Mattia Samory

Indira Sen, Dennis Assenmacher, Mattia Samory et al.

NLP models are used in a variety of critical social computing tasks, such as detecting sexist, racist, or otherwise hateful content. Therefore, it is imperative that these models are robust to spurious features. Past work has attempted to tackle such spurious features using training data augmentation, including Counterfactually Augmented Data (CADs). CADs introduce minimal changes to existing training data points and flip their labels; training on them may reduce model dependency on spurious features. However, manually generating CADs can be time-consuming and expensive. Hence in this work, we assess if this task can be automated using generative NLP models. We automatically generate CADs using Polyjuice, ChatGPT, and Flan-T5, and evaluate their usefulness in improving model robustness compared to manually-generated CADs. By testing both model performance on multiple out-of-domain test sets and individual data point efficacy, our results show that while manual CADs are still the most effective, CADs generated by ChatGPT come a close second. One key reason for the lower performance of automated methods is that the changes they introduce are often insufficient to flip the original label.

Indira Sen, Mattia Samory, Claudia Wagner et al.

Counterfactually Augmented Data (CAD) aims to improve out-of-domain generalizability, an indicator of model robustness. The improvement is credited with promoting core features of the construct over spurious artifacts that happen to correlate with it. Yet, over-relying on core features may lead to unintended model bias. Especially, construct-driven CAD -- perturbations of core features -- may induce models to ignore the context in which core features are used. Here, we test models for sexism and hate speech detection on challenging data: non-hateful and non-sexist usage of identity and gendered terms. In these hard cases, models trained on CAD, especially construct-driven CAD, show higher false-positive rates than models trained on the original, unperturbed data. Using a diverse set of CAD -- construct-driven and construct-agnostic -- reduces such unintended bias.

Franco Della Negra, Mattia Samory, Matteo Cinelli

Online communities are a global phenomenon, but assessing their actual geographical spread requires accurate and scalable measurement. We propose and evaluate methods that infer the time zone of online communities solely from their temporal activity patterns, requiring nothing beyond hourly activity counts. Grounding our approach in the well-established finding that posting rhythms encode circadian structure, we compare time-domain and frequency-domain methods against a parsimonious heuristic: that activity reaches its minimum around 4 a.m. local time. On Reddit, we show that the best-performing method is accurate to a sub-30-minute resolution, and that fewer than a thousand comments are sufficient to reach peak performance. Similarly, our heuristic almost matches the accuracy of more complex methods, recovering the correct time zone within a one-hour margin on average. This simple method correlates significantly with the actual distribution of Reddit's geographical spread; we validate its generalizability across communities organized around diverse cultural phenomena, from sports to finance, and apply it at scale to characterize the geographic evolution of Reddit from its founding to the present. Our method is portable across platforms and requires no user disclosure, making it a practical baseline for any study that must account for the geographic structure of online behavior.

Shruti Phadke, Mattia Samory, Tanushree Mitra

The disruptive offline mobilization of participants in online conspiracy theory (CT) discussions has highlighted the importance of understanding how online users may form radicalized conspiracy beliefs. While prior work researched the factors leading up to joining online CT discussions and provided theories of how conspiracy beliefs form, we have little understanding of how conspiracy radicalization evolves after users join CT discussion communities. In this paper, we provide the empirical modeling of various radicalization phases in online CT discussion participants. To unpack how conspiracy engagement is related to radicalization, we first characterize the users' journey through CT discussions via conspiracy engagement pathways. Specifically, by studying 36K Reddit users through their 169M contributions, we uncover four distinct pathways of conspiracy engagement: steady high, increasing, decreasing, and steady low. We further model three successive stages of radicalization guided by prior theoretical works. Specific sub-populations of users, namely those on steady high and increasing conspiracy engagement pathways, progress successively through various radicalization stages. In contrast, users on the decreasing engagement pathway show distinct behavior: they limit their CT discussions to specialized topics, participate in diverse discussion groups, and show reduced conformity with conspiracy subreddits. By examining users who disengage from online CT discussions, this paper provides promising insights about conspiracy recovery process.

Zehui Yu, Indira Sen, Dennis Assenmacher et al.

