Research On Cross-domain Recommendation Method Based On System Correlation

With the full coverage of the Internet,large amounts of data continue to generate and accumulate,which makes it difficult for them to get the information they need quickly and accurately.The recommendation system can filter out a large number of useless information to a certain extent,but the traditional single domain recommendation system has cold start and data sparsity,which results in low recommendation accuracy and poor pertinence,which affects user experience.As the social software of various functions emerge in endlessly,more and more people are involved in many fields of interest,leaving rich user behavior data.If these data can be integrated properly,the user model can be improved to achieve better recommendation results.The possible commercial potential of cross-Multi-domain recommendation,as well as the challenges of information integration and migration between domains,have driven the extensive and in-depth research by experts and scholars.Many experts and scholars have carried out extensive and in-depth research and application of cross-domain recommendation methods according to specific recommendation tasks.There are two typical recommendation scenarios:(1)there is shared user information among the systems;(2)there is neither shared user information nor shared project information among the systems,but there is some context based similarity and relevance.In view of the above two cross domain recommendation scenarios,scholars have proposed some solutions,but there are still the following shortcomings:(1)In view of the scene one,many systems share a large number of user information with users as a bridge,and how to extract more comprehensively from these information through effective methods.It is still to be further explored to integrate and integrate multiple user features to provide better personalized recommendation services for users.(2)In view of scene two,how to make full use of the similarity and relevance of the inter system context to build a relationship bridge between the system and the density of the more active system of the user.To improve personalized recommendation service quality of users' inactive system,we need to propose more effective methods.In view of the above problems,the following research works have been carried out in this paper.(1)Based on the user information shared by multiple systems,a cross-domain recommendation algorithm based on multi-layer network community detection is proposed.Because of the heterogeneity of multi-source user behavior data,and considering that user behavior is driven by its own interest,it is often promoted by social group behavior.This method extracts and unify multi user characteristics through machine learning and other methods,and uses a variety of common collaborative filtering ideas for reference to single domain recommendation,and uses a variety of methods.The user features construct the user similarity network respectively,thus the user interest community is found on the multi-layer network to implement the collaborative filtering recommendation,and the correlation weight between each layer is also calculated.(2)A cross-domain recommendation algorithm based on correlation between systems is proposed based on contextual information correlation among multiple systems.Inspired by context aware high-order tensor decomposition method,considering the similarity and relevance of contextual information between systems,Moving the potential factors from the dense data set to the target domain to implement the score prediction and project recommendation can make the inactive system use the data of the active system to improve its own recommendation effect.The algorithm also introduces a low rank approximation to truncate the tensor dimension,and effectively solves the problem of tensor dimension non alignment in the process of cross domain potential factor migration.The results show that the first method can fuse the multi user characteristics,find the community of interest effectively,and calculate the relationship weight between the multiple features.It is not only superior to the single domain recommendation algorithm in the recommendation accuracy,but also better than the multi-domain collaborative filtering in the community with the multi network weighted fusion as a single network.Recommended methods.The second methods in this paper can effectively learn the potential factors from the data intensive auxiliary domain by using the contextual correlation between the systems.It is more effective than the single domain recommendation algorithm in the target domain.