Go to ACL ARR 2025 February homepageiNews: A Multimodal Dataset for Modeling Personalized Affective Responses to News
Abstract: Current NLP approaches to emotion detection predominantly focus on the writer's perspective and overlook the subjective affective experience across groups and individuals, relying on aggregation techniques that obscure important nuances in reader reactions. To address this gap, we introduce iNews, a novel large-scale dataset capturing subjective affective responses to news headlines from reader's perspective. Our dataset comprises annotations from 291 participants across 2,899 multimodal Facebook news posts from major UK outlets, with an average of 5.18 annotators per sample. For each post, annotators provide multifaceted labels including valence, arousal, dominance, discrete emotion labels, content relevance, sharing likelihood, and modality importance (text, image, or both). Furthermore, we collect rich persona variables covering demographics, personality traits, media trust, and consumption patterns, which explain 15.2\% of annotation variance - higher than existing NLP datasets. iNews will enhance research in LLM personalization, subjectivity, affective computing, and individual-level human behavior simulation.
Paper Type: Long
Research Area: Human-Centered NLP
Research Area Keywords: human factors in NLP, human-centered evaluation, emotion detection and analysis, corpus creation, multimodality
Languages Studied: English
Submission Number: 2238
Loading