Kiran Garimella

Punyajoy Saha, Kiran Garimella, Narla Komal Kalyan et al.

Recently, social media platforms are heavily moderated to prevent the spread of online hate speech, which is usually fertile in toxic words and is directed toward an individual or a community. Owing to such heavy moderation, newer and more subtle techniques are being deployed. One of the most striking among these is fear speech. Fear speech, as the name suggests, attempts to incite fear about a target community. Although subtle, it might be highly effective, often pushing communities toward a physical conflict. Therefore, understanding their prevalence in social media is of paramount importance. This article presents a large-scale study to understand the prevalence of 400K fear speech and over 700K hate speech posts collected from Gab.com. Remarkably, users posting a large number of fear speech accrue more followers and occupy more central positions in social networks than users posting a large number of hate speech. They can also reach out to benign users more effectively than hate speech users through replies, reposts, and mentions. This connects to the fact that, unlike hate speech, fear speech has almost zero toxic content, making it look plausible. Moreover, while fear speech topics mostly portray a community as a perpetrator using a (fake) chain of argumentation, hate speech topics hurl direct multitarget insults, thus pointing to why general users could be more gullible to fear speech. Our findings transcend even to other platforms (Twitter and Facebook) and thus necessitate using sophisticated moderation policies and mass awareness to combat fear speech.

S M Mehedi Zaman, Kiran Garimella

Hundreds of millions of users now hold detailed, multi-turn conversations with ChatGPT and similar LLM assistants. We measure two privacy-relevant features of these conversations on a corpus of complete ChatGPT histories donated by over 1,000 users in four Global South countries (Brazil, India, Nigeria, Pakistan). First, on explicit disclosure: 34.5% of user messages contain personal information across a twenty-category taxonomy, with the median user first revealing identifying content within the first 14% of their conversation history. Second, on inference beyond explicit disclosure: we restrict to a cohort whose conversations contain no messages flagged by an LLM-based filter for explicit demographic self-identification (a separate NER pass marks PII for the disclosure audit but does not drive cohort exclusion). On this filtered cohort, an off the shelf large language model still recovers each user's age, gender, and country at weighted F1 of 0.84, 0.90, and 0.88, respectively, with the median user identified from the first 5% of their conversation history. Reading the model's natural-language reasoning traces, we identify four recurring stereotype patterns that drive both successful inference and an asymmetric error distribution concentrating on women in technical fields, older users with contemporary skills, and Global South tech professionals. We also compare ChatGPT against the same users' Google Search and YouTube histories as inference surfaces, and find it competitive with these older substrates that have driven behavioral advertising for two decades. Message-level PII removal is insufficient on its own as a privacy intervention for conversational AI data.

Shreya Gupta, Ojasva Saxena, Arghodeep Nandi et al.

Podcasts have become a central arena for shaping public opinion, making them a vital source for understanding contemporary discourse. Their typically unscripted, multi-themed, and conversational style offers a rich but complex form of data. To analyze how podcasts persuade and inform, we must examine their narrative structures -- specifically, the narrative frames they employ. The fluid and conversational nature of podcasts presents a significant challenge for automated analysis. We show that existing large language models, typically trained on more structured text such as news articles, struggle to capture the subtle cues that human listeners rely on to identify narrative frames. As a result, current approaches fall short of accurately analyzing podcast narratives at scale. To solve this, we develop and evaluate a fine-tuned BERT model that explicitly links narrative frames to specific entities mentioned in the conversation, effectively grounding the abstract frame in concrete details. Our approach then uses these granular frame labels and correlates them with high-level topics to reveal broader discourse trends. The primary contributions of this paper are: (i) a novel frame-labeling methodology that more closely aligns with human judgment for messy, conversational data, and (ii) a new analysis that uncovers the systematic relationship between what is being discussed (the topic) and how it is being presented (the frame), offering a more robust framework for studying influence in digital media.

Gauri Nayak, Farhana Shahid, Aditya Vashistha et al.

