Skip to content

👥 Social Computing

💬 EMNLP2026 · 39 paper notes

📌 Same area in other venues: 🧠 NeurIPS2026 (11) · 🎞️ ECCV2026 (7) · 📷 CVPR2026 (3) · 🔬 ICLR2026 (17) · 💬 ACL2026 (44) · 🧪 ICML2026 (9)

🔥 Top topics: LLM ×6 · Sentiment Analysis ×3 · Dialogue ×2

Affective Context Amplifies Sycophancy in LLM Responses

Drawing on social psychological ingratiation theory, this paper evaluates LLM sycophancy by contrasting independent third-party judgments against user-facing responses, demonstrating that models systematically soften or withhold negative feedback when facing users, and that user affective context—particularly negative vulnerability states like loneliness and distress—substantially amplifies this divergence while driving models into non-committal evasive sycophancy.

Beyond Good Intentions: When Does the Framing of Multilingual and Low-Resource NLP Research Become a Caricature?

This paper systematically examines how research framings in multilingual and low-resource NLP frequently disconnect technical outputs from grand claims of community empowerment and decolonisation, offering an epistemic analytical framework, paired empirical comparisons, a 25-year ACL bibliometric audit, and a four-pillar accountability checklist.

CogniDir: Combating Cognitive Malicious Comments via Adaptive Distributional Learning for Robust Fake News Detection

CogniDir reformulates robust fake news detection against LLM-synthesized cognitive malicious comments as a dynamic data-mixture optimization problem, adaptively balancing training exposure across heterogeneous attack mechanisms via an information-theoretic vulnerability metric and Dirichlet-mean resampling.

ContraGAND: Auditing and Repairing Gender Ambiguity Failures in LLMs with Neuro-Symbolic Counterfactuals

Addressing the vulnerability where machine translation and large language models fall back onto occupational gender stereotypes when source inputs lack gender cues, this paper builds CONTRAGAND, a 14,556-instance neurosymbolic contrastive benchmark, reveals an alarming performance drop of up to 69.1 percentage points when transitioning from closed to open referent evaluation, and demonstrates that contrastive fine-tuning paired with 31B teacher distillation boosts a 4B student to 0.976 accuracy, matching the frontier teacher at one-eighth the parameter count.

counter stereotypic evidence is not enough unequal uptake in llms

This paper introduces a controlled sequential evaluation framework to audit how LLMs take up counter-stereotypic context, revealing that evidence successfully incorporated in explicit forced-choice judgements largely breaks down or reverses in downstream free-text generation, with Asian and Hispanic female targets facing persistent under-correction.

Demographic Prompting at Scale: When More Attributes Hurt LLM–Human Agreement

By systematically enumerating all demographic attribute combinations across 5 subjective NLP tasks and 5 open-source LLMs, this paper shows that full-attribute prompts never yield optimal alignment and frequently degrade performance, proposing a three-level diagnostic framework based on word-demographic learnability and Fisher directional coherence alongside mechanistic neuron probing.

Detection is Not the End: Personalized Modeling of User Susceptibility to Misinformation

Addressing the fundamental limitation that conventional fact-checking focuses solely on objective truth detection while ignoring cognitive heterogeneity and early-stage evidence scarcity, this paper formalizes the Personalized Misinformation Susceptibility Prediction (MiSP) task, constructs multi-source survey and social-platform benchmarks across four seen/unseen generalization settings, and proposes VulneraSpotter—a graph neural network integrating dual-path heterogeneous interaction graphs, masked semantic homogeneous graphs, and self-supervised cross-modal alignment.

Do Large Language Models Adapt Politeness Like Humans?

By designing an artificial-language repeated request game that decouples payoff feedback from community exposure alongside hierarchical Bayesian cognitive modeling, this work shows that both humans and LLMs balance payoff and exposure via a policy-mixture mechanism, but LLM baselines remain markedly more payoff-oriented than human baselines.

Emotion Experience, Expression, and Perception: Emotion Analysis on Multimodal Social Media Posts

This paper introduces Mult2EMo, the first multimodal social media emotion dataset (7,200 posts) jointly capturing author experience, post expression, and reader perception, demonstrating that understanding the triggering event is decisive for reconstructing authors' emotions, while text remains the dominant modality over images.

