Research Of Collaborative Filtering Algorithm Based On Deep Neural Network Fusion

Nowadays,with the rapid development of society,the amount of data is increasing rapidly,and the rate of data generation is getting higher and higher.For the public,it is difficult to find and quickly filter out the information that meets their own needs in the massive data.Because of this problem,the recommendation algorithm emerges as the times require,and has made rapid development.Now it has also made some achievements in theory and application.A collaborative filtering algorithm is one of the main recommended algorithms,but the sparsity of data and the scalability of the method limit the performance.The rapid rise of artificial intelligence has been favored by a large number of researchers.The field of deep learning in artificial intelligence has become the first choice to solve problems.By studying the neural system of the human brain and evolving it into a deep neural network,it can effectively capture non-linear and non-trivial user/item relationships,and can code more complex abstract data into higher-level data representation.It has high performance for non-linear and complex high-dimensional data processing,and high performance for feature extraction and construction of complex data structure.The recommendation system based on deep learning,which overcomes the obstacles of the traditional model,obtains high recommendation quality,and has been widely concerned.To obtain the potential features and improve the recommendation performance accurately,users can understand the origin of the recommendation results and make more intelligent and accurate decisions.Firstly,a collaborative filtering algorithm based on deep neural network fusion(CF-DNNF)is proposed.CF-DNNF makes full use of the implicit attributes of data,among which text attributes are extracted through the LSTM network and other attributes are extracted through depth neural network to obtain the user and project feature matrix containing attribute information.The feature matrix is the initial input of the DBN depth confidence neural network,and the prediction score is obtained through the output probability of the DBN network.At the same time,the feature matrix and the There are comments as input.To verify the effectiveness of the algorithm,compared with PMF,SVD,RBM-CF algorithm in the Movie Lens dataset,and Amazon Product dataset,the best index RMSE value of the CF-DNNF is increased by 2.015%,and the MAE value is increased by 2.222%.The results show that CF-DNNF model can effectively improve the recommended performance.A large number of modern recommendation algorithms try to use potential features to represent users and items.The deep neural network is more like a black box,which may lead to the lack of transparency in the recommendation system,the result is not easy to predict,and will affect the further understanding and debugging of the algorithm model.When it comes to e-commerce sites,this kind of transparency problem may become very serious,because the potential functions are not easy to be marked out.To build trust between the recommendation system and its users,it is very important to supplement the recommendation with explanations,so that users can understand why they recommend a specific project and trust the recommendation results more.Based on the above analysis,this paper proposes an interpretable collaborative filtering algorithm based on deep neural network fusion based on CF-DNNF,ICFDNNF),after the feature matrix is obtained,the features and known comments are taken as the input of the Seq2 Seq model.Through the encoder and decoder of Seq2 Seq model,the comments corresponding to the prediction score are output as the explanation.The performance is verified on the Amazon Product data set.Compared with PMF,SVD,and RBM-CF algorithms,the tf-idf similarity index is the highest and superior to the three algorithms.