WhatsApp is one of the most widely used messaging platforms globally, with billions of users sharing information in private groups. Yet, it offers little infrastructure to support moderation and group governance. In the absence of platform-level oversight, group admins bear the responsibility of governing group behavior. In this paper, we explore how WhatsApp group admins collaborate with AI tools to create, enforce, and maintain group rules. Drawing on a two-phase speculative design study with 20 admins in India, we examine how participants interacted with an AI assistant (Meta AI) to co-create rules and responded to a series of probes illustrating AI-assisted moderation features. Our findings show that while admins appreciated the AI's ability to surface overlooked rules and reduce their moderation burden, they were highly sensitive to issues of relational trust, data privacy, tone, and social context. We identify how group type and admin style shaped their willingness to delegate authority, and surface the limitations of current chatbot interfaces in supporting collaborative rule-making. We conclude with design implications for building moderation tools that center human judgment, relational nuance, contextual adaptability, and collective governance.

S M Mehedi Zaman, Joao Couto, Kiran Garimella

Despite Facebook's central role in American civic life, a clear, evidence-based understanding of users' long-term information environments has remained elusive, hindering assessments of the platform's societal impact. This study addresses that gap by analyzing a unique decade-long dataset, constructed by collecting the full list of public pages and groups followed by over 1,100 American users. This approach allows us to examine the potential information exposure of these users by analyzing hundreds of millions of posts from 2012 to 2023. We find that political content constitutes a modest 18% of a user's potential information diet, which is predominantly composed of lifestyle and entertainment topics. This aggregate view, however, masks a deeply stratified reality: we uncover significant and persistent disparities in the volume and ideological leaning of political content across age, gender, and racial lines. Furthermore, we quantify the porous boundaries between content categories, showing how political discourse frequently permeates non-political spaces. Leveraging the dataset's longitudinal nature, we also assess the impact of major platform interventions. We find that Meta's 2018 "Meaningful Social Interactions" update dramatically increased the share of political content by contracting the visibility of non-political posts. By providing a granular, decade-long map of potential information exposure, our study offers one of the first representative and longitudinal picture drawn from platform-independent data. Our findings underscore the critical need for researchers to measure exposure, not merely engagement, and to account for the significant volume of political content that circulates in non-political spaces.

Peixian Zhang, Qiming Ye, Zifan Peng et al.

LLM-based Search Engines (LLM-SEs) introduces a new paradigm for information seeking. Unlike Traditional Search Engines (TSEs) (e.g., Google), these systems summarize results, often providing limited citation transparency. The implications of this shift remain largely unexplored, yet raises key questions regarding trust and transparency. In this paper, we present a large-scale empirical study of LLM-SEs, analyzing 55,936 queries and the corresponding search results across six LLM-SEs and two TSEs. We confirm that LLM-SEs cites domain resources with greater diversity than TSEs. Indeed, 37% of domains are unique to LLM-SEs. However, certain risks still persist: LLM-SEs do not outperform TSEs in credibility, political neutrality and safety metrics. Finally, to understand the selection criteria of LLM-SEs, we perform a feature-based analysis to identify key factors influencing source choice. Our findings provide actionable insights for end users, website owners, and developers.

Naquee Rizwan, Nayandeep Deb, Sarthak Roy et al.

Tackling toxic behavior in digital communication continues to be a pressing concern for both academics and industry professionals. While significant research has explored toxicity on platforms like social networks and discussion boards, podcasts despite their rapid rise in popularity remain relatively understudied in this context. This work seeks to fill that gap by curating a dataset of political podcast transcripts and analyzing them with a focus on conversational structure. Specifically, we investigate how toxicity surfaces and intensifies through sequences of replies within these dialogues, shedding light on the organic patterns by which harmful language can escalate across conversational turns. Warning: Contains potentially abusive/toxic contents.

Ashwin Singh, Mallika Subramanian, Anmol Agarwal et al.

In the last two decades, ICTs have played a pivotal role in empowering rural populations in India by making knowledge more accessible. Digital Green (DG) is one such ICT that employs a participatory approach with smallholder farmers to produce instructional videos that encompass content specific to them. With help of human mediators, they disseminate these videos using projectors to improve the adoption of agricultural practices. DG's web-based data tracker stores attendance and adoption logs of millions of farmers, videos screened and their demographic information. We leverage this data for a period of ten years between 2010-2020 across five states in India and use it to conduct a holistic evaluation of the ICT. First, we find disparities in adoption rates of farmers, following which we use statistical tests to identify different factors that lead to these disparities and gender-based inequalities. Second, to provide assistance to farmers facing challenges, we model the adoption of practices from a video as a prediction problem and experiment with different model architectures. Our classifier achieves accuracies ranging from 79% to 90% across the five states, demonstrating its potential for assisting future ethnographic investigations. Third, we use SHAP values in conjunction with our model for explaining the impact of various network, content and demographic features on adoption. Our research finds that farmers greatly benefit from past adopters of a video from their group and village. We also discover that videos with a low content-specificity benefit some farmers more than others. Next, we highlight the implications of our findings by translating them into recommendations for community building, revisiting participatory approach and mitigating inequalities. We conclude with a discussion on how our work can assist future investigations into the lived experiences of farmers.

