Research And Implementation Of Cross Domain Recommendation Algorithm Based On Cross User

With the vigorous development of the Internet industry,users need to face a huge amount of information every day,how to select important information has become an important issue in the Internet industry.In order to solve the problem of information overload,people put forward a variety of solutions,one of the most successful and most widely used is the personalized recommendation system.The recommendation system predicts the user's preferences for the items that have not been viewed by analyzing the user history information and summarizing the user's interest characteristics.After 20 years of development,personalized recommendation system in academic reflect a lot of research results,are widely used in business.At present,when users visit the e-commerce website,video website,music website,and other multimedia sites such as Q & A website,they can get the relevant recommendation,eliminating the need for repeated search,screening boring work to improve efficiency while also improving the user experience.For the e-commerce site,the personalized recommendation allows users to more likely to buy related products,improve sales performance,for video and other content categories of Web sites,personalized recommendations can make users interested in browsing more content and then improve the user Of the residence time.So far,the research on the recommendation system has long been in the stage of improving the efficiency or accuracy of the proposed algorithm,while ignoring the essence that the user is a person,each person will generate data on multiple websites,thus Cross-domain recommendation system that can integrate multiple domain data has become a new research hotspot.The cross-domain recommendation system can improve the accuracy of the recommendation in the target domain and alleviate the sparse and cold start of the single domain recommendation system by migrating the data of the users on the source platform to the target domain.The implementation of the cross-domain recommendation system has a variety of technical means,including the use of matrix decomposition method to migrate the source-built model to the target domain,the use of clustering and artificial neural networks to migrate the knowledge of the source domain to the target domain.The above methods have achieved some results in specific areas.This thesis focuses on the cross users between different domains,that is,the same person for different users of different domains,the users of different domains seemingly unrelated,and the data accumulated in the original domain is used in the target domain for recommendation,and is intended to provide more accurate recommendations for cross users in sparse data.In this thesis,the cross-domain recommendation system has made the following achievements:1.The concept of cross-user,that is,the same person in two different sites need to register two users,this user is called the cross-user;cross-users in the two sites should have a similar performance,if you can identify all of the cross-users,the user registration of a new platform after the user will not be completely cold start,and his similar interest in other sites will help provide recommended items in the target site.2.The user similarity calculation formula of the single domain recommendation system uses only data of one domain.Different domain evaluation system may be very different,can not directly improve the original user similarity calculation formula-Pearson correlation coefficient,adding the user in the source field of data,more accurate.A cross-user recommendation algorithm is designed.3.In order to solve the problem of cross user identification,a cookie-based cross-user authentication system is constructed.After the cross user is identified,this thesis proposes a cross-domain recommendation system based on cross-users.Finally,the online data is used to verify the cross-domain recommendation system.The experimental results show that the proposed cross-domain recommendation system has a good effect in resolving the cold start of the user.The recommendation new users get when they first visit are targeted,reflecting the part of the user's interest.At the same time in the target domain data sparse case,it can also give more accurate recommendations.