Research On The Matrix Factorization Recommendation Algorithm Based On User And Item Co-occurrence Regularity

Big data technology has opened up a new wave of computer technology development,generating massive amounts of information every day,flooding the network,and data and users have begun exponential growth.In fact,this is a great development for the computer industry,but to a certain extent,it is not very friendly to the users served by the computer.Users may have a lot of other data when they get the information they want.Affected.Therefore,how to better recommend information has gradually become a hot research topic for computer scientists.For this reason,the recommendation algorithm is generated.The recommendation algorithm is one of the main methods for accurate recommendation of bulky data.The recommendation algorithm mainly uses the behavior information such as the user's relevant purchase consumption score to intelligently analyze the user behavior,predicts the user's interest,and recommends the content related to the interest of the user.The principal approach can make great use of effective data to generate maximum benefits and generate great value.With the rapid advance of computer technology and artificial intelligence,many computational scientists are working on recommending algorithms.Although many classic traditional recommendation algorithms have been commercialized on a large scale,they still face numerous unresolved theoretical and technical problems.For example,the more common problems are cold start problems,Dataset sparsely problems,and recommended accuracy problems,plus data scale issues and the current lack of processing power of computers.Aiming at some of the above problems,this paper proposes a collaborative filtering class recommendation algorithm: a matrix decomposition recommendation algorithm based on user co-occurrence and project co-occurrence regularity,hereinafter referred to as UIC-MF model(User and Item co-occurrence on Matrix Factorization).To some extent,the recommended effects and data sparseness are better handled.The work of this article is as follows:First,the user preferences are redefined,and the user threshold preference setting is made for each user in the recommendation system by using the concept of the average value,thereby obtaining the favorite item definition and the dislike item definition corresponding to each user.Secondly,being aimed at the problem of insufficient information mining in the matrix decomposition part algorithm model,two kinds of user co-occurrence matrix decomposition information are integrated: the common user/disliked user co-occurrence matrix decomposition information.Therefore,the UIC-MF model of this paper uses three observations to obtain better recommendations: user-item matrix,common like/disliked item co-occurrence matrix,common like/disliked user co-occurrence matrix,to increase Potential factors between users and users,between items and items.The article uses the word embedding technique and user preference definition to process the user-item scoring matrix to generate additional four matrices.Then the paper sets the loss function.On the basis of the traditional matrix decomposition,the decomposition of the four matrices is integrated to obtain a different loss function.The loss function is optimized by the ALS method to obtain the algorithm parameter update rules in the UIC-MF model.The parameters are iteratively updated until the model converges,and the matrix information of the user and the item is issued.Finally,the UIC-MF model proposed in this paper is validated on the dataset,and several experimental general indicators compare with other algorithms.The results show that the UICMF model proposed in this paper is more admirable to some extent.The theory of this article.