Ashkan Kazemi, Kiran Garimella, Gautam Kishore Shahi et al.

There is currently no easy way to fact-check content on WhatsApp and other end-to-end encrypted platforms at scale. In this paper, we analyze the usefulness of a crowd-sourced "tipline" through which users can submit content ("tips") that they want fact-checked. We compare the tips sent to a WhatsApp tipline run during the 2019 Indian national elections with the messages circulating in large, public groups on WhatsApp and other social media platforms during the same period. We find that tiplines are a very useful lens into WhatsApp conversations: a significant fraction of messages and images sent to the tipline match with the content being shared on public WhatsApp groups and other social media. Our analysis also shows that tiplines cover the most popular content well, and a majority of such content is often shared to the tipline before appearing in large, public WhatsApp groups. Overall, our findings suggest tiplines can be an effective source for discovering content to fact-check.

Pushkal Agarwal, Aravindh Raman, Damilola Ibosiola et al.

WhatsApp is a popular messaging app used by over a billion users around the globe. Due to this popularity, understanding misbehavior on WhatsApp is an important issue. The sending of unwanted junk messages by unknown contacts via WhatsApp remains understudied by researchers, in part because of the end-to-end encryption offered by the platform. We address this gap by studying junk messaging on a multilingual dataset of 2.6M messages sent to 5K public WhatsApp groups in India. We characterise both junk content and senders. We find that nearly 1 in 10 messages is unwanted content sent by junk senders, and a number of unique strategies are employed to reflect challenges faced on WhatsApp, e.g., the need to change phone numbers regularly. We finally experiment with on-device classification to automate the detection of junk, whilst respecting end-to-end encryption.

Ashkan Kazemi, Kiran Garimella, Devin Gaffney et al.

Manual fact-checking does not scale well to serve the needs of the internet. This issue is further compounded in non-English contexts. In this paper, we discuss claim matching as a possible solution to scale fact-checking. We define claim matching as the task of identifying pairs of textual messages containing claims that can be served with one fact-check. We construct a novel dataset of WhatsApp tipline and public group messages alongside fact-checked claims that are first annotated for containing "claim-like statements" and then matched with potentially similar items and annotated for claim matching. Our dataset contains content in high-resource (English, Hindi) and lower-resource (Bengali, Malayalam, Tamil) languages. We train our own embedding model using knowledge distillation and a high-quality "teacher" model in order to address the imbalance in embedding quality between the low- and high-resource languages in our dataset. We provide evaluations on the performance of our solution and compare with baselines and existing state-of-the-art multilingual embedding models, namely LASER and LaBSE. We demonstrate that our performance exceeds LASER and LaBSE in all settings. We release our annotated datasets, codebooks, and trained embedding model to allow for further research.

Maurice Jakesch, Kiran Garimella, Dean Eckles et al.

Political organizations worldwide keep innovating their use of social media technologies. In the 2019 Indian general election, organizers used a network of WhatsApp groups to manipulate Twitter trends through coordinated mass postings. We joined 600 WhatsApp groups that support the Bharatiya Janata Party, the right-wing party that won the general election, to investigate these campaigns. We found evidence of 75 hashtag manipulation campaigns in the form of mobilization messages with lists of pre-written tweets. Building on this evidence, we estimate the campaigns' size, describe their organization and determine whether they succeeded in creating controlled social media narratives. Our findings show that the campaigns produced hundreds of nationwide Twitter trends throughout the election. Centrally controlled but voluntary in participation, this hybrid configuration of technologies and organizational strategies shows how profoundly online tools transform campaign politics. Trend alerts complicate the debates over the legitimate use of digital tools for political participation and may have provided a blueprint for participatory media manipulation by a party with popular support.

Punyajoy Saha, Binny Mathew, Kiran Garimella et al.

