A Hybrid YOLOv11 and LightFM Model for Emotion-Driven Anime Recommendation

Abstract

Existing anime recommendation systems predominantly focus on genre preferences and viewing history without considering users' emotional states, leading to context-blind recommendations that may exacerbate negative moods and reduce viewing satisfaction. This study addresses this gap by developing an emotion-based anime recommendation system integrating YOLOv11 for facial emotion recognition with hybrid collaborative filtering using LightFM. The research objectives are to achieve superior emotion classification accuracy, enhance recommendation quality through hybrid modeling, and prevent filter bubbles through diversification mechanisms. The methodology employed the KDEF dataset (3,597 images, five emotion classes) for training YOLOv11 with data augmentation, and the MyAnimeList dataset (744,330 interactions) for recommendation modeling. Emotion-to-genre mappings informed by neuropsychological research were implemented, and Maximum Marginal Relevance (MMR) diversification with λ=0.7 was applied to balance relevance and variety. The YOLOv11 model achieved 97.62% training accuracy and 93.70% validation accuracy, outperforming existing CNN-LSTM approaches by 37.55 percentage points. The hybrid recommendation model demonstrated test AUC of 0.8567 and Precision@10 of 0.1457, representing a 417% improvement over pure collaborative filtering with high statistical significance (Cohen's d = 0.9837, p<0.001). The diversification strategy successfully recommended anime spanning 15 unique genres, preventing monotonous suggestions. This system has practical applications for streaming platforms, mental health support systems, and personalized entertainment services requiring real-time affective computing. The findings confirm that integrating real-time emotion detection with hybrid collaborative filtering effectively enhances recommendation quality while addressing context-unawareness, cold-start problems, and filter bubbles, though future work should address limitations in Precision@10 performance and cross-cultural validation.

GS Cited Analysis