A Context-Aware Movie Recommender System Through Semantic Analysis and Visual Comprehension-Driven Multimodal Fusion

Recommender systems have been a kind of effective tool in online movie websites. However, existing recommender systems mostly relied on user ratings and tag information, neglecting deep-level situational awareness and visual understanding. In response to this issue, this paper proposes a context-aware movie recommendation system based on semantic analysis and visual comprehension. Firstly, semantic analysis for users is conducted to perceive their emotional tendencies and interest preferences. At the same time, visual understanding techniques are used to extract and analyze the visual features of movies, in order to increase diversity and accuracy of recommendation results. Finally, similarity between users’ interest vectors and movies’ feature vectors is calculated to obtain preliminary recommendation results. And a context-aware module is introduced to consider users’ current emotional state and environmental information, in order to adjust and optimize initial recommendation results. In addition, some experiments are conducted on the MovieLens-1M dataset and the Avazu dataset. The experimental results show that the proposal improves recommendation accuracy compared to several typical methods. And the satisfaction of users with recommendation results has been effectively improved.