Refusals must be resilient, not brittle.” Yet guarding refusals against adversarial phrasing and shifting user contexts remains difficult: large language models (LLMs) still yield to jailbreak prompts that evade safety filters and surface harmful content. We propose Refusal Activation Steering (RAS), a training-free, inference-time method that uses contrastive activations to shift LLM responses, biasing generation trajectories toward refusals without altering model weights. The approach is modular and domain-targetable, avoiding collateral refusals on benign queries while strengthening activation- space boundaries for unsafe content. On adversarial evaluations with an 8B instruction-tuned model, we find that steering improves refusal rate by ∼ 52% and reduces attack success rate by ∼ 40%, establishing a lightweight and interpretable safety layer for robust refusal consistency. To foster further research in this domain, we have made our implementation publicly available.

We study when users end a session on X using high-resolution interaction logs from 215 US participants collected over four weeks. Sessions are defined via data-driven inter-activity gaps, and each session is encoded by fine-grained activity counts and duration (versus a simple activity ratio baseline). Fine-grained activity features substantially outperform the activity ratio baseline (C-index ≈ 0.76 vs. 0.62 for future sessions; 0.72 vs. 0.60 for unseen users), indicating that the composition of activity types is a strong predictor of disengagement. At the app level, we analyze retention over early adoption windows and find that the ratio of active activity in the first three days is most predictive of later usage. These results highlight session composition and early on-platform behavior as practical levers for forecasting and mitigating premature drop-off.

User behavior on social media---from scrolling and viewing to liking, reposting, and posting---yet most research relies on self-reports that obscure fine-grained usage patterns. We analyze high-resolution activity logs from 209 U.S. X (Twitter) users tracked over four weeks to identify distinct behavioral profiles based on session-level features. Latent profile analysis reveals three groups---interactors (32.52%), lurkers (60.45%), and creators (7.03%) that differ in engagement intensity, demographics, and content exposure. Interactors and lurkers skew younger and Democratic, whereas creators skew older and more Republican, consuming more ideological and low-credibility content. These results link behavioral heterogeneity to systematically different information environments and suggest that platform interventions may operate unevenly across user types. Our findings demonstrate the value of log-based behavioral segmentation for understanding online participation and motivate profile-aware platform governance and content moderation strategies.

Perceived trustworthiness underpins how users navigate online information, yet it remains unclear whether large language models (LLMs), increasingly embedded in search, recommendation, and conversational systems, represent this construct in psychologically coherent ways. We analyze how instruction-tuned LLMs (Llama 3.1 8B, Qwen 2.5 7B, Mistral 7B) encode perceived trustworthiness in web-like narratives using the PEACE-Reviews dataset annotated for cognitive appraisals, emotions, and behavioral intentions. Across models, systematic layer- and head-level activation differences distinguish high- from low-trust texts, revealing that trust cues are implicitly encoded during pretraining. Probing analyses show linearly decodable trust signals and fine-tuning effects that refine rather than restructure these representations. Strongest associations emerge with appraisals of fairness, certainty, and accountability-self– dimensions central to human trust formation online. These findings suggest that modern LLMs internalize psychologically grounded trust signals without explicit supervision, offering a representational foundation for designing credible, transparent, and trustworthy AI systems in the web ecosystem. Code and appendix are available at: https://github.com/GerardYeo/TrustworthinessLLM.

The prevalence of multi-modal content on social media complicates automated moderation strategies. This calls for an enhancement in multi-modal classification and a deeper understanding of understated meanings in images and memes. Although previous efforts have aimed at improving model performance through fine-tuning, few have explored an end-to-end optimization pipeline that accounts for modalities, prompting, labelling, and fine-tuning. In this study, we propose an end-to-end conceptual framework for model opti- mization in complex tasks. Experiments support the efficacy of this traditional yet novel framework, achieving the highest accuracy and AUROC. Ablation experiments demonstrate that isolated optimisations are not ineffective on their own.

Large language models (LLMs) are trained on vast amounts of publicly available text. However, the current training frameworks take for granted that these annotations are accurate reflections of the authors’ true intents. This study questions that assumption by examining the gaps between writers’ actual psychological states and the inferences made by third-party annotators. We explore how readers interpret psychological cues in text and demonstrate that third-person annotations often fail to align with first-person realities. By integrating both first- and third-person annotations, we develop computational models that reveal significant biases in how psychological states are perceived and the downstream effects these perceptions have on reader behavior. Our findings challenge the foundational assumptions of LLM training, suggesting that the reliance on potentially flawed third-person annotations could impact model accuracy and real-world applications.

