Research On Combine Deep Learning And Collaborative Filter Recommendation Algorithm

In the era of exponential growth of data scale,information overload has become an urgent problem to be solved.Recommendation algorithms can obtain the information that meets the needs of users from the mass data according to their interests.Among them,collaborative filtering is the most widely studied recommendation algorithm,However,it suffers from data sparsity and cold start problem.With the wide application of deep learning,combining deep learning and recommendation algorithm is a new research direction.To solve the problem of sparse data and cold start,this paper propose a recommendation algorithm combining deep learning and collaborative filtering.The main contents of this paper are as follows.Firstly Consindering the problem of cold start in the recommendation algorithms,this paper proposed a Top-N recommendation algorithm(abbreviation for DL-NAIS)which incorporates deep learning model into Neural Attentive Item Similarity algorithm.In the feature extraction,the algorithm use the deep learning model(Convolutional Neural Network,Multi Layer Perception and Connection Layer)to obtain the nonlinear feature vector of target item.At the same time,NAIS model use the historical interaction items of users to obtain the hidden feature vectors of users.Finally,we use the feature embedding vectors of target item and the user to predict the probability value of the user’s preference for the target item.Secondly,Aiming at the sparsity of rating data,combining the deep learning model with explicit feedback and implicit feedback,this paper proposed a rating prediction algorithm named MF-Neu Rec which combining matrix factorization and Neu Rec algorithm.We use the user-based and the item-based Neu Rec algorithm to obtain the feature vector of users and items by using the implicit feedback data respectively,and then combine the obtained user and item feature vector with the user and item feature vector which is obtained by using the explicit feedback data in the matrix factorization algorithm in a certain proportion.Finally,we use the feature vectors of item and user to predict the user’s rating of the item.Finally,Using the public dataset Movie Lens 1m to verify the DL-NAIS algorithm designed in this paper.We use HR and NDCG to evaluate the recommendation performance.Compared with the baseline methods,the evaluation metrics of the DL-NAIS algorithm has been greatly improved;Using the public dataset Movie Lens 1m and Movie Lens 100 k to verify the MF-Neu Rec algorithm designed in this paper.We use RMSE to evaluate the rating performance.The experimet results have proved that MF-Neu Rec algorithm can improve the accuracy of rating prediction.