Recommendation Research Based On Latent Factors Prediction

With the continuous development of the network,the information explosion makes users face a more and more serious problem that they cannot select useful information efficiently.In order to avoid the problems caused by information overload,each major network site has launched their own recommendation system to provide personalized recommendation services for users,in order to improve users' network experience and increase user stickiness.The recommendation system analyzes users' historical behavior data,mining users;interests and other attributes,and gives appropriate recommendations.The recommendation algorithm is divided into three categories:content-based recommendation,collaborative filtering recommendation and hybrid recommendation.Matrix decomposition in collaborative filtering is the most common method,which has a high accuracy rate for scoring prediction.At present,most of the recommendation algorithms are based on the score prediction,which ignore the time factors.In real-world scenarios,user interests or popular trends change over time,but traditional recommendation algorithms do not capture these trends.Therefore,this paper proposes a recommendation algorithm based on hidden factor prediction,which regards user consumption behavior as a time series and models various changes of user interest in the recommendation scenario,so as to improve the recommendation effect.The proposed algorithm based on hidden factor prediction mainly includes two parts:matrix decomposition and recurrent neural network.In that field of recommendation system,matrix factorization is an effective method for extracting hidden factor of response items or user latent attributes,and the recurrent neural network has a strong ability of sequence modeling,the algorithm proposed in this paper combines the two well.In this paper,in the field of recommendation with rating feedback data,the matrix decomposition algorithm is used to decompose the scoring data of items to obtain the hidden factor vectors of users and items,and then the hidden factor vector sequences of consumer items are input into the training sequence model of the recurrent neural network.After the model is trained,the hidden vectors of the items to be consumed are predicted,and finally,a recommendation list is generated by the k nearest neighbor hidden vectors of the predicted hidden vector in the hidden factor space.Compared with the common recommendation algorithm,the proposed algorithm based on hidden factor has a good recommendation effect.