using real-world observational data, undermining our ability to identify the extent to which platform strategies produce their intended outcomes. Here, we measure the causal effect of Twitter’s soft and hard interventions on the within- and cross-platform spread of Trump’s misinformation tweets during the 2020 election. We find evidence of a Streisand effect on Twitter itself, with labeled tweets experiencing more engagement than similar unlabeled tweets. We identify the same messages across Facebook, Instagram, and Reddit, finding that Twitter’s interventions have heterogeneous effects on the spread of the same messages across other platforms.
Certifiably True: The Impact of Self-Certification on Misinformation

Aaron Nichols, Nina Mazar, Swapneel Mehta, Tejovan Parker, Gordon Pennycook, David Rand, and Marshall van Alstyne

Nature Communications (Revise and Resubmit), 2025

Abs

Combating the spread of misinformation at scale requires platform-level tools that disincentivize the sharing of misinformation without censorship. Across three pre-registered online experiments (N = 5,277 participants; 127,824 responses), we test the potential of Truth Warrants– a decentralized mechanism that allows individuals to voluntarily signal that the information they share is true by vouchsafing its accuracy and accepting financial responsibility. In Experiment 1, we investigate how warrants affect social media users’ news sharing behavior. Experiment 2 tests whether warrants uniquely affect perceptions of news accuracy, comparing their impact to other news sharing signals. Experiment 3 replicated and extended results of Experiments 1 and 2, generalizing to more realistic settings where participants gain experience with the warrant system before evaluating the accuracy of news headlines. Results indicated that introducing truth warrants increased the quality of headlines shared, the overall number of headlines shared, and the perceived accuracy of headline claims. Exploratory analysis revealed that, truth warrants also increased sharing quality for decisions that they did not endorse as true. Furthermore, although truth warrants affected the sharing behavior of both politically polarized and moderate participants, warrants uniquely affected perceptions of accuracy for politically moderate participants. Our findings demonstrate that truth warrants offer a promising tool for combating misinformation without censorship, while also increasing engagement.
Curbing Strategic Deception in Digital Markets

Swapneel Mehta, Aaron Nichols, Nina Mazar, and Marshall van Alstyne

Workshop on Information Systems and Economics (Oral Talk), 2025

HTML
A Framework for Social Media Safety: Comparing Platform Interventions

Caitlyn Vergara, Jhagrut Lalwani, Hriday Ranka, Neel Kothari, and Swapneel Mehta

Research Conference on Communications, Information and Internet Policy (TPRC), 2025

Abs HTML PDF

Social media platforms deploy a wide variety of interventions to reduce online harms, yet efforts to compare or assess these measures are hampered by fragmented definitions and inconsistent evaluation methods across the industry. Our research addresses this gap by introducing an action-oriented framework that organizes digital safety interventions according to four key parameters: focus, scope, driver, and user journey. Through a systematic analysis of recent interventions across youth-focused social platforms, including TikTok, Instagram, and YouTube, we show how this taxonomy creates a common language for classifying interventions and enables more consistent assessment of their design and impact. Our case study demonstrates that the proposed framework not only streamlines cross- platform comparisons, but also supplies actionable guidance for designers and policymakers to identify gaps, measure effectiveness, and build safer digital environments. By bridging inconsistencies in language and approach, this work lays a pragmatic foundation for stronger evidence-based policymaking and supports collaborative efforts to address evolving challenges in the digital safety landscape.
Accelerating Narrative Identification in Conflict Regions with Artificial Intelligence

