Movie Recommendation Model Based on Attention Mechanism for Dynamically Capturing User Interest Evolution
Shiliang Gu, Chaobin Wang, Gang Zhao, Linfeng Wu · 2023
With the development of the Internet and big data, there is an increasing demand for movie recommendation systems. Faced with a massive amount of movie information online, it is difficult for users to quickly find movies they like. In this paper, we propose a movie recommendation model called Interest Dynamic Evolution Model (IDEM) based on a recurrent neural network (RNN) and incorporated with an attention mechanism to dynamically capture user interest evolution. The model captures temporal interests from user’s historical behavior sequences and introduces an auxiliary loss function to supervise interest extraction at each step. The attention mechanism is then embedded into the interest sequences, giving higher weights to interest sequences with higher similarity to the set of movies to be recommended. Finally, the vector representing the user’s interest sequence is concatenated with other information and fed into a deep neural network. Experimental results on publicly available datasets, MovieLens and Amazon (Electro), demonstrate that IDEM outperforms other state-of-the-art solutions significantly.