Research On Movie Recommendation Method Based On Multi-head Attention Mechanism

Today,with the continuous development and maturity of computer and multimedia technologies,watching movies has gradually become people’s daily entertainment,but the arrival of Internet big data has also brought about the problem of information overload.How to find what the user likes in a large number and variety of movies is also the primary issue in the research of recommendation systems.In terms of movie recommendation,the traditional collaborative filtering recommendation algorithm mainly uses the user’s rating of the movie as the basis for recommendation.However,most important information such as the user’s age,occupation,hobbies,movie introduction,and comments are not fully utilized by users and movies,the disadvantage of this is the lack of precise positioning of user behavior preferences.This paper proposes a movie recommendation model CBLSC based on user preference learning.In addition,in order to solve the problem of parallel execution and data sparsity,this paper proposes a user personalized movie recommendation model CBLAMF based on the self-attention mechanism based on the CBLSC model.The comparative experimental analysis of different models shows that our proposed CBLAMF model can obtain better performance than other methods.The detailed research contents of this article are as follows:(1)Propose a movie recommendation CBLSC model based on user preference learning.The model can divide the introduction of a single movie into multiple sentences,and then extract the feature representation through the convolutional layer neural network in order to fully understand the feature information of the introduction of the movie.Then use LSTM to sequentially integrate these sentence features to construct the entire sentence feature representation,so that it can combine the context information in the movie introduction to capture the subtle differences between words,so as to more accurately obtain the key information in the movie introduction.After obtaining the features in the movie profile,and finally combining the user and other attribute features of the movie and the rating information of the movie to perform similarity calculation,the recommendation results of the top N movies recommended for the user can be obtained.(2)Propose a user preference movie recommendation model CBLAMF combining self-attention mechanism.The purpose of incorporating the multi-head self-intentional mechanism in the model is to perform parallel calculation of multiple scaling dot products,then splice together independent attention calculation units,and finally convert it into a dimensional output of the desired size through a linear unit.Therefore,the CBLAMF model can obtain the global dependence of input and output through the self-attention mechanism to learn the internal dependence of the input sequence,then input this dependency to each multi-head self-attention mechanism layer,and combine the information output last time to generate the next feature representation.This also allows our model to adaptively combine contextual information to obtain more accurate feature information.In addition,after calculating the feature output,matrix decomposition is used to calculate the feature score to provide recommendations for users,which also solves the problem of data sparsity..(3)Experimental results and analysis.Based on the evaluation criteria for the performance of the recommendation system,this article is based on the proposed method compared with other movie recommendation methods,and has better performance.