Research On Recommendation Algorithm Based On Semi-supervised Learning

When computing users' similarity,the collaborative filtering recommendation method usually only considers a single type of similarities,and the computation complexity increases significantly as the number of users grow.Meanwhile,the influence of user preferences and user tag information on recommendation results has not been fully considered.To address these issues,a semi-supervised hybrid clustering-based collaborative filtering recommendation method is proposed in this paper,which adopts the semi-supervised clustering method to identify similar users more accurately so as to reduce computation time and improve the recommendation accuracy.The main contents are as follows:1.In order to effectively integrate of the labeled data and improve the accuracy of clustering,the artificial bee colony algorithm has been adopted to conduct semi-supervised clustering,and a semi-supervised artificial bee colony clustering algorithm named SSABC is presented in this paper.The weight parameter is set to balance the weight of the labeled data and the unlabeled data,and the target function is reconstructed to determine the clustering center point.The adaptive learning mechanism is used to assign values to the weight parameters,and combining with clustering algorithm,the semi-supervised hybrid clustering algorithm of parameter adaptive learning is proposed,which combines the determination of parameters with the clustering process to accelerate the clustering process.Finally,Through the UCI dataset and the MovieLens dataset,the clustering results of the SSABC algorithm and the APL-SSHC algorithm are analyzed,and then these two algorithms are compared with several existing methods.Accuracy,Recall and F-Score are regarded as the evaluation metrics on the experimental results.Experiments show that the clustering results of the APL-SSHC algorithm are better than other algorithms with higher Accuracy and F-Score.2.Based on the APL-SSHC algorithm,a semi-supervised hybrid clustering based collaborative filtering algorithm abbreviated as SSHC-CF is proposed.The algorithm firstly uses APL-SSHC to cluster the users,and select similar users of the end user from the cluster that the end user belongs to,which considers the user's multi-dimensional attributes,and confirms the similar users more accurately,resulting in overcoming the calculation problem of the single type of user similarity and shortening the computation time and improving the recommendation efficiency of the algorithm.3.Considering the effect of user preferences on the recommendation results,a users' preferences and semi-supervised hybrid clustering based collaborative filtering algorithm named UP-SSHC-CF is proposed.It applies the Latent Dirichlet Allocation(LDA)model to analyze the relationship among users,items and labels,it uses the APL-SSHC algorithm to figure out the similar users of the end user,and identify the items with greater preference probabilities of the similar users according to the user-item preference probability matrix,which will contribute to form the final recommendation list.On the basis of the MovieLens dataset,the SSHC-CF algorithm and UP-SSHC-CF algorithm are compared with other recommendation algorithms.The experimental results show that UP-SSHC-CF algorithm stands out with higher accuracy and F-measure.