Deep Collaborative Filtering Recommendation Algorithm Based On Attention Mechanism

The traditional collaborative filtering recommendation algorithm analyzes the user's behavior data,mines the user's behavior pattern,models the user's interest,and recommends the user's favorite items according to the user's interest preference,guiding the user to discover them from a large number of options.Products or services of interest.However,when calculating the similarity between items,only the historical item scores are considered,and the influence of historical item preferences on them is neglected,so that the recommendation accuracy is not ideal,the personalized service is lacking,and the recommended interpretability is poor.To solve this issue,this paper proposes a movie recommendation algorithm based on deep learning and attention mechanism.Considering that traditional collaborative filtering lacks personalized recommendations,the attention mechanism is introduced to assign individualized weights,and multi-layer perceptrons are used to increase the flexibility and nonlinearity of the system.For the implicit feedback problem with only binary 0 or 1,the item-level and feature-level attention mechanism framework is designed for user history preferences.First,on the feature-level attention frame,starting from the item content feature extraction network,learning the preferences of item features.Then the item feature preferences and item features are weighted to obtain the item content feature vector.Finally,in the item-level feature attention frame,we obtain the final recommendation results through the scores of the item preferences learned by the item content feature vector.The experiment was conducted on two public datasets,MovieLens100 K and MovieLens 1M.The results show that the proposed recommendation algorithm improves the accuracy of similarity calculation,and at the same time,it has higher degree of recommendation accuracy than traditional algorithms,and enhances the interpretability of personalized recommendation,showing superior recommendation performance.This paper has 27 figures,9 tables and 70 references.