Haris Naveed, and Swapneel Mehta

Working Paper (collaboration with global development agency), 2025

Abs PDF

Intelligence-gathering in conflict regions is of vital importance to the success of international humanitarian missions in some of the most sensitive conflict regions in the world. Human analysts working with developmental agencies routinely spend several hours manually searching through transcripts for entities of interest, summarizing their findings, and validating their sources. We accelerate this process with a retrieval augmented generation-based search system that recovers transcripts that are most relevant to an input query, and frames a response based on the facts presented in the transcripts. We propose a novel question-answering system to support narrative identification in conflict regions using audio datasets. Our system not only produces an answer but also highlights its context in audio clips, along with the source citations used to generate the answer. We employ different retrieval modules, testing the efficiency of our system for achieving high-quality search result generation, composing a novel GraphRAG-based system with a multi-query retriever that improves retrieval performance over other methods. We compose an audio dataset from radio broadcasts across countries in Africa and show that using our system could reduce the time spent by human analysts tasked with intelligence-gathering from hours to minutes without compromising on the quality of output they produce.
Estimating the Effects of Blocking on User Behavior

Dhvani Shah, Shristi Shetty, and Swapneel Mehta

Working Paper; Stanford TSRC (Poster), 2025

Abs PDF

Decentralized social media platforms like Bluesky offer protocols for open data access and empower researchers with a unique lens to study the protective behaviors such as blocking of other accounts by a user on the platform. Our study examines how the issuance of a block by one account to another shapes online activity on the platform. Using a difference-in-differences (DiD) approach, we find that blocking increases user activity on the platform. We employ topic modeling to identify shifts in the themes of posts after they issue a block while controlling for user attributes and content engagement. Through an Interrupted Time Series (ITS) analysis to verify the robustness of our assumptions, we find that the mentions of a user’s account handle do not influence the issuance of blocks on Bluesky. Our findings underscore how blocking increases the volume of posts that a user creates on the platform and influences the diversity of information and external content shared. This contributes to a novel understanding of how user level safety controls influence user activity and content production on Bluesky.
Who Trumps What: News ’Republication’ on Truth Social

Gautam Malpani, and Swapneel Mehta

Working Paper; Stanford TSRC (Lightning Talk), 2025

Abs HTML

News-sharing on social media platforms frequently involves republication of the same links shared in prior posts, with new opinions and commentary, without explicitly citing the original post. Although prior research underscores news quality as a key determinant of its virality on related platforms like X (Twitter), we find that user profile attributes, including their follower count, tenure on the platform, and history of audience engagement, may supersede the quality of the news source in predicting how it is republished and amplified on the platform. We collect over two million posts from over 100,000 Truth Social users (Feb 2022 – Feb 2024) to investigate the driving factors for news republication. For news links shared within posts on Truth Social, we find that news may not gain prominence the first time a post publishes the article on the platform but often rises in visibility upon being republished. Users with a large following and a longer platform tenure cause the increased visibility of news, primarily sourced from low-quality providers, which receives nearly a three times higher volume of engagement than news from high-quality providers. News from dubious sources receives 33% of online engagement and is reposted between 2 and 3 times more often than highly credible news sources. We demonstrate how dynamics consistent with echo-chamber effects and social capital theory create information cascades on Truth Social. Exploring the drivers of republished content popularity is critical for understanding the information consumption patterns of users on this platform. Our results underscore how news republication departs from traditional notions of news visibility on social media platforms.
Multi-agent Debates for Deliberative Content Moderation on X

Viral Dalal, Aaryamonvikram Singh, and Swapneel Mehta

Working Paper; Stanford TSRC (Poster), 2025

HTML
Truth Warrants Increase Economic Value and Accelerate Product Sales in Digital Marketplaces

Swapneel Mehta, Aaron Nichols, Nina Mazar, and Marshall van Alstyne

BU Platform Strategy Research Symposium, 2025

Abs PDF

Misleading advertisements in digital marketplaces deceive buyers of online goods, posing significant challenges to platform integrity and buyer protection. Existing content moderation and reputation systems have proven insufficient in addressing the root incentives that drive false advertising. This paper introduces a novel market-based economic intervention, “truth warrants” that allows online advertisers to guarantee the accuracy of their claims, and presents buyers with the ability to seek recourse when they are misled. Through a series of controlled online experiments in a competitive two-sided marketplace, we demonstrate the economic value of truth warrants in comparison to existing reputation and rating systems. The introduction of truth warrants penalizes cheating advertisers that make false claims, and promotes accountability among online advertisers, without affecting the profits or sales of honest advertisers. Empirical results(n = 250 users and 5208 rounds of product sales) and demonstrate that truth warrants nearly double the economic value provided by reputation signals and reduce the time it takes for the first sale of a product by nearly half the full duration of our experiment.

