Research On Matrix Factorization Recommendation Algorithm Base On Deep Learning Methods

With the development of the mobile Internet and mobile phone,a variety of user-centered and content-centered web application services have rapidly developed.We have entered the era of information explosion.It is important to recommend interested information to the users from the massive information.Therefore,personalized recommendation algorithm has been widely concerned and studied by academia and industry.The recommendation algorithm based on collaborative filtering is the mainstream recommendation algorithm at present,but the problem of cold start and data sparseness restricts the performance of collaborative filtering algorithm.In addition to the rating matrix,if the auxiliary information of users and items can be effectively utilized in the algorithm model,the problem of data sparseness and cold start can be improved to a certain extent.We design and implement a matrix factorization hybrid collaborative filtering model based on deep autoencoder and Long-Short Term Memory(LSTM).It extracts the potential features of users and items from the auxiliary information of users and items through deep autoencoder and LSTM,and combines them with traditional matrix factorization collaborative filtering model to improve the performance of the recommendation algorithm and to some extent solves the problem of data sparseness and cold start.In order to verify the effectiveness of the algorithm,a contrast experiment is conducted in three real data sets of MovieLens with other algorithms.The data set is user-item rating data.The movie description is crawled as auxiliary information of items,and the user's statistical information is used as the auxiliary information of users.The root mean square error(RMSE)is used as evaluation index.The results show that our hybrid model is superior to other algorithms in the RMSE,which indicates that our model can effectively extract the potential features of users and items for rating prediction,and solves the problem of cold start and data sparseness to some extent.