Positive feedback via likes and awards is central to online governance, yet which attributes of users’ posts elicit rewards—and how these vary across authors and communities—remains unclear. To examine this, we combine quasi-experimental causal inference with predictive modeling on 11M posts from 100 subreddits. We identify linguistic patterns and stylistic attributes causally linked to rewards, controlling for author reputation, timing, and community context. For example, overtly complicated language, tentative style, and toxicity reduce rewards. We use our set of curated features to train models that can detect highly-upvoted posts with high AUC. Our audit of community guidelines highlights a “policy-practice gap”—most rules focus primarily on civility and formatting requirements, with little emphasis on the attributes identified to drive positive feedback. These results inform the design of community guidelines, support interfaces that teach users how to craft desirable contributions, and moderation workflows that emphasize positive reinforcement over purely punitive enforcement.

In recent years, various interventions have been developed to reduce individuals’ susceptibility to false and misleading content. Among these, inoculation-based interventions—which preemptively expose people to common manipulation techniques to reduce their susceptibility to falsehoods—have gained popularity in both academic and applied settings, including public education campaigns and platform-level deployments. However, most evaluations have been conducted in controlled settings where each falsehood contains a single manipulation technique, raising concerns about their ecological validity. To inform more robust intervention design, we present two complementary studies that examine the limitations of current inoculation interventions. Study 1 analyzes large-scale COVID-19 and Non-COVID false information datasets to assess the prevalence, co-occurrence, and virality of manipulation techniques, and finds that current interventions rely on oversimplified assumptions about the structure of false information. Study 2 tests whether inoculation against single-technique (emotional language) transfers to multi-technique false information (e.g., emotional language & polarization), and shows its transferability is limited. Taken together, these findings highlight the need to re-evaluate how prevailing inoculation-based interventions are designed and to better adapt them to the complexity of real-world false information environments.

Large language models (LLMs) are now ubiquitous in user-facing applications, yet they still generate undesirable toxic outputs, including profanity, vulgarity, and derogatory remarks.Although numerous detoxification methods exist, most apply broad, surface-level fixes and can therefore easily be circumvented by jailbreak attacks.In this paper we leverage sparse autoencoders (SAEs) to identify toxicityrelated directions in the residual stream of models and perform targeted activation steering using the corresponding decoder vectors.We introduce three tiers of steering aggressiveness and evaluate them on GPT-2 Small and Gemma-2-2B, revealing trade-offs between toxicity reduction and language fluency.At stronger steering strengths, these causal interventions surpass competitive baselines in reducing toxicity by up to 20%, though fluency can degrade noticeably on GPT-2 Small depending on the aggressiveness.Crucially, standard NLP benchmark scores upon steering remain stable, indicating that the model's knowledge and general abilities are preserved.We further show that featuresplitting in wider SAEs hampers safety interventions, underscoring the importance of disentangled feature learning.Our findings highlight both the promise and the current limitations of SAE-based causal interventions for LLM detoxification, further suggesting practical guidelines for safer language-model deployment. 1 ‡ Both authors contributed equally. 1 Code: https://github.com/CrowdDynamicsLab/SAE- Detoxification.Normal

Large language models (LLMs) have shown great potential in flagging harmful content in online communities.Yet, existing approaches for moderation require a separate model for every community and are opaque in their decision-making, limiting real-world adoption.We introduce Mixture of Moderation Experts (MoMoE), a modular, cross-community framework that adds post-hoc explanations to scalable content moderation.MoMoE orchestrates four operators-Allocate , Predict , Aggregate , Explain -and is instantiated as seven community-specialized experts (MoMoE Community ) and five norm-violation experts (MoMoE NormVio ).On 30 unseen subreddits, the best variants obtain Micro-F1 scores of 0.72 and 0.67, respectively, matching or surpassing strong fine-tuned baselines while consistently producing concise and reliable explanations.Although community-specialized experts deliver the highest peak accuracy, norm-violation experts provide steadier performance across domains.These findings show that MoMoE yields scalable, transparent moderation without needing per-community fine-tuning.More broadly, they suggest that lightweight, explainable expert ensembles can guide future NLP and HCI research on trustworthy human-AI governance of online communities.1

Key point extraction is an important task in argument summarization, which involves extracting high-level short summaries from arguments.Existing approaches for KP extraction have been mostly evaluated on the popular ArgKP21 dataset.In this paper, we highlight some of the major limitations of the ArgKP21 dataset and demonstrate the need for new benchmarks that are more representative of actual human conversations.Using SoTA large language models (LLMs), we curate a new argument key point extraction dataset called ArgCMV comprising of ∼ 12K arguments from actual online human debates spread across ∼ 3K topics.Our dataset exhibits higher complexity such as longer, coreferencing arguments, higher presence of subjective discourse units, and a larger range of topics over ArgKP21.We show that existing methods do not adapt well to ArgCMV and provide extensive benchmark results by experimenting with existing baselines and latest open source models.This work introduces a novel KP extraction dataset for long-context online discussions, setting the stage for the next generation of LLM-driven summarization research.1

Xianyang Zhan, Agam Goyal, Yilun Chen, Eshwar Chandrasekharan, Koustuv Saha. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.