Recommendation System Based On Fusion Of Deep Learning And Matrix Factorization

For today's data explosion,it is difficult for people to choose their favorite goods or services through numerous suppliers.In order to overcome the phenomenon of information overload and help the users quickly get valuable information during the huge amount of information so as to save the users' time on browsing websites,selecting goods and services,the user experience,improve customer loyalty and change potential purchase into actual benefits.Based on this,recommender system is emerging and plays an increasingly vital role.In recent years,as one of the typical representatives of deep learning algorithm,convolutional neural network(CNN)model has been applied in various fields such as computer vision and natural language processing and achieved good results.In this paper,first we study a recommendation algorithm based on context,which improves the prediction accuracy through incorporating the text information processed by the CNN model into the probabilistic matrix decomposition model for the recommendation task.Then,we analyze the characteristics of the convolutional neural network model applied in the field of natural language.In order to further mine the information between project texts and improve the recommendation accuracy and performance of the model,an improved algorithm is proposed for improving convolutional neural network's partially embedded layer and convolutional layer.Experimental results show that the algorithm not only improves the accuracy of the model but also improves the system performance.In addition,the improved algorithm is in essence a probabilistic matrix decomposition model,although it solves the problem of partial sparse matrix compared with the traditional MF,it ignores problems such as user ratings and users'implicit historical feedback information.Based on this,this paper proposes a matrix decomposition recommendation algorithm blending together deep learning.In this algorithm,the convolutional neural network is also used to model the text information such as comments in the data set and associate it with the implicit factor part of items in SVD++model,meanwhile,it makes recommendations taking advantage of the user ratings and text information.The experimental results also prove that this algorithm makes full use of the implicit information of users and effective information of comments both in the field of film recommendation and e-commerce which improves the recommendation quality to some extent.