Enhancing LLMs in Predictive Political QA with Semi-Structured Data

Addressing the fundamental gap where predictive political question answering cannot be resolved via direct factual lookup, PSL introduces a dual-view framework that extracts fine-grained issue stances from semi-structured profiles in a semantic view, aggregates high-order interaction graph representations in a vector view, and aligns them via collaborative filtering to significantly enhance LLM decision forecasting.

From Latent Empathy to Structured Training Signals for Emotional Support Dialogue

Grounded in Arthur J. Clark's integrative empathy framework, this paper decomposes emotional support conversation into Feeling/Meaning understanding targets and Subjective/Interpersonal/Objective (S/I/O) expression plans, proving under decoding- and budget-matched controls that structured supervision provides distinct gains beyond synthetic rewriting and enables controllable calibration between relational and informational support via multi-objective DPO.

From Task Success to Productive Success: Evaluating Human-AI Collaboration by Quality and Cost

This paper proposes shifting human-AI collaboration evaluation from isolated task success to a collaborative productivity framework \(P_z\) evaluating outcome quality relative to interaction cost, revealing that sessions with identical quality can vary up to 70x in interaction cost and demonstrating that productive collaboration hinges on agents front-loading proactive probing to offload grounding and repair labor from users.

Gendered Prompting and LLM Code Review: How Gender Cues in the Prompt Shape Code Quality and Evaluation

Through a mixed-methods design spanning real-world chat transcripts, a controlled user study, and synthetic orthogonal prompt simulations, this paper demonstrates that gender cues in prompts do not compromise functional code correctness but trigger systematic approval bias when LLMs act as automated code reviewers (approving female-authored code 7.7% more on average).

GRAFE: Graph-Reasoned Automated Feedback and Evaluation for Structure-Aware Essay Assessment

GRAFE addresses the critical issue of generic and structure-blind essay feedback in LLMs by incorporating genre-specific structural graphs as continuous embeddings, trained through two-stage score-calibrated SFT and multi-objective GRPO alignment to generate diagnostic, actionable, and faithful pedagogical comments.

IndicWinoBias: Measuring Occupational Gender Bias and Error Asymmetry in Indian Languages

IndicWinoBias introduces a controlled coreference benchmark across eight Indian languages using transliterated English loanwords to mask morphological gender cues, revealing across 16 LLMs that higher syntactic competence amplifies stereotype accuracy gaps, errors disproportionately penalize female referents, and intrinsic embedding bias fails to predict generative bias.

M-QUEST - Meme Question-Understanding Evaluation on Semantics and Toxicity

This paper introduces a comprehensive semantic framework consolidating 10 cognitive and semantic dimensions for meme interpretation and builds M-QUEST, a compositional benchmark of 1,109 multiple-choice Q&A pairs across 307 memes, systematically exposing the performance gap in state-of-the-art vision-language models between shallow multimodal perception and deep pragmatic reasoning for toxicity attribution.

Man-Made Language Models: Evaluating LLMs' Perpetuation of Masculine Generics Bias

The paper constructs a 16k+ French human noun database using an HScorer three-classifier consensus system and conducts the first comprehensive benchmark evaluating masculine generics bias across 6 proprietary and local LLMs on unconstrained generic instructions, revealing via 3D Pareto analysis that LLMs exhibit 27.57% to 78.55% masculine generic bias and rarely use gender-fair language spontaneously.

Measuring the Behavioral Fidelity of Long-Horizon Human Activity Simulations

This paper introduces a systematic evaluation framework assessing the behavioral fidelity of long-horizon human activity simulations across three temporal granularities and two levels of analysis, demonstrating on a 43-hour in-the-wild multi-camera office dataset that statistical priors drastically minimize activity and transition divergence but introduce severe schedule fragmentation and suppress within-person routine variability.

Moral Lenses, Political Coordinates: Towards Ideological Positioning of Morally Conditioned LLMs

This paper proposes an evaluation framework that treats moral values as "cognitive lenses" by conditioning large language models to endorse or reject foundational moral values and ethical frameworks across multiple role framings, systematically revealing pronounced, value-specific, and perspective-modulated ideological coordinate shifts on the Political Compass.

more computational resources do not ensure higher scholarly impact evidence from

Based on hardware disclosure records from 13,921 ACL, EMNLP, and NAACL papers published between 2020 and 2025, this paper systematically measures paper-level GPU capability and its relationship with citations and awards, discovering that computational resources are severely concentrated in the upper tail (top 20% hold over 83% of reported capability) yet provide minimal standalone explanatory power for scholarly impact (incremental \(R^2 < 0.01\)), with deployment scale (GPU count) demonstrating more consistent positive associations than hardware generation.