Large language models (LLMs) play a key role in generating evidence-based and stylistic counter-arguments, yet their effectiveness in real-world applications has been underexplored. Previous research often neglects the balance between evidentiality and style, which are crucial for persuasive arguments. To address this, we evaluated the effectiveness of stylized evidence-based counter-argument generation in Counterfire, a new dataset of 38,000 counter-arguments generated by revising counter-arguments to Reddit’s ChangeMyView community to follow different discursive styles. We evaluated generic and stylized counter-arguments from basic and fine-tuned models such as GPT-3.5, PaLM-2, and Koala-13B, as well as newer models (GPT-4o, Claude Haiku, LLaMA-3.1) focusing on rhetorical quality and persuasiveness. Our findings reveals that humans prefer stylized counter-arguments over the original outputs, with GPT-3.5 Turbo performing well, though still not reaching human standards of rhetorical quality nor persuasiveness. Additionally, our work created a novel argument triplets dataset for studying style control, with human preference labels that provide insights into the tradeoffs between evidence integration and argument quality.

The use of LLMs in natural language reasoning has shown mixed results, sometimes rivaling or even surpassing human performance in simpler classification tasks while struggling with social-cognitive reasoning, a domain where humans naturally excel. These differences have been attributed to many factors, such as variations in prompting and the specific LLMs used. However, no reasons appear conclusive, and no clear mechanisms have been established in prior work. In this study, we empirically evaluate how role-playing persona-based prompting influences Theory-of-Mind (ToM) reasoning capabilities. Grounding our research in psychological theory, we found that, beyond the inherent variance in the complexity of reasoning tasks, ToM performance differences arise because of socially-motivated prompting differences. In an era where prompt engineering with role-play is a typical approach to adapt LLMs to new contexts, our research advocates caution as models that adopt specific personas might potentially result in errors in social-cognitive reasoning.

Supervised machine-learning models often underperform in predicting user behaviors from conversational text, hindered by poor crowdsourced label quality and low NLP task accuracy. We introduce the Metadata-Sensitive Weighted-Encoding Ensemble Model (MSWEEM), which integrates annotator meta-features like fatigue and speeding. First, our results show MSWEEM outperforms standard ensembles by 14% on held-out data and 12% on an alternative dataset. Second, we find that incorporating signals of annotator behavior, such as speed and fatigue, significantly boosts model performance. Third, we find that annotators with higher qualifications, such as Master's, deliver more consistent and faster annotations. Given the increasing uncertainty over annotation quality, our experiments show that understanding annotator patterns is crucial for enhancing model accuracy in user behavior prediction.

Achieving fairness in Large Language Models (LLMs) continues to pose a persistent challenge, as these models are prone to inheriting biases from their training data, which can subsequently impact their performance in various applications. There is a need to systematically explore whether structured prompting techniques can offer opportunities for debiased text generation by LLMs. In this work, we designed an evaluative framework to test the efficacy of different prompting techniques for debiasing text along different dimensions. We aim to devise a general structured prompting approach to achieve fairness that generalizes well to different texts and LLMs.

Shaz Furniturewala, Surgan Jandial, Abhinav Java, Pragyan Banerjee, Simra Shahid, Sumit Bhatia, Kokil Jaidka. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Online games are dynamic environments where players interact with each other, which offers a rich setting for understanding how players negotiate their way through the game to an ultimate victory. This work studies online player interactions during the turn-based strategy game, Diplomacy. We annotated a dataset of over 10,000 chat messages for different negotiation strategies and empirically examined their importance in predicting long- and short-term game outcomes. Although negotiation strategies can be predicted reasonably accurately through the linguistic modeling of the chat messages, more is needed for predicting short-term outcomes such as trustworthiness. On the other hand, they are essential in graph-aware reinforcement learning approaches to predict long-term outcomes, such as a player's success, based on their prior negotiation history. We close with a discussion of the implications and impact of our work.

The rapid growth of information and communication technologies in recent years, and the different forms of digital connectivity, have profoundly affected how news is generated and consumed. Digital traces and computational methods offer new opportunities to model and track the provenance of news. This project is the first study to characterize and predict how prominent news outlets make edits to news frames and their implications for geopolitical relationships and attitudes. We evaluate the feasibility of training few-shot learners on the editing patterns of articles discussing different countries, for understanding their wider implications in preserving or damaging geopolitical relationships.