2024

A history of transparency regulations: interdisciplinary strategies for shaping social media regulation and self-governance

Caitlyn Vergara, Raghav Jain, and Swapneel Mehta

In Proceedings of the 25th Annual International Conference on Digital Government Research, 2024

Abs PDF

Widespread calls for social media transparency have resulted in the passage of landmark regulations across the globe to hold social media platforms accountable and improve online safety. In response, countries have enacted acts, such as the Online Safety Act by the United Kingdom (2023) and the Digital Services Act by the European Union (2023). While regulation was hailed as a much-needed governance lever to enforce accountability mechanisms on monopolistic social platforms, there remain significant challenges to navigate regarding its long-term influence and efficacy. One of the challenges through this process has been the development of equitable standards and a shared understanding of platform mechanisms, for which there has been significant consultation between the public, platforms, policymakers, and nongovernmental organizations. While regulating social platforms is a novel example, the prevalence of consumer harms in an opaque, unregulated ecosystem has occurred in various prior instances in the fields of healthcare, finance, and aviation in the United States. In each of these fields, the harms perpetrated by the inefficiencies of the system resulted in calls for regulation and transparency at a similar or greater magnitude as that of social media. We draw parallels between the harms generated and corresponding safety regulations across different industries, with the goal of highlighting the challenges navigated by past regulatory efforts to introduce transparency into each of these complex ecosystems. Our work provides one of the first comparative analyses of a set of flagship regulations in adjacent fields that can inform the development of upcoming regulations aimed at improving transparency efforts by digital platforms.
Improving Economic Welfare in Reputation-based Marketplaces with Truth Warrants

Swapneel Mehta, and Marshall van Alstyne

Workshop on Information Systems and Economics (Oral Talk), 2024

HTML
Market Design Interventions for Safer Agentic AI

Swapneel Mehta, Aaron Nichols, Nina Mazar, and Marshall van Alstyne

Columbia MarkTech Conference; Yale AI/ML Conference (Oral Talks), 2024

HTML
Examining the Implications of Deepfakes for Election Integrity

Hriday Ranka, Mokshit Surana, N. Kothari, V. Pariawala, Pratyay Banerjee, A. Surve, S.R. Sankepally, Raghav Jain, J. Lalwani, and Swapneel Mehta

In AAAI Workshop on AI for Credible Elections, 2024

Abs PDF

It is becoming cheaper to launch disinformation operations at scale using AI-generated content, in particular ’deepfake’ technology. We have observed instances of deepfakes in political campaigns, where generated content is employed to both bolster the credibility of certain narratives (reinforcing outcomes) and manipulate public perception to the detriment of targeted candidates or causes (adversarial outcomes). We discuss the threats from deepfakes in politics, highlight model specifications underlying different types of deepfake generation methods, and contribute an accessible evaluation of the efficacy of existing detection methods. We provide this as a summary for lawmakers and civil society actors to understand how the technology may be applied in light of existing policies regulating its use. We highlight the limitations of existing detection mechanisms and discuss the areas where policies and regulations are required to address the challenges of deepfakes.
How Decentralization Affects User Agency on Social Platforms

Aditya Surve, Aneesh Shamraj, and Swapneel Mehta

In Data Challenge, AAAI International Conference on Web and Social Media (ICWSM), 2024