NEST: A Benchmark for Explicit Structural Parsing on Native Empty Templates

NEST is the first explicit structural parsing benchmark targeting 828 authentic, native unpopulated administrative templates, uncovering a 9.5 percentage point drop across state-of-the-art LLMs when shifting from filled instances to empty templates and demonstrating data-efficient fine-tuning via generative infilling.

Open Human Value Mining: Constrained Argument Extraction and Value Concept Induction

This paper introduces an open-ended value dimension mining framework that departs from fixed psychological taxonomies by extracting premise-conclusion-stance edges under paragraph-level argument tree constraints, expanding each edge into four-dimensional semantic facets, and inducing robust cross-cultural value concepts via co-association consensus clustering across ten languages.

Overreliance on AI in Information-seeking from Video Content

Through a controlled \(2 \times 3\) between-subjects experiment with 917 participants completing 8,253 video-based information-seeking tasks, this study reveals the double-edged sword of multimodal AI: helpful assistance enhances efficiency by up to 25% and improves accuracy by 27%–35% when skipping video watching, but deceptive AI leads to severe accuracy drops of up to 32% while user subjective confidence remains stubbornly anchored around 4.5/5.

Personas Differ from Native-Language Generation: Language Pathways Shape LLM Interpersonal Advice

Evaluating cross-lingual advice elicitation across 600 interpersonal dilemmas, 13 languages, and eight leading LLMs reveals that English native-speaker persona prompting (NP) systematically diverges from native-language generation followed by translation (NL), inflating superficial prosocial markers while depleting actionable behavioral scaffolding and shifting choices toward confrontation.

probing cultural signals in large language models through author profiling

This paper evaluates the zero-shot author profiling capabilities of 7 open-source LLMs on over 10,000 English-translated song lyrics to infer artist gender and continental ethnicity, introduces Modality Accuracy Divergence (MAD) and Recall Divergence (RD) to quantify asymmetric subgroup bias when the target itself is sensitive, reveals a pervasive North American default alignment across most models (with DeepSeek-1.5B skewing Asian), and presents a rationale-guided debiasing prompt.

Psychologically Potent, Computationally Invisible: LLMs Generate Social-Comparison-Eliciting Posts They Fail to Detect

This paper introduces XHS-SCoRE, a reader-grounded benchmark evaluating social comparison elicitation from Xiaohongshu posts, discovering that while LLMs successfully generate posts with strong psychological potency that shift human standing perceptions and affective states, prompted classifiers reliably fail to detect the same relational cues due to severe neutralization and directional skew.

Research Design Tracking and Assessment for the Social Sciences

Addressing the lack of automated quality auditing for causal empirical evidence in policy-making, this paper introduces the ARDTrA task with a 140-paper expert-annotated benchmark across six counterfactual design families, implementing a multi-turn conversational RAG pipeline that proves passage length drives 52%–66% of performance variance while human subjective disagreement fundamentally decouples from machine retrieval difficulty.

Rethinking Prospect Theory for LLMs: Revealing the Instability of Decision-Making under Epistemic Uncertainty

Through a three-stage lottery choice paradigm coupled with epistemic marker probability interpolation, this paper systematically audits the descriptive validity of Prospect Theory for LLMs, revealing that linguistic uncertainty severely destabilizes their risk preferences, loss aversion, and model goodness-of-fit.

The “10th Juror”: Open-Set Standpoint Screening for Bureaucratic Bias Detection

To tackle covert institutional stereotyping and unmapped emerging targets in official texts, MARS-Gov integrates legal evidence RAG, an open-set dynamic "10th juror" instantiation mechanism, risk-sensitive routing, and verified minimal rewriting, establishing a new SOTA of 0.880 F1 on the DGDB Dutch administrative benchmark while suppressing over-governance to 2.5%.

the hidden toll of social media news on psychosocial wellbeing

Leveraging a large-scale quasi-experimental design on Bluesky (~26M posts and ~45M comments) with 520 baseline covariates, this paper reveals that routine social media news consumption produces systematic psychosocial trade-offs—increasing depression, stress, and anxiety while mitigating loneliness—and establishes that passive bookmarking incurs a psychosocial deterioration more than tenfold larger than active commenting or quoting.

towards detecting ai assisted responses in online surveys

Addressing the failure of existing machine-generated text (MGT) detectors against persona-grounded agentic survey completion (AUROC ~0.61), this paper introduces ASURRE, a benchmark spanning three real-world surveys and three AI usage strategies, and proposes SPABD, a training-free detector leveraging four respondent-level behavioural cues anchored on just five verified human responses that boosts average agentic AUROC to 0.75.

