An Emotional Label-Based Recommendation Method for Electronic Platforms

Personalized recommendation systems aim to model user preferences, but traditional collaborative filtering methods often rely solely on numerical ratings while neglecting emotional context, which limits their accuracy and applicability. To address this gap, this paper proposes an emotional label and SenticNet-based collaborative filtering approach (LSCF) that integrates facial emotion features into recommendation modeling. With user consent, facial expressions are captured and classified via deep learning to generate emotional labels. The method reduces ambiguity by calculating the similarity between emotional labels and item labels, and further accounts for the potential impact of users’ pre-interaction emotional states—emotional fluctuations are measured using variance to avoid biases caused by sustained non-interactive emotions. In neighbor selection, both rating similarity and emotional response similarity are considered. Experiments on MovieLens datasets demonstrate that the proposed method better aligns with user preferences and significantly improves recommendation performance.