Abs PDF

Mainstream social media platforms function as "walled garden" ecosystems that restrict user agency, control, and data portability. They have demonstrated a lack of transparency that contributes to a multitude of online harms. Our research investigates how decentralization might present promise as an alternative model to walled garden platforms. Specifically, we describe the user-driven content moderation through blocks as an expression of agency on Bluesky, a decentralized social platform. We examine the impact of providing users with more granular control over their online experiences, including what they post, who can see it, and whose content they are exposed to. We describe the patterns identified in user-driven content moderation and suggest directions for further research.
Can Social Media Platforms Transcend Political Labels? An Analysis of Neutral Conversations on Truth Social

Chaitya Shah, Ritesh Konka, Gautam Malpani, Swapneel Mehta, and Lynnette Ng

In DARE Workshop, AAAI International Conference on Web and Social Media (ICWSM), 2024

Abs PDF

There is a prevailing perception that content on a social media platform generally have the same political leaning. These platforms are often viewed as ideologically congruent entities, reflecting the majority opinion of their users; a prime example of this is Truth Social. While this perception may exist, it is essential to verify the platform’s credibility, acknowledging that such platforms contain meaningful insights with neutral stances. To this end, we examine the dissemination of Wikipedia links on the alt-right platform, Truth Social. Wikipedia is recognized for enforcing content neutrality and serves as a unique lens to analyze the objectivity of user-generated content on Truth Social. By scrutinizing Truths with and without Wikipedia links, identifying toxicity trends & recognizing coordinated networks, we observe a lower level of engagement and a tendency for Truths shared on Truth Social to cover more neutral topics when it includes Wikipedia links (Wiki Truths). Given the significantly different engagement and nature of content shared of Wiki Truths against Non-Wiki Truths, we emphasize that we should not generalize the techno-political affiliation of a social media platform, but rather should investigate the content closely.
Reducing Misleading Claims in Digital Marketplaces using Truth Warrants

Swapneel Mehta, Zoey Yang, Changxuan Fan, Daian Liu, Nina Mazar, and Marshall van Alstyne

IC2S2 (Oral Talk), 2024

2023

Multiagent simulators for social networks

Aditya Surve, Archit Rathod, Mokshit Surana, Gautam Malpani, Aneesh Shamraj, Sainath Reddy Sankepally, Raghav Jain, and Swapneel S Mehta

ICML Workshop on AI for Agent-based Modeling, 2023

Abs PDF

Multiagent social network simulations are an avenue that can bridge the communication gap between the public and private platforms in order to develop solutions to a complex array of issues relating to online safety. While there are significant challenges relating to the scale of multiagent simulations, efficient learning from observational and interventional data to accurately model micro and macro-level emergent effects, there are equally promising opportunities not least with the advent of large language models that provide an expressive approximation of user behavior. In this position paper, we review prior art relating to social network simulation, highlighting challenges and opportunities for future work exploring multiagent security using agent-based models of social networks

2022

Compartmental Models for COVID-19 and Control via Policy Interventions

Swapneel Mehta, and Noah Kasmanoff

arXiv preprint arXiv:2203.02860, 2022

PDF
Estimating the Impact of Coordinated Inauthentic Behavior on Content Recommendations in Social Networks

Swapneel S Mehta, Atilim Gunes Baydin, Richard Bonneau, Jonathan Nagler, and Philip Torr

In ICML 2022 Workshop AI for Agent-Based Modelling, 2022

Abs PDF

Online disinformation is a dynamic and pervasive problem on social networks as evidenced by a spate of public disasters in light of active efforts to combat it. Since the massive amounts of content generated each day on these platforms is impossible to manually curate, ranking and recommendation algorithms are a key apparatus that drive user interactions. However, the vulnerability of ranking and recommendation algorithms to attack from coordinated campaigns spreading misleading information has been established both theoretically and anecdotally. Unfortunately it is unclear how effective countermeasures to disinformation are in practice due to the limited view we have into the operation of such platforms. We develop a multiagent simulation of a popular social network, Reddit, that aligns with the state-action space available to real users based on the platform’s affordances. We collect millions of real-world interactions from Reddit to estimate the network for each user in our dataset and utilise Reddit’s self-described content ranking strategies to compare the impact of coordinated activity on content spread by each algorithm. We expect that this will inform the design of robust content distribution systems that are resilient against targeted attacks by groups of malicious actors.
Expanding Access to ML Research through Student-led Collaboratives