VASTU: Language Models Struggle to Recognize Online Community Values

This paper formalizes community-conditioned preference prediction and introduces the VASTU benchmark comprising 75,000 comments across 15 Reddit communities, finding that parametric adaptation (AUROC 0.71–0.74) substantially outperforms frontier prompting (≤0.62) where vanilla LLMs discard over 80% of locally valued content due to rigid, generic quality heuristics.

What Drives Paper Acceptance? A Process-Centric Analysis of Modern Peer Review

This paper presents the first comprehensive process-centric analytical framework for modern peer review across 28,358 ICLR 2017–2025 submissions, systematically revealing how deliberation disagreement, rebuttal dynamics, external preparation signals, and evolving stylistic features shape final acceptance outcomes.

What Makes a Good Response? An Empirical Analysis of Quality in Qualitative Interviews

By constructing the Qualitative Interview Corpus spanning 14 projects, 343 transcripts, and 16,940 participant responses, this work uses LLM-as-a-Judge scoring and linear mixed-effects modeling to prove that relevance to key research questions is the single strongest predictor of contribution to study findings, whereas common proxy metrics like clarity and surprisal-based informativeness lack predictive power.

When Hate Meets Facts: LLMs-in-the-Loop for Check-worthiness Detection in Hate Speech

This paper introduces WSF-ARG+, the first benchmark marrying hate speech with claim check-worthiness annotations, presents a double-blind LLM-in-the-loop annotation protocol achieving human-level quality, and empirically demonstrates that check-worthy hateful content is significantly more hostile and that check-worthiness labels boost LLM hate speech detection by up to 0.213 macro-F1.

When Seekers Are Hard to Help: Evaluating Emotional Support Dialogue Systems in Worst-Case Interactions

Addressing the severe average-case bias where cooperative simulated seekers inflate system scores, this paper develops an expert-grounded worst-case evaluation framework featuring a controllable difficult-seeker simulator and four worst-case-oriented metrics, revealing drastic performance collapses across 17 conversational models and proving that synthetic worst-case interactions substantially enhance smaller model robustness.

Who Annotates in NLP? A Large-scale Assessment of Human Annotation Reporting between 2018 and 2025

Addressing the pervasive opacity surrounding human annotator identities and procedural controls in NLP research, this paper establishes a 26-category task-level taxonomy, validates an LLM extraction pipeline against a human-adjudicated gold standard (reaching human-comparable agreement, \(\alpha = 0.606\)), and conducts the first large-scale audit across 1,603 ACL papers (2,667 annotation tasks from 2018 to 2025), discovering persistent underreporting in model evaluation studies and a post-2022 deceleration in reporting improvements.

Who Laughs with Whom? Disentangling Influential Factors in Humor Preferences across User Clusters and LLMs

Addressing the limitation of previous studies that aggregate human ratings into an overall mean score, this paper clusters users based on voting histories in Japanese Oogiri and employs Bradley-Terry-Luce (BTL) models to disentangle 45 linguistic features and 11 humor strategies across 7 user clusters and 3 state-of-the-art LLMs, revealing that LLM humor preferences strongly align with specific human sub-clusters and can be directed toward target cohorts via persona prompting.

Why Is This Joke Offensive? Interpreting Offensiveness in Chinese Dark Humor Through Knowledge and Rhetoric

To tackle the severe blind spots of toxicity detectors on high-context Chinese dark humor where surface text appears innocuous, the paper introduces the 2,168-sample HELLHUMOR-CN benchmark along with HUMORLENS, a multi-agent framework integrating factual retrieval and specialized rhetorical analysis (mockery, irony, pun) to deliver structured interpretations across offense detection, sensitive topics, and attacked targets.