Machine learning (ML)-based content moderation tools are essential to keep online spaces free from hateful communication. Yet, ML tools can only be as capable as the quality of the data they are trained on allows them. While there is increasing evidence that they underperform in detecting hateful communications directed towards specific identities and may discriminate against them, we know surprisingly little about the provenance of such bias. To fill this gap, we present a systematic review of the datasets for the automated detection of hateful communication introduced over the past decade, and unpack the quality of the datasets in terms of the identities that they embody: those of the targets of hateful communication that the data curators focused on, as well as those unintentionally included in the datasets. We find, overall, a skewed representation of selected target identities and mismatches between the targets that research conceptualizes and ultimately includes in datasets. Yet, by contextualizing these findings in the language and location of origin of the datasets, we highlight a positive trend towards the broadening and diversification of this research space.

Shayan Alipour, Indira Sen, Mattia Samory et al.

Large language models (LLMs) are known to exhibit demographic biases, yet few studies systematically evaluate these biases across multiple datasets or account for confounding factors. In this work, we examine LLM alignment with human annotations in five offensive language datasets, comprising approximately 220K annotations. Our findings reveal that while demographic traits, particularly race, influence alignment, these effects are inconsistent across datasets and often entangled with other factors. Confounders -- such as document difficulty, annotator sensitivity, and within-group agreement -- account for more variation in alignment patterns than demographic traits alone. Specifically, alignment increases with higher annotator sensitivity and group agreement, while greater document difficulty corresponds to reduced alignment. Our results underscore the importance of multi-dataset analyses and confounder-aware methodologies in developing robust measures of demographic bias in LLMs.

Jacopo Nudo, Mario Edoardo Pandolfo, Edoardo Loru et al.

We investigate how Large Language Models (LLMs) behave when simulating political discourse on social media. Leveraging 21 million interactions on X during the 2024 U.S. presidential election, we construct LLM agents based on 1,186 real users, prompting them to reply to politically salient tweets under controlled conditions. Agents are initialized either with minimal ideological cues (Zero Shot) or recent tweet history (Few Shot), allowing one-to-one comparisons with human replies. We evaluate three model families (Gemini, Mistral, and DeepSeek) across linguistic style, ideological consistency, and toxicity. We find that richer contextualization improves internal consistency but also amplifies polarization, stylized signals, and harmful language. We observe an emergent distortion that we call "generation exaggeration": a systematic amplification of salient traits beyond empirical baselines. Our analysis shows that LLMs do not emulate users, they reconstruct them. Their outputs, indeed, reflect internal optimization dynamics more than observed behavior, introducing structural biases that compromise their reliability as social proxies. This challenges their use in content moderation, deliberative simulations, and policy modeling.

Xi Chen, Mattia Samory, Scott Hale et al.

Understanding the writing frame of news articles is vital for addressing social issues, and thus has attracted notable attention in the fields of communication studies. Yet, assessing such news article frames remains a challenge due to the absence of a concrete and unified standard dataset that considers the comprehensive nuances within news content. To address this gap, we introduce an extended version of a large labeled news article dataset with 16,687 new labeled pairs. Leveraging the pairwise comparison of news articles, our method frees the work of manual identification of frame classes in traditional news frame analysis studies. Overall we introduce the most extensive cross-lingual news article similarity dataset available to date with 26,555 labeled news article pairs across 10 languages. Each data point has been meticulously annotated according to a codebook detailing eight critical aspects of news content, under a human-in-the-loop framework. Application examples demonstrate its potential in unearthing country communities within global news coverage, exposing media bias among news outlets, and quantifying the factors related to news creation. We envision that this news similarity dataset will broaden our understanding of the media ecosystem in terms of news coverage of events and perspectives across countries, locations, languages, and other social constructs. By doing so, it can catalyze advancements in social science research and applied methodologies, thereby exerting a profound impact on our society.

Shayan Alipour, Shruti Phadke, Seyed Shahabeddin Mousavi et al.

