Research On Hybrid Recommendation Algorithm Based On Matrix Decomposition And Clustering

In recent years,with the popularization and application of the Internet,the amount of information on the network shows a blowout growth,and recommendation system has become an important way to solve the problem of information overload.Recommendation algorithm is the core of recommendation system and determines the final recommendation effect.Matrix factorization based recommendation algorithm and memory based collaborative filtering recommendation algorithm as classical recommendation algorithms are widely used in various recommendation systems.However,the recommendation algorithm based on matrix factorization has some problems,such as single data source and data sparsity.The memorybased collaborative filtering recommendation algorithm has some problems,such as poor realtime performance,low accuracy of rating prediction and complete dependence of similarity calculation model on common rating users,which affect the recommendation effect.In order to solve the above problems,this paper makes the following research:(1)A collaborative filtering algorithm based on SVD++ and user clustering is proposed.Firstly,the user attribute information was introduced to increase the data source,and K-Means was used to cluster the vectorized user attributes to reduce the number of user similarity calculations,which solved the single data source problem of SVD++ algorithm based on matrix factorization model.Then,the time weight factor is added to the user similarity calculation model to obtain the similar neighboring users of the target user over time,which improves the real-time performance of the whole algorithm.Finally,the similar neighbor is applied to the bias adjustment term,and the bias adjustment term is used as the first prediction score generated by SVD++ algorithm to make the second prediction,and the final prediction result is obtained.Through experimental comparison,the prediction error of the proposed collaborative filtering algorithm based on SVD++ and user clustering is lower than that of the SVD++ algorithm and the user-based collaborative filtering algorithm,and it has high accuracy of rating prediction.(2)A collaborative filtering algorithm based on SVD and item clustering is proposed.Firstly,SVD matrix decomposition method was used to reduce the dimension of user item score matrix,extract item feature vectors,and K-Means clustering was used to cluster item feature vectors,which effectively solved the problem of data sparsity.Then,an improved project similarity calculation method based on KL divergence is proposed.This method calculates the similarity between projects according to the probability distribution of project scores,and at the same time,it adds a penalty factor for popular projects,which avoids the frequent recommendation of popular projects,and solves the problem that the project similarity calculation method relies too much on the common evaluation of users;Finally,the project similarity calculation method based on Pearson similarity and improved KL divergence similarity is combined to obtain the nearest neighbor projects and generate recommendations,taking into account the probability distribution of project rating and the influence of common evaluation users.Through experimental comparison,the collaborative filtering algorithm based on SVD and item clustering proposed in this paper solves the problem of data sparsity to a certain extent,and has high recommendation accuracy and effectiveness.