Research On The Streaming Recommendation Algorithms Based On Context

Recommendation system is an effective way to deal with the problem of information overload,which is based on a large number of choices of user’s historical data and provides personalized suggestions for users.With the increasing popularity of recommendation systems,the timing,continuity,high-speed and time-varying characteristics of data in recommendation system make the data satisfy the streaming property,and pose a huge challenge to the traditional recommendation system,and the recommendation system based on streaming data can deal with streaming data more effectively.The stream recommendation algorithm based on matrix decomposition can not only solve the problem that the traditional matrix decomposition algorithm cannot deal with data sparsity,but also solve the recommendation problem with stream input data.Among them,Movie Recommendation via Markovian Factorization of Matrix Processes,which combines the matrix decomposition technology via Markovian,and effectively solves how to update the feature vectors of users and items in the system in real time.But MFMP still has the following problems:first,with the passage of time,MFMP cannot deal with the new users and new items in the recommendation system;second,the algorithm does not consider the real-time changing context environment of users.In view of the above problems in MFMP,the main work of this paper are as follows:(1)First of all,in view of the new users and new items in the recommendation system,this paper proposes a Markov Matrix Decomposition Recommendation Algorithm for New Users and New Items(iMFMP algorithm).iMFMP solves the problem of how to effectively obtain the hidden eigenvectors of new users and new items without any historical data.Finally,though two groups of comparative experiments based on the Movielens-1M data set,the performance of rate prediction and recommendation of the iMFMP is verified,and the effectiveness of the method of obtaining new users and new items hidden eigenvectors in the algorithm is proved.(2)In order to verify whether the context environment in which users rate items will affect their preferences,a Streaming Recommendation Algorithm based on Context(C-iMFMP algorithm)is proposed based on the iMFMP.In order to effectively select the context information related to rating from noisy context,a correlation analysis method of context information and rate based on information gain is proposed.On the basis of this method,the selected context information is divided into subjective context and objective context,and the two types of context information and algorithm are integrated in turn.Finally,two groups of experiments based on LDOS-CoMoDa data set show that the C-iMFMP is superior to the other comparison algorithms in both rate prediction performance and recommendation performance.