Volunteer moderators play a crucial role in sustaining online dialogue, but they often disagree about what should or should not be allowed. In this paper, we study the complexity of content moderation with a focus on disagreements between moderators, which we term the ``gray area'' of moderation. Leveraging 5 years and 4.3 million moderation log entries from 24 subreddits of different topics and sizes, we characterize how gray area, or disputed cases, differ from undisputed cases. We show that one-in-seven moderation cases are disputed among moderators, often addressing transgressions where users' intent is not directly legible, such as in trolling and brigading, as well as tensions around community governance. This is concerning, as almost half of all gray area cases involved automated moderation decisions. Through information-theoretic evaluations, we demonstrate that gray area cases are inherently harder to adjudicate than undisputed cases and show that state-of-the-art language models struggle to adjudicate them. We highlight the key role of expert human moderators in overseeing the moderation process and provide insights about the challenges of current moderation processes and tools.

Mattia Samory, Diana Pamfile, Andrew To et al.

Online communities rely on a mix of platform policies and community-authored rules to define acceptable behavior and maintain order. However, these rules vary widely across communities, evolve over time, and are enforced inconsistently, posing challenges for transparency, governance, and automation. In this paper, we model the relationship between rules and their enforcement at scale, introducing ModQ, a novel question-answering framework for rule-sensitive content moderation. Unlike prior classification or generation-based approaches, ModQ conditions on the full set of community rules at inference time and identifies which rule best applies to a given comment. We implement two model variants - extractive and multiple-choice QA - and train them on large-scale datasets from Reddit and Lemmy, the latter of which we construct from publicly available moderation logs and rule descriptions. Both models outperform state-of-the-art baselines in identifying moderation-relevant rule violations, while remaining lightweight and interpretable. Notably, ModQ models generalize effectively to unseen communities and rules, supporting low-resource moderation settings and dynamic governance environments.

Shruti Phadke, Mattia Samory, Tanushree Mitra

Online discussion platforms offer a forum to strengthen and propagate belief in misinformed conspiracy theories. Yet, they also offer avenues for conspiracy theorists to express their doubts and experiences of cognitive dissonance. Such expressions of dissonance may shed light on who abandons misguided beliefs and under which circumstances. This paper characterizes self-disclosures of dissonance about QAnon, a conspiracy theory initiated by a mysterious leader Q and popularized by their followers, anons in conspiracy theory subreddits. To understand what dissonance and disbelief mean within conspiracy communities, we first characterize their social imaginaries, a broad understanding of how people collectively imagine their social existence. Focusing on 2K posts from two image boards, 4chan and 8chan, and 1.2 M comments and posts from 12 subreddits dedicated to QAnon, we adopt a mixed methods approach to uncover the symbolic language representing the movement, expectations, practices, heroes and foes of the QAnon community. We use these social imaginaries to create a computational framework for distinguishing belief and dissonance from general discussion about QAnon. Further, analyzing user engagement with QAnon conspiracy subreddits, we find that self-disclosures of dissonance correlate with a significant decrease in user contributions and ultimately with their departure from the community. We contribute a computational framework for identifying dissonance self-disclosures and measuring the changes in user engagement surrounding dissonance. Our work can provide insights into designing dissonance-based interventions that can potentially dissuade conspiracists from online conspiracy discussion communities.

Mattia Samory, Indira Sen, Julian Kohne et al.

Research has focused on automated methods to effectively detect sexism online. Although overt sexism seems easy to spot, its subtle forms and manifold expressions are not. In this paper, we outline the different dimensions of sexism by grounding them in their implementation in psychological scales. From the scales, we derive a codebook for sexism in social media, which we use to annotate existing and novel datasets, surfacing their limitations in breadth and validity with respect to the construct of sexism. Next, we leverage the annotated datasets to generate adversarial examples, and test the reliability of sexism detection methods. Results indicate that current machine learning models pick up on a very narrow set of linguistic markers of sexism and do not generalize well to out-of-domain examples. Yet, including diverse data and adversarial examples at training time results in models that generalize better and that are more robust to artifacts of data collection. By providing a scale-based codebook and insights regarding the shortcomings of the state-of-the-art, we hope to contribute to the development of better and broader models for sexism detection, including reflections on theory-driven approaches to data collection.