arXiv subjects

Charlotte Lambert

Publications and source records attributed to Charlotte Lambert.

8 recordsLinked to original sources

arXiv2026

VASTU: Language Models Struggle to Recognize Online Community Values

Online communities develop distinct norms for content they collectively value, yet it remains unclear whether current language models can recognize locally valued contributions in context. We formalize this as \textbf{community-conditioned preference prediction} and introduce \textsc{Vastu} (\underline{V}alue-\underline{A}ware \underline{S}ocial \underline{Tu}ning), a benchmark of 75,000 Reddit comments from 15 communities spanning Gaming, Science, Q\&A, Advice, and Politics. We evaluate four model families---prompted LLMs, LoRA-adapted SLMs, supervised encoders, and feature-based classifiers---across global, local, and context-conditioned settings. Our central finding is that parametric adaptation consistently outperforms prompting: supervised encoders reach 0.74 AUROC and fine-tuned SLMs 0.64--0.71, while the best prompted result is only 0.62. This gap is not merely quantitative---vanilla prompting yields over 80\% false-negative rates, systematically discarding content communities actually value. Conversational context narrows but does not close this divide. Together, these results suggest that local preference recognition requires community-specific training signal, not just better prompting. Our work supports future research on community-aware reward modeling, feed curation, and positive moderation.

cs.HC↗

arXiv2025

Mind Your Ps and Qs: Supporting Positive Reinforcement in Moderation Through a Positive Queue

Online communities are constantly growing, with dozens of platforms housing millions of users. Large and small communities alike rely on volunteer moderators to maintain order. Despite their key role, moderators are given a toolbox of punishments and asked to fend off barrages of harmful content. However, prior research shows that positive feedback may proactively encourage higher quality contributions and discourage norm violations. Moreover, moderators themselves have requested support for locating and rewarding content to encourage in their communities. These requests notwithstanding, there is a tangible lack of practical support through tools. Building off moderators' ideas, we build a novel moderation system, the Positive Queue, that augments Reddit's existing moderator interface with features to discover and reward desirable content. Through a user study of moderators, we find that the system has value to vastly different moderation settings. We present design directions and insights for incorporating positive moderation strategies into existing spaces.

cs.HC↗

arXiv2025

The Language of Approval: Identifying the Drivers of Positive Feedback Online

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.

cs.HC↗

arXiv2025

BAP v2: An Enhanced Task Framework for Instruction Following in Minecraft Dialogues

Developing interactive agents that can understand language, perceive their surroundings, and act within the physical world is a long-standing goal of AI research. The Minecraft Collaborative Building Task (MCBT) (Narayan-Chen, Jayannavar, and Hockenmaier 2019), a two-player game in which an Architect (A) instructs a Builder (B) to construct a target structure in a simulated 3D Blocks World environment, offers a rich platform to work towards this goal. In this work, we focus on the Builder Action Prediction (BAP) subtask: predicting B's actions in a multimodal game context (Jayannavar, Narayan-Chen, and Hockenmaier 2020) - a challenging testbed for grounded instruction following, with limited training data. We holistically re-examine this task and introduce BAP v2 to address key challenges in evaluation, training data, and modeling. Specifically, we define an enhanced evaluation benchmark, featuring a cleaner test set and fairer, more insightful metrics that also reveal spatial reasoning as the primary performance bottleneck. To address data scarcity and to teach models basic spatial skills, we generate different types of synthetic MCBT data. We observe that current, LLM-based SOTA models trained on the human BAP dialogues fail on these simpler, synthetic BAP ones, but show that training models on this synthetic data improves their performance across the board. We also introduce a new SOTA model, Llama-CRAFTS, which leverages richer input representations, and achieves an F1 score of 53.0 on the BAP v2 task and strong performance on the synthetic data. While this result marks a notable 6 points improvement over previous work, it also underscores the task's remaining difficulty, establishing BAP v2 as a fertile ground for future research, and providing a useful measure of the spatial capabilities of current text-only LLMs in such embodied tasks.

