Research On Collaborative Filtering Recommendation Algorithm Based On Two-stage Joint Hash

In the Internet age,the rapid development of Internet technologies and e-commerce has generated massive amounts of information.It is becoming increasingly difficult for users,to search information that they are interested in from vast amounts of information.The recommendation system links the user with the information,and finds the items/services that are of interest to the user,it may even find novel interesting things to the user.However,the traditional collaborative filtering algorithms depend on the similarity calculation of users or items,it's limited when facing large volume of data in the system.A two-stage joint hashing collaborative filtering algorithm is proposed in this paper,it map the user and the item to the low-dimensional space by retaining the user's preference for the item,recommending is converted to searching in the low-dimensional space as a result,which eliminating the similarity calculations between users or items,and achieving efficient recommendation performance.Here are the main tasks of this thsis:First of all,by applying Principal Component Analysis techniques from user or item perspective to obtain low-dimensional feature representation of the perspective,the author applied Iterative Quantification techniques to generate binary codes for this perspective,which realized the extraction of the global features of rating data and prepared for the further generation of binary codes of another perspective.Secondly,basing on the binary code of one of the perspectives obtained in the previous process,the author used the existing rating information restrict the distance between the user and the item in the Hamming space,thereby generated the binary code of another perspective,which caught the local features of the rating data.The binary codes of users and items is used to efficiently recommend.Finally,basing on the proposed two-stage joint hash that effectively reduced the computational and storage consumption in the recommendation process,the author proposed a fast and efficient recommendation algorithm called collaborative filtering recommendation algorithm based on two-stage joint hash.The simulation results on the MovieLens-1m data sets shows that the proposed algorithm can significantly improve the quality and the efficiency of the recommendation.