Research On Collaborative Filtering Recommendation Algorithm Under CPS

Cyber Physical System(CPS)is an important part of Industry 4.0 which play an important role in future design and development.It realizes the combination and coordination of computing resources and physical resources by integrating computing,network and physical environment.However,while CPS brings convenience to people,it also brings information overload problem,processing huge amounts of data from multiple sources has become the top priority.Collaborative filtering recommendation algorithm is an effective solution to information overload.It processes existing user interaction data,extracts similarities,and finally gives users appropriate recommendations.However,in the current CPS environment,with the rapid increase in the number of users and the number of items,the defects of traditional collaborative filtering recommendation algorithms are gradually exposed,especially in the cold start of new users,new items and new systems,and sparse user behavior data.On the other hand,these problems lead to the degradation of collaborative filtering recommendation performance.In order to solve the poor recommendation effect caused by these problems,this research adopts the cross-domain collaborative filtering recommendation method.Through the improved cross-domain recommendation algorithm,the auxiliary domain data is used as a reference to extract the effective information in the auxiliary domain data and transfer the recommendation knowledge.To the data of the target domain,it can expand the target domain data through the effective recommendation information contained in other fields,so as to alleviate the data sparsity and cold start problems,resulting in a good recommendation effect.The main contents and contributions of this study are as follows:(1)By analyzing the traditional collaborative filtering algorithm and the cross-domain collaborative filtering recommendation algorithm,understand the ideas of cross-domain recommendation and traditional collaborative filtering recommendation,compare and analyze the similarities and differences between the algorithms,and clarify the design ideas.(2)Based on the Orthogonal Nonnegative Matrix Tri-factorization(ONMTF),a spectral clustering algorithm is proposed to improve the existing Code Book Transform Learning(CBT)algorithm to make it It has higher accuracy when extracting information from auxiliary domains,and improves the accuracy of cross-domain knowledge transfer.(3)On the basis of the improved CBT algorithm,the characteristics of spectral clustering are used to incorporate more information contained in the data,designed a weighted similarity method of time information into the user-side data processing,propose an improved time decay function,designed a weighted similarity of category information into item measurement.The method integrates multi-information into the process of cross-domain recommendation to improve the accuracy of recommendation.(4)Build an experimental environment,using Douban dataset and Movie Lens-Latest dataset to build experimental data for bidirectional cross-domain migration recommendation experiments,and compare and analyze with existing algorithms and traditional collaborative filtering algorithms.The results show that the improved cross-domain recommendation algorithm It can alleviate the data sparse problem and effectively improve the recommendation accuracy.