cs.CL↗

arXiv2024

Uncovering the Internet's Hidden Values: An Empirical Study of Desirable Behavior Using Highly-Upvoted Content on Reddit

A major task for moderators of online spaces is norm-setting, essentially creating shared norms for user behavior in their communities. Platform design principles emphasize the importance of highlighting norm-adhering examples and explicitly stating community norms. However, norms and values vary between communities and go beyond content-level attributes, making it challenging for platforms and researchers to provide automated ways to identify desirable behavior to be highlighted. Current automated approaches to detect desirability are limited to measures of prosocial behavior, but we do not know whether these measures fully capture the spectrum of what communities value. In this paper, we use upvotes, which express community approval, as a proxy for desirability and examine 16,000 highly-upvoted comments across 80 popular sub-communities on Reddit. Using a large language model, we extract values from these comments across two years (2016 and 2022) and compile 64 and 72 $\textit{macro}$, $\textit{meso}$, and $\textit{micro}$ values for 2016 and 2022 respectively, based on their frequency across communities. Furthermore, we find that existing computational models for measuring prosociality were inadequate to capture on average $82\%$ of the values we extracted. Finally, we show that our approach can not only extract most of the qualitatively-identified values from prior taxonomies, but also uncover new values that are actually encouraged in practice. Our findings highlight the need for nuanced models of desirability that go beyond preexisting prosocial measures. This work has implications for improving moderator understanding of their community values and provides a framework that can supplement qualitative approaches with larger-scale content analyses.

cs.HC↗

arXiv2024

Does Positive Reinforcement Work?: A Quasi-Experimental Study of the Effects of Positive Feedback on Reddit

Social media platform design often incorporates explicit signals of positive feedback. Some moderators provide positive feedback with the goal of positive reinforcement, but are often unsure of their ability to actually influence user behavior. Despite its widespread use and theory touting positive feedback as crucial for user motivation, its effect on recipients is relatively unknown. This paper examines how positive feedback impacts Reddit users and evaluates its differential effects to understand who benefits most from receiving positive feedback. Through a causal inference study of 11M posts across 4 months, we find that users who received positive feedback made more frequent (2% per day) and higher quality (57% higher score; 2% fewer removals per day) posts compared to a set of matched control users. Our findings highlight the need for platforms, communities, and moderators to expand their perspective on moderation and complement punitive approaches with positive reinforcement strategies to foster desirable behavior online.

cs.HC↗

arXiv2024

Creator Hearts: Investigating the Impact Positive Signals from YouTube Creators in Shaping Comment Section Behavior

Much of the research in online moderation focuses on punitive actions. However, emerging research has shown that positive reinforcement is effective at encouraging desirable behavior on online platforms. We extend this research by studying the "creator heart" feature on YouTube, quantifying their primary effects on comments that receive hearts and on videos where hearts have been given. We find that creator hearts increased the visibility of comments, and increased the amount of positive engagement they received from other users. We also find that the presence of a creator hearted comment soon after a video is published can incentivize viewers to comment, increasing the total engagement with the video over time. We discuss the potential for creators to use hearts to shape behavior in their communities by highlighting, rewarding, and incentivizing desirable behaviors from users. We discuss avenues for extending our study to understanding positive signals from moderators on other platforms.

cs.HC↗

arXiv2021

Inter-Sense: An Investigation of Sensory Blending in Fiction

This study reports on the semantic organization of English sensory descriptors of the five basic senses of sight, hearing, touch, taste, and smell in a large corpus of over 8,000 fiction books. We introduce a large-scale text data-driven approach based on distributional-semantic word embeddings to identify and extract these descriptors as well as analyze their mixing interconnections in the resulting conceptual and sensory space. The findings are relevant for research on concept acquisition and representation, as well as for applications that can benefit from a better understanding of perceptual spaces of sensory experiences, in fiction, in particular, and in language in general.

cs.CL↗