Recommendation Model Based On Improved Matrix Decomposition And Cross-channel CNN

With the rapid development of information technology and the "5G era","information overload" has caused great trouble to people's lives.At the same time,with the rapid rise of social networks and e-commerce platforms,a large amount of user behavior data has been generated.In the face of such rich user data,many recommendation models only consider the user's rating data or project description content of the project and ignore it.A large number of users' text comments on the project.In addition,the recommendation system is also generally problematic in that the recommendation accuracy is not high and the recommendation model is not scalable.Through research and summary of a large number of literatures and materials,this paper proposes a recommendation model(MF & cross-channel CNN)that improves matrix decomposition and cross-channel CNN in view of the above problems in the current recommendation system.The model combines cross-channel CNN with model-based collaborative filtering recommendations,making full use of user ratings and text review information to obtain user preferences to improve model scalability and recommendation accuracy.Specifically,this paper mainly does the following research work:(1)From the perspective of the rapid development of the Internet and the "information overload",the research background and significance of the recommendation technology are analyzed.For the problems encountered in the current recommendation system,the traditional recommendation algorithm,the recommendation algorithm based on deep learning,the applicable field and The advantages and disadvantages of the research were studied,and the advantages and disadvantages of the recommended algorithms of the two categories were analyzed to prepare for the proposed model.(2)Based on matrix decomposition and deep learning techniques,the convolution matrix decomposition(ConvMF)model proposed by Kim et al.in 2016 is improved.First,the user and project paranoia are added on the basis of the probability matrix decomposition to improve the scalability of the model.Second,the cross-channel CNN is used to replace the traditional CNN.Third,the neural network is not to identify the project profile but to the user.Text comments perform feature learning to obtain user preferences for the project.Combine the improved matrix decomposition with the cross-channel CNN to make full use of user ratings and user review data to improve the accuracy of model recommendation.(3)Pre-processing the experimental data in this paper,constructing the user scoring matrix by user scoring,constructing the word vector of user comments by TF-IDF(Term Frequency-Inverse Document Frequency,TF-IDF)method,and performing superparameters in pre-processing and model Settings.Finally,the proposed model for the proposed improved matrix decomposition and cross-channel CNN is designed and analyzed from two aspects: MovieLens 1M,MovieLens 10 M and AIV.First,the influence of the selection of parameters on the model;second,the comparison between the MF&cross-channel CNN and the commonly used recommendation model prediction results.The experimental results show that: in the MF & cross-channel CNN model,when the user and project regularization coefficients are taken as 100 and 10 respectively,the matrix decomposition hidden factor number K is 200,and the convolution kernel number is 300,the overall prediction effect of the model is best.Compared with PMF,the MF & cross-channel CNN model improved the prediction results by 4.80%,12.03% and 2.34% respectively on the above three data sets;compared with CTR,it increased by 7.33%,10.14% and 6.62% respectively;compared with ConvMF,increased by 0.41%,2.84% and 0.66% respectively.As the training data density increases in the same dataset,the prediction effect of the MF&cross-channel CNN model is further improved.