Research On Personalized Recommendation Algorithm Based On Rough Set

With the development of the scientific and technology, the information on the Internet has exploded, then, people enter into an era with rich information and poor knowledge. How to find knowledge they need with the mass information is a serious problem. Thus, the recommendation algorithm emerged, so that users can then find the array of goods they interest in. Currently, the research on recommendation algorithm mainly focuses on collaborative filtering and content-based algorithm. Although the recommendation technique has developed very fast, there also