Cross-domain Recommendation Research And Parallelization Implementation Based On Migration Learning

The convergence of information technology and economic society makes the amount of information data grow exponentially.The explosive growth of information has made the amount of data that individual users have come into contact with every day exceeds the range of capabilities that people can handle,causing information overload problems.The recommender system came into being at this background,allowing users to obtain what they are really interested in from massive information and provide solutions to the problem of “information overload”.At present,most of the research on the recommender system focuses on the single-domain environment.The data sources exist independently,and there are data sparseness problems and “cold start-up” problems,leading to a serious decline in the recommendation performance.Using the relevance of the auxiliary field and the target field in data,the effective information in the auxiliary field is migrated to the target field and crossdomain recommendation is made,which can greatly improve the accuracy and efficiency of the recommender system.CBT is a classic codebook transfer learning model.It assumes that all domains share a scoring model,and transfers the cluster-level scoring model from the auxiliary data to the target data.TRBT is based on the consideration of domain-specific patterns and it predict the missing ratings jointly by user-side and item-side triple-bridge transfer learning based on shared pattern,latent factors and adjacency graphs.However,this model has the problems of high computational complexity,low accuracy,but little improvement,and artificially determining coefficients when integrating user-end and project-end model scoring,which has the disadvantage of poor adaptability.This paper analyzes the advantages and disadvantages of TRBT-based multiple migration learning models,improves the existing models,and obtains a cross-domain recommendation algorithm based on the clustering implicit factor migration model.First,the data set is analyzed.In the case where users and projects overlap in each domain,the Canopy-Kmeans clustering algorithm is applied to users and projects in the corresponding auxiliary fields,and the obtained clustering information is used to extract the user feature matrix.Project eigenmatrix,user sharing grading model,project sharing grading model,and data migration learning at the user end and project end.Finally,the linear regression model is used to integrate the scores of the user end and the project end to obtain the final scoring prediction matrix.The algorithm creatively clusters users and projects,accurately obtains the number of users and project clusters,and improves the accuracy of cross-domain recommendation of the algorithm.In addition,the use of linear regression model dynamic fusion score overcomes the defect of poor adaptability.Through verification on the three public datasets MovieLens10 M,Amazon-meta,and Book-crossing,the crossdomain recommendation algorithm based on clustering implicit factor migration model has lower average absolute error and root mean square error,and the proposed algorithm's recommendation accuracy.Better than some of today's most advanced recommendation algorithms.Finally,this paper presents the application of the cross-domain recommendation algorithm based on the clustering implicit factor migration model on the Spark platform to achieve parallelization.It shows that parallelization can improve the efficiency of the algorithm operation and provide a valuable reference for the actual recommendation system development work.