Deep Gandhi, Raghav Jain, Jay Gala, Jhagrut Lalwani, and Swapneel S Mehta

In NeurIPS 2022 Workshop on Broadening Research Collaborations, 2022

Abs PDF

We provide concrete examples of the guiding principles that led to the evolution of the collaborative from a reading group into a research group and eventually launching a non-profit software product to help non-technical stakeholders leverage artificial intelligence (AI), improving access to advanced technologies, and promoting open science. Our goal is to lay out a template to launch similar small-scale collaborative organisations at different institutes around the world.

2021

Open-domain Trending Hashtag Recommendation for Videos

Swapneel Mehta, Somdeb Sarkhel, Xiang Chen, Saayan Mitra, Vishwanathan Swaminathan, Ryan Rossi, Ali Aminian, Han Guo, and Kshitiz Garg

In IEEE International Symposium on Multimedia (ISM), 2021

Abs PDF

We describe a novel algorithm for an open-domain trending hashtag recommendation task using zero-shot hashtag prediction in an online learning paradigm. Our method utilizes joint representation learning of latent embeddings for features extracted from long-form videos and semantic embeddings of hashtags trending on social platforms. In particular, we apply graph convolutional networks to a link prediction task using videos and hashtags as nodes in a heterogeneous graph. Comparing it to the existing models for closely related tasks, we demonstrate state-of-the-art results in trending hashtag recommendations for videos. The architecture is designed to be modular in a plug-and-play fashion to enable quick and easy incorporation of the latest advances in natural language understanding and image and video processing, with a practical view to its implementation in a real-time, online setting.

2020

Deepjetcore

J Kieseler, M Stoye, M Verzetti, P Silva, SS Mehta, A Stakia, Y Iiyama, E Bols, SR Qasim, H Kirschenmann, and others

Zenodo, doi, 2020

2018

A Big Data Architecture for Log Data Storage and Analysis

Swapneel Mehta, Prasanth Kothuri, and Daniel Lanza Garcia

In Integrated Intelligent Computing, Communication and Security, 2018

Abs PDF

We propose an architecture for analysing database connection logs across different instances of databases within an intranet comprising over 10,000 users and associated devices. Our system uses Flume agents to send notifications to a Hadoop Distributed File System for long-term storage and ElasticSearch and Kibana for short-term visualisation, effectively creating a data lake for the extraction of log data. We adopt machine learning models with an ensemble of approaches to filter and process the indicators within the data and aim to predict anomalies or outliers using feature vectors built from this log data.
Auto-Grading for 3D Modeling Assignments in MOOCs

Swapneel Mehta, Chirag Raman, Nitin Ayer, and Sameer Sahasrabudhe

In 2018 IEEE 18th International Conference on Advanced Learning Technologies (ICALT), 2018

Abs PDF

Bottlenecks such as the latency in correcting assignments and providing a grade for Massive Open Online Courses (MOOCs) could impact the levels of interest among learners. In this proposal for an auto-grading system, we present a method to simplify grading for an online course that focuses on 3D Modeling, thus addressing a critical component of the MOOC ecosystem that affects. Our approach involves a live auto-grader that is capable of attaching descriptive labels to assignments which will be deployed for evaluating submissions. This paper presents a brief overview of this auto-grading system and the reasoning behind its inception. Preliminary internal tests show that our system presents results comparable to human graders.
Anomaly Detection for Network Connection Logs

Swapneel Mehta, Prasanth Kothuri, and Daniel Lanza Garcia

arXiv preprint arXiv:1812.01941, 2018

PDF
DeepJet: A Machine Learning Environment for High-energy Physics

Swapneel Mehta, Mauro Verzetti, Jan Kieseler, and Markus Stoye

2018

PDF