WhatsApp is the most popular messaging app in the world. Due to its popularity, WhatsApp has become a powerful and cheap tool for political campaigning being widely used during the 2019 Indian general election, where it was used to connect to the voters on a large scale. Along with the campaigning, there have been reports that WhatsApp has also become a breeding ground for harmful speech against various protected groups and religious minorities. Many such messages attempt to instil fear among the population about a specific (minority) community. According to research on inter-group conflict, such `fear speech' messages could have a lasting impact and might lead to real offline violence. In this paper, we perform the first large scale study on fear speech across thousands of public WhatsApp groups discussing politics in India. We curate a new dataset and try to characterize fear speech from this dataset. We observe that users writing fear speech messages use various events and symbols to create the illusion of fear among the reader about a target community. We build models to classify fear speech and observe that current state-of-the-art NLP models do not perform well at this task. Fear speech messages tend to spread faster and could potentially go undetected by classifiers built to detect traditional toxic speech due to their low toxic nature. Finally, using a novel methodology to target users with Facebook ads, we conduct a survey among the users of these WhatsApp groups to understand the types of users who consume and share fear speech. We believe that this work opens up new research questions that are very different from tackling hate speech which the research community has been traditionally involved in.

Kirill Martynov, Kiran Garimella, Robert West

Body measurements, including weight and height, are key indicators of health. Being able to visually assess body measurements reliably is a step towards increased awareness of overweight and obesity and is thus important for public health. Nevertheless it is currently not well understood how accurately humans can assess weight and height from images, and when and how they fail. To bridge this gap, we start from 1,682 images of persons collected from the Web, each annotated with the true weight and height, and ask crowd workers to estimate the weight and height for each image. We conduct a faceted analysis taking into account characteristics of the images as well as the crowd workers assessing the images, revealing several novel findings: (1) Even after aggregation, the crowd's accuracy is overall low. (2) We find strong evidence of contraction bias toward a reference value, such that the weight (height) of light (short) people is overestimated, whereas that of heavy (tall) people is underestimated. (3) We estimate workers' individual reference values using a Bayesian model, finding that reference values strongly correlate with workers' own height and weight, indicating that workers are better at estimating people similar to themselves. (4) The weight of tall people is underestimated more than that of short people; yet, knowing the height decreases the weight error only mildly. (5) Accuracy is higher on images of females than of males, but female and male workers are no different in terms of accuracy. (6) Crowd workers improve over time if given feedback on previous guesses. Finally, we explore various bias correction models for improving the crowd's accuracy, but find that this only leads to modest gains. Overall, this work provides important insights on biases in body measurement estimation as obesity related conditions are on the rise.

Julio C. S. Reis, Philipe de Freitas Melo, Kiran Garimella et al.

WhatsApp was alleged to be widely used to spread misinformation and propaganda during elections in Brazil and India. Due to the private encrypted nature of the messages on WhatsApp, it is hard to track the dissemination of misinformation at scale. In this work, using public WhatsApp data, we observe that misinformation has been largely shared on WhatsApp public groups even after they were already fact-checked by popular fact-checking agencies. This represents a significant portion of misinformation spread in both Brazil and India in the groups analyzed. We posit that such misinformation content could be prevented if WhatsApp had a means to flag already fact-checked content. To this end, we propose an architecture that could be implemented by WhatsApp to counter such misinformation. Our proposal respects the current end-to-end encryption architecture on WhatsApp, thus protecting users' privacy while providing an approach to detect the misinformation that benefits from fact-checking efforts.

Pushkal Agarwal, Kiran Garimella, Sagar Joglekar et al.

Social media has been on the vanguard of political information diffusion in the 21st century. Most studies that look into disinformation, political influence and fake-news focus on mainstream social media platforms. This has inevitably made English an important factor in our current understanding of political activity on social media. As a result, there has only been a limited number of studies into a large portion of the world, including the largest, multilingual and multi-cultural democracy: India. In this paper we present our characterisation of a multilingual social network in India called ShareChat. We collect an exhaustive dataset across 72 weeks before and during the Indian general elections of 2019, across 14 languages. We investigate the cross lingual dynamics by clustering visually similar images together, and exploring how they move across language barriers. We find that Telugu, Malayalam, Tamil and Kannada languages tend to be dominant in soliciting political images (often referred to as memes), and posts from Hindi have the largest cross-lingual diffusion across ShareChat (as well as images containing text in English). In the case of images containing text that cross language barriers, we see that language translation is used to widen the accessibility. That said, we find cases where the same image is associated with very different text (and therefore meanings). This initial characterisation paves the way for more advanced pipelines to understand the dynamics of fake and political content in a multi-lingual and non-textual setting.