A Movie Recommendation Algorithm Combining Time Weight GRU and Attention

In movie recommendation, how to analyze the users’ needs according to their historical behavior to realize the accurate recommendation of movies is a hotspot in the research of movie recommendation algorithm. The traditional GRU model simply infers the user’s final interest preference according to the historical behavior sequence, but does not consider the time decay of user’s interest preference and the weight factor of interest preference, which results in the same weight allocation of interest preferences learned from different types of movies, reducing the recommendation accuracy of movies, and lacking diversity of recommendation results. In order to solve this problem, this paper constructs a movie recommendation algorithm - ATGRU, which integrates time weight GRU and attention mechanism. Through analyzing the user’s historical behavior record, we can capture the user-related interest changes and predict the user’s real-time interest preference accurately. Experiments on the open dataset Movie Lens 10M show that compared with two traditional recommendation algorithms, the HR and NDCG indicators are improved by 14% and 4% and 28.3% and 3.6% respectively. The experimental results show that the proposed ATGRU movie recommendation algorithm has higher recommendation performance.