At the end of October 2022, Elon Musk concluded his acquisition of Twitter. In the weeks and months before that, several questions were publicly discussed that were not only of interest to the platform's future buyers, but also of high relevance to the Computational Social Science research community. For example, how many active users does the platform have? What percentage of accounts on the site are bots? And, what are the dominating topics and sub-topical spheres on the platform? In a globally coordinated effort of 80 scholars to shed light on these questions, and to offer a dataset that will equip other researchers to do the same, we have collected all 375 million tweets published within a 24-hour time period starting on September 21, 2022. To the best of our knowledge, this is the first complete 24-hour Twitter dataset that is available for the research community. With it, the present work aims to accomplish two goals. First, we seek to answer the aforementioned questions and provide descriptive metrics about Twitter that can serve as references for other researchers. Second, we create a baseline dataset for future research that can be used to study the potential impact of the platform's ownership change.

Novel AI-generated audio samples are evaluated for descriptive qualities such as the smoothness of a morph using crowdsourced human listening tests. However, the methods to design interfaces for such experiments and to effectively articulate the descriptive audio quality under test receive very little attention in the evaluation metrics literature. In this paper, we explore the use of visual metaphors of image-schema to design interfaces to evaluate AI-generated audio. Furthermore, we highlight the importance of framing and contextualizing a descriptive audio quality under measurement using such constructs. Using both pitched sounds and textures, we conduct two sets of experiments to investigate how the quality of responses vary with audio and task complexities. Our results show that, in both cases, by using image-schemas we can improve the quality and consensus of AI-generated audio evaluations. Our findings reinforce the importance of interface design for listening tests and stationary visual constructs to communicate temporal qualities of AI-generated audio samples, especially to naïve listeners on crowdsourced platforms.

This abstract proposes an approach towards goal-oriented modeling of the detection and modeling complex social phenomena in multiparty discourse in an online political strategy game. We developed a two-tier approach that first encodes sociolinguistic behavior as linguistic features then use reinforcement learning to estimate the advantage afforded to any player. In the first tier, sociolinguistic behavior, such as Friendship and Reasoning, that speakers use to influence others are encoded as linguistic features to identify the persuasive strategies applied by each player in simultaneous two-party dialogues. In the second tier, a reinforcement learning approach is used to estimate a graph-aware reward function to quantify the advantage afforded to each player based on their standing in this multiparty setup. We apply this technique to the game Diplomacy, using a dataset comprising of over 15,000 messages exchanged between 78 users. Our graph-aware approach shows robust performance compared to a context-agnostic setup.

Psychological concepts can help computational linguists to better model the latent semantic spaces of emotions, and understand the underlying states motivating the sharing or suppressing of emotions. This abstract applies the understanding of agency and social interaction in the happiness semantic space to its role in positive emotion. First, BERT-based fine-tuning yields an expanded seed set to understand the vocabulary of the latent space. Next, results benchmarked against many emotion datasets suggest that the approach is valid, robust, offers an improvement over direct prediction, and is useful for downstream predictive tasks related to psychological states.

In contexts where debate and deliberation is the norm, participants are regularly presented with new information that conflicts with their original beliefs. When required to update their beliefs (belief alignment), they may choose arguments that align with their worldview (confirmation bias). We test this and competing hypotheses in a constraint-based modeling approach to predict the winning arguments in multi-party interactions in the Reddit ChangeMyView dataset. We impose structural constraints that reflect competing hypotheses on a hierarchical generative Variational Auto-encoder. Our findings suggest that when arguments are further from the initial belief state of the target, they are more likely to succeed.

In contexts where debate and deliberation are the norm, the participants are regularly presented with new information that conflicts with their original beliefs.When required to update their beliefs (belief alignment), they may choose arguments that align with their worldview (confirmation bias).We test this and competing hypotheses in a constraintbased modeling approach to predict the winning arguments in multi-party interactions in the Reddit Change My View and Intelligence Squared debates datasets.We adopt a hierarchical generative Variational Autoencoder as our model and impose structural constraints that reflect competing hypotheses about the nature of argumentation.Our findings suggest that in most settings, predictive models that anticipate winning arguments to be further from the initial argument of the opinion holder are more likely to succeed.

This paper motivates and presents the Twitter Deliberative Politics dataset, a corpus of political tweets labeled for its deliberative characteristics. The corpus was randomly sampled from replies to US congressmen and women. It is expected to be useful to a general community of computational linguists, political scientists, and social scientists interested in the study of online political expression, computer-mediated communication, and political deliberation. The data sampling and annotation methods are discussed and classical machine learning approaches are evaluated for their predictive performance on the different deliberative facets. The paper concludes with a discussion of future work aimed at developing dictionaries for the quality assessment of online political talk in English. The dataset and a demo dashboard are available at https://github.com/kj2013/twitter-deliberative-politics.

Kokil Jaidka, Andrea Ceolin, Iknoor Singh, Niyati Chhaya, Lyle Ungar. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Learning word representations from large corpora relies on the distributional hypothesis that words present in similar contexts tend to have similar meanings. Recent work has shown that word representations learnt in this manner lack sentiment information which, can be introduced using external knowledge. Our work addresses the question: Can affect lexica improve word representations learnt from a corpus ? In this work, we propose techniques to incorporate affect lexica, which capture fine-grained information about a word's psycholinguistic and emotion orientation, into the training for Word2Vec SkipGram, Word2Vec CBOW, and GloVe using a Joint Learning approach. We use affect scores from the Warriner's affect lexicon to regularize the vector representations learnt from an unlabeled corpus. Our proposed method outperforms previously methods on standard tasks for word similarity detection, outlier detection, and sentiment analysis. We also show the usefulness of our approach for the prediction of formality, frustration, and politeness in text.

Individuals express their locus of control, or "control", in their language when they identify whether or not they are in control of their circumstances.Although control is a core concept underlying rhetorical style, it is not clear whether control is expressed by how or by what authors write.We explore the roles of syntax and semantics in expressing users' sense of control -i.e.being "controlled by" or "in control of" their circumstances-in a corpus of annotated Facebook posts.We present rich insights into these linguistic aspects and find that while the language signaling control is easy to identify, it is more challenging to label it is internally or externally controlled, with lexical features outperforming syntactic features at the task.Our findings could have important implications for studying selfexpression in social media.

The large scale of scholarly publications poses a challenge for scholars in information seeking and sensemaking. Information retrieval~(IR), bibliometric and natural language processing (NLP) techniques could enhance scholarly search, retrieval and user experience but are not yet widely used. To this purpose, we propose the third iteration of the Joint Workshop on Bibliometric-enhanced Information Retrieval and Natural Language Processing for Digital Libraries (BIRNDL). The workshop is intended to stimulate IR, NLP researchers and Digital Library professionals to elaborate on new approaches in natural language processing, information retrieval, scientometrics, text mining and recommendation techniques that can advance the state-of-the-art in scholarly document understanding, analysis, and retrieval at scale. The BIRNDL workshop will incorporate multiple invited talks, paper sessions, a poster session and the 4th edition of the Computational Linguistics (CL) Scientific Summarization Shared Task.

Several studies have demonstrated how language models of user attributes, such as personality, can be built by using the Facebook language of social media users in conjunction with their responses to psychology questionnaires. It is challenging to apply these models to make general predictions about attributes of communities, such as personality distributions across US counties, because it requires 1. the potentially inavailability of the original training data because of privacy and ethical regulations, 2. adapting Facebook language models to Twitter language without retraining the model, and 3. adapting from users to county-level collections of tweets. We propose a two-step algorithm, Target Side Domain Adaptation (TSDA) for such domain adaptation when no labeled Twitter/county data is available. TSDA corrects for the different word distributions between Facebook and Twitter and for the varying word distributions across counties by adjusting target side word frequencies; no changes to the trained model are made. In the case of predicting the Big Five county-level personality traits, TSDA outperforms a state-of-the-art domain adaptation method, gives county-level predictions that have fewer extreme outliers, higher year-to-year stability, and higher correlation with county-level outcomes.

The large scale of scholarly publications poses a challenge for scholars in information seeking and sensemaking. Bibliometrics, information retrieval (IR), text mining and NLP techniques could help in these search and look-up activities, but are not yet widely used. This workshop is intended to stimulate IR researchers and digital library professionals to elaborate on new approaches in natural language processing, information retrieval, scientometrics, text mining and recommendation techniques that can advance the state-of-the-art in scholarly document understanding, analysis, and retrieval at scale. The BIRNDL workshop at SIGIR 2017 will incorporate an invited talk, paper sessions and the third edition of the Computational Linguistics (CL) Scientific Summarization Shared Task.

This study investigates how politicians and citizens cooperate to create an e-democracy based on information dissemination and political awareness in an ICT environment, especially during elections. This research is conducted in the Indian context, where new ICT channels, such as Facebook and Twitter, were extensively used for campaigning and citizen engagement prior to elections. We downloaded 98,396 tweets posted by the official Twitter accounts of the top ten political parties during a two month period prior to election and conducted a three-level analysis to identify the overall trend in usage, the interactive characteristics of tweets and the functions driving the Twitter usage of political parties. Our findings show that the more successful parties used Twitter to push timely updates on online and offline campaign activities, to their followers. The exemplary use of Twitter for campaigns was by the Bharatiya Janta Party (BJP) Twitter account for interacting with the public, and by the Aam Aadmi Party (AAP) account for self-promotion and highlighting its party manifesto. Further, we identify the new paradigms created by political parties to engage and inform voters, driven on modern ICT. Our study is the first in analyzing Twitter usage by Indian political parties, and its findings corroborate seminal research in other developing countries.