Research And Implementation Of Recommendation Algorithm Based On Collaborative Filtering

Facing the increasingly serious "information overload" problem,many studies have proposed to use information retrieval technology to solve,but the technology still has problems such as dependence on search keywords and the inability to provide personalized services.In order to solve the above problems,a recommendation system is proposed.The collaborative filtering algorithm is a commonly used recommendation algorithm in recommendation systems,but it still faces challenges such as data sparsity,cold start,scalability,and the limitations of the scoring data itself.Therefore,this paper takes the above problems as a starting point,combined with different analysis methods,merges multi-source information,and conducts research,improvement and practical application on the basis of two traditional collaborative filtering algorithms based on memory and matrix decomposition.The work content is as follows:The effect of Collaborative Filtering model based on Clustering and Bipartite Network(CBNRank)on alleviating the limitations of data sparseness,scalability,and scoring data in collaborative filtering algorithms is studied.First,use the combined clustering and collaborative filtering algorithm to process the scoring data,cluster the users,and use the user-based collaborative filtering algorithm to fill the scoring matrix in each cluster cluster to reduce the algorithm's computing time and data sparsity.Then,the scoring data is processed into a pair of preference data sets,and the corresponding bipartite network structure is constructed to alleviate the limitation of the scoring data itself.Finally,a graphbased ranking algorithm is used to calculate the similarity between items to implement Top N recommendation.Experimental results show that the CBNRank model plays a positive role in alleviating the above problems and promotes prediction accuracy.The effect of Collaborative Filtering model based on Multi-source Information and Deep Matrix Factorization(MIDMFRank)on mitigating data sparsity and cold start problems in collaborative filtering algorithms is studied.First,the user and project information are processed through different network layers to obtain the characteristics of the user and project.Then,use the scoring data as the input of the deep matrix decomposition model,learn the user and project hidden features separately,fuse the user and project features into the user and project hidden features,and continue to train the model through the fused hidden features to ease the cold start problem.Finally,continue to learn according to the normalized cross-entropy loss function,get the final score prediction matrix,and implement Top N recommendation.The experimental results show that the MIDMFRank model can promote the above problems and improve the accuracy of scoring prediction.A movie recommendation system based on MIDMFRank model is designed.Starting from requirements analysis,the system designed the overall framework,functional modules and database,and finally implemented the recommendation service,which verified the feasibility of the MIDMFRank model in practical application.