Leveraging Descriptions of Emotional Preferences in Recommender Systems

2025-05-26

AI Summary

This paper introduces the novel paradigm of *affective matching recommendation*, which aims to recommend items capable of evoking target emotional states—based on users’ explicit, fine-grained affective preferences (e.g., “feeling pleasantly surprised by the ending”)—rather than merely modeling coarse-grained liking/disliking. Methodologically, it pioneers the use of open-domain, explicit affective descriptions as primary recommendation signals; constructs the first large-scale book review dataset annotated for affective preferences; and employs a Transformer-based dual-encoder architecture to jointly encode user affective texts and item content, enabling cross-modal affect–content alignment. Compared to conventional collaborative filtering and content-based approaches, the proposed model achieves a 19.3% improvement in affective matching accuracy. This work advances recommender systems from *preference modeling* toward *affective intent modeling*, establishing a foundation for emotion-aware personalization.

Technology Category

• Cognitive Modeling & Cognitive Systems: Affective Computing
• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Recommender Systems

Application Category

• User Modeling, Personalization and Recommendation: Psychology-informed user models and recommender systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economics and fairness of platforms and recommendation systems

Abstract

The affective attitude of liking a recommended item reflects just one category in a wide spectrum of affective phenomena that also includes emotions such as entranced or intrigued, moods such as cheerful or buoyant, as well as more fine-grained affective states, such as"pleasantly surprised by the conclusion". In this paper, we introduce a novel recommendation task that can leverage a virtually unbounded range of affective states sought explicitly by the user in order to identify items that, upon consumption, are likely to induce those affective states. Correspondingly, we create a large dataset of user preferences containing expressions of fine-grained affective states that are mined from book reviews, and propose a Transformer-based architecture that leverages such affective expressions as input. We then use the resulting dataset of affective states preferences, together with the linked users and their histories of book readings, ratings, and reviews, to train and evaluate multiple recommendation models on the task of matching recommended items with affective preferences. Experiments show that the best results are obtained by models that can utilize textual descriptions of items and user affective preferences.

Problem

• Expanding recommender systems to incorporate diverse user emotional preferences
• Creating dataset from book reviews to capture fine-grained affective states
• Developing Transformer models to match items with user affective preferences

Innovation

• Leveraging user affective states in recommendations
• Transformer-based architecture for affective expressions
• Utilizing textual descriptions for affective preferences