Research On Hybrid Collaborative Filtering Recommendation Algorithm In Data Sparse Environment

In today's society,the Internet has become the main source of people's access to information.Massive data information makes it difficult for people to quickly obtain information that meets their needs,but a recommendation system can help people achieve this goal.Collaborative filtering recommendation algorithm is the most mature and widely used algorithm among recommendation algorithms,but its data sparsity problem often restricts the quality of recommendation.In this thesis,the collaborative filtering recommendation algorithm under the sparse data environment is researched,and corresponding improved algorithms are proposed from different aspects to improve the accuracy of recommendation.This thesis first studies the accuracy of similarity calculation between users,and proposes a collaborative filtering recommendation algorithm PRE-CF based on rating information entropy.The algorithm combines the user rating information entropy with the improved Pearson similarity calculation method to form a new similarity calculation method.Experiments show that the algorithm can more accurately calculate the similarity between users and improve the recommendation accuracy.Then from the aspect of sparse data of the original scoring matrix,this thesis proposes a collaborative filtering recommendation algorithm SVDBCF based on matrix pre-filling.The algorithm first uses the slope one algorithm to perform score prediction and initial filling of the original user-item score matrix,and then uses the SVD algorithm to perform score prediction on the initially filled matrix,and fills the predicted scores into the initial score matrix,and then uses the similarity calculation formula of the fused Bhattacharyya coefficient to calculate the similarity of the filled score matrix.The comparative experiment proves that the algorithm improves the prediction scoring deviation caused by the sparse scoring matrix,and has a better recommendation effect than the traditional recommendation algorithm.Then from the perspective of hybrid algorithm and user data credibility,this thesis proposes a hybrid recommendation algorithm CPEBCTCF that incorporates trust factors.The algorithm first uses the slope one algorithm and the SVD algorithm to fill in the data of the scoring matrix,and then adds the trust factor between users to the similarity formula of the weight fusion to form a hybrid similarity calculation method for similarity calculation,and conducts comparison experiments on the data set.The results show that compared with the two algorithms mentioned before,the algorithm has higher recommendation accuracy in a data sparse environment.Finally,a simple movie recommendation system is designed based on the CPEBCTCF algorithm.This thesis discusses the functional requirements of the system,analyzes and designs the system architecture,functional modules and database,and shows the effect of each function of the system;thus verifies the CPEBCTCF algorithm has a certain practicability.