Research On Personalized Hybrid Recommendation Algorithm And Model Based On Collaborative Filtering

With the development of information science and technology,people have gradually step forward from the IT(Information Technology)era to the DT(Data Technology)era.In the time that full of massive information,how to validly alleviate the problem of information overload has become one of the significant researches.The traditional personalized recommendation system based on collaborative filtering can effectively solve the above problems.However,those collaborative filtering recommendation algorithms have the data sparsity and cold start problems.At the same time,it is difficult for a single recommendation algorithm to improve the recommendation accuracy effectively.In view of the above problems,the following work made by this thesis are as follows:Firstly,a novel personalized recommendation algorithm based on user's context has been proposed in this paper.through introducing the device log context information of users and items.this algorithm can extra the information features of users,items and contexts and dig the potential relationships among them.The simulation results show that compared with other mainstream recommendation algorithms,the data sparsity problem can alleviated effectively,the algorithm proposed in this dissertation achieves the optimal recommendation results.Secondly,a novel matrix decomposition personalized recommendation algorithm based on users and items side information has been proposed in this thesis.As the traditional matrix decomposition recommendation model does not take the effect of users and items side information into account.this paper integrates users and items side information into the matrix factorization model effectively.Based on those side information,matrix completion technology is combined to effectively complete the original rating matrix,and the least square method is used to solve the optimal solution of this model.The simulation results show that compared with other matrix decomposition recommendation algorithms,this matrix decomposition recommendation based on users and items side information can validly relieves the cold start problem and achieves better recommendation result and higher recommendation accuracy.Finally,a novel hybrid personalized recommendation model based on Learning To Rank has been proposed in this paper.Since the traditional single recommendation algorithm often has some shortcomings,for example,the content-based algorithm recommendation accuracy is not satisfied and the collaborative filtering recommendation algorithm has the cold start problem.Also,it is difficult to effectively merge different types of recommendation algorithms.Therefore,this paper proposes a hybrid personalized recommendation model based on learning to rank,which effectively fuses multiple recommendation algorithms with the help of pairwise approaches,and automatically allocates the optimum fusion weight between different algorithms.Simulation results show that the hybrid personalized recommendation model can effectively make up for the shortcomings of a single recommendation algorithm,solving the fusion problem of different kinds of recommendation algorithms,relieving overfitting in the fusion process and providing higher recommendation accuracy.