arXiv subjects

Erfan Samieyan Sahneh

Publications and source records attributed to Erfan Samieyan Sahneh.

4 recordsLinked to original sources

arXiv2026

Auditing the Reliability of Multimodal Generative Search

Multimodal Large Language Models (MLLMs) increasingly function as generative search systems that retrieve and synthesize answers from multimedia content, including YouTube videos. Although these systems project authority by citing specific videos as evidence, the extent to which these citations genuinely substantiate the generated claims remains underexplored. We present a large-scale audit of the Gemini 2.5 Pro multimodal search system, analyzing 11,943 claim-video pairs generated across Medical, Economic, and General domains. Through automated verification using three independent LLM judges (87.7\% inter-rater agreement), validated against human annotations, we find that depending on the judge's strictness, between 3.7\% and 18.7\% of video-grounded claims are not supported by their cited sources. The dominant failure modes are not outright contradictions but rather unverifiable specificities and overstated claims, suggesting the system injects precise but ungrounded details from parametric knowledge while citing videos as evidence. Exploratory post-hoc analysis via logistic regression reveals properties associated with these failures: claims departing from source vocabulary ($β= -1.6$ to $-3.1$, $p < 0.01$) and claims with low semantic similarity to the video transcript ($β= -2.1$ to $-11.6$, $p < 0.01$) are significantly more likely to be unsupported. These findings characterize the current trustworthiness of video-based generative search and highlight the gap between the confidence these systems project and the fidelity of their outputs. The dataset is available at https://anonymous.4open.science/r/icwsm-gemini-audit-04DF .

cs.CY↗

arXiv2026

Audit of takedown delays across social media reveals failure to reduce exposure to illegal content

Illegal content on social media poses significant societal harm and necessitates timely removal. However, the impact of the speed of content removal on prevalence, reach, and exposure to illegal content remains underexplored. This study examines the relationship with a systematic audit of takedown delays using data from the EU Digital Services Act Transparency Database, covering five major platforms over a one-year period. We find substantial variation in takedown delay, with some content remaining online for weeks or even months. To evaluate how these delays affect the prevalence and reach of illegal content and exposure to it, we develop an agent-based model and calibrate it to empirical data. We simulate illegal content diffusion, revealing that rapid takedown (within hours) significantly reduces prevalence, reach, and exposure to illegal content, while the longer delays measured by the audit fail to reduce its spread. Though the link between delay and spread is intuitive, our simulations quantify exactly how takedown speed shapes exposure to illegal content. Building on these results, we point to the benefits of faster content removal to effectively curb the spread of illegal content, while also considering the limitations of strict enforcement policies.

cs.SI↗

arXiv2025

A longitudinal analysis of misinformation, polarization and toxicity on Bluesky after its public launch

Bluesky is a decentralized, Twitter-like social media platform that has rapidly gained popularity. Following an invite-only phase, it officially opened to the public on February 6th, 2024, leading to a significant expansion of its user base. In this paper, we present a longitudinal analysis of user activity in the two months surrounding its public launch, examining how the platform evolved due to this rapid growth. Our analysis reveals that Bluesky exhibits an activity distribution comparable to more established social platforms, yet it features a higher volume of original content relative to reshared posts and maintains low toxicity levels. We further investigate the political leanings of its user base, misinformation dynamics, and engagement in harmful conversations. Our findings indicate that Bluesky users predominantly lean left politically and tend to share high-credibility sources. After the platform's public launch, an influx of new users, particularly those posting in English and Japanese, contributed to a surge in activity. Among them, several accounts displayed suspicious behaviors, such as mass-following users and sharing content from low-credibility news sources. Some of these accounts have already been flagged as spam or suspended, suggesting that Bluesky's moderation efforts have been effective.

cs.SI↗

arXiv2024

The Dawn of Decentralized Social Media: An Exploration of Bluesky's Public Opening

Bluesky is a Twitter-like decentralized social media platform that has recently grown in popularity. After an invite-only period, it opened to the public worldwide on February 6th, 2024. In this paper, we provide a longitudinal analysis of user activity in the two months around the opening, studying changes in the general characteristics of the platform due to the rapid growth of the user base. We observe a broad distribution of activity similar to more established platforms, but a higher volume of original than reshared content, and very low toxicity. After opening to the public, Bluesky experienced a large surge in new users and activity, especially posting English and Japanese content. In particular, several accounts entered the discussion with suspicious behavior, like following many accounts and sharing content from low-credibility news outlets. Some of these have already been classified as spam or suspended, suggesting effective moderation.

cs.SI↗