Research On Recommendation Algorithm Based On Semi-supervised AP Clustering And Adaptive Transfer Clustering

In recent years,with the rapid development of the Internet and mobile Internet,the scale is increasing.Network application platforms such as JD.com and Iqiyi are springing up like bamboo shoots after a spring rain,and a variety of recommendation systems have been widely used.The collaborative filtering recommendation algorithm is one of the most mature and widely used recommendation algorithms in the recommendation system.However,with the increasingly complex data structure and the increase of data types,the traditional collaborative filtering recommendation algorithm have exposed some insurmountable problems,making its recommendation performance is difficult to meet the needs of merchants and users.In collaborative filtering recommendation algorithm,more and more clustering algorithms are used to pre-classify user sets and item sets,this collaborative filtering recommendation algorithm is limited by the precision of unsupervised clustering,so this study introduces semi-supervised AP clustering to improve clustering accuracy.Another bottleneck that constrains the recommendation performance of the collaborative filtering recommendation algorithm is the similarity measurement method,which is due to the complexity and diversity of the measurement objects.Therefore,this study considers improving the data sparsity by improving the similarity measurement method.In addition,the cross-domain collaborative filtering recommendation algorithm also has the problem of negative transfer and data sparsity.Therefore,this study introduces an adaptive migration clustering algorithm and a new type of cross-domain similarity matrix.Improvement of single-domain collaborative filtering algorithm.In the single-domain collaborative filtering recommendation algorithm based on clustering,the clustering accuracy of unsupervised clustering are difficult to meet the recommendation requirements,and because of the complexity and diversity of the measurement objects,the traditional similarity measurement method is difficult to accurately reflect the true similarity between objects,resulting in unsatisfactory recommendation results.Therefore,this study proposes a Collaborative Filtering algorithm combining Semi-supervised AP clustering and Improved Similarity Measurement.Firstly,the algorithm uses k-nearest neighbor and pairwise constraints into AP clustering to obtain Semi-supervised AP clustering based on K-nearest Neighbor Density estimation.Secondly,the similarity measurement was improved by integrating active user penalty factor and trajectory similarity into Pearson similarity.Then,the user preference matrix for the item category was then obtained using semi-supervised AP clustering and improved similarity measures.Finally,the loss of points was predicted by using the preference matrix.Improvement of cross-domain collaborative filtering algorithm.In the traditional crossdomain collaborative filtering recommendation algorithm,if there is a large difference between domains,there will often be negative transfer,resulting in undesirable recommendation results.Therefore,this study proposes a Collaborative Filtering algorithm combining Adaptive Transfer Clustering and Matrix Factorization.Firstly,a user-based cross-domain similarity matrix and a cluster-based cross-domain similarity matrix are constructed between the target domain and the auxiliary domain,and the two matrices are integrated into the Co-Clustering algorithm to obtain the adaptive transfer clustering algorithm.Secondly,the adaptive transfer clustering algorithm was used to solve the optimal cross-domain similarity matrix,and then calculate the transfer regularization term.Then,the target domain score matrix was decomposed by non-negative matrix factorization,and the transfer regularization term is used to improve the target user latent factor.Finally,the scoring matrix was reconstructed to achieve the goal of scoring prediction.To verify the validity of the proposed algorithm,experiments are carried out on six clustering datasets such as Vowel,Abalon,Yeast,and Covtype,as well as two recommendation algorithm datasets Movielens and Book-Crossing.The results show that the data sparsity problem has been improved to a certain extent,and the recommendation performance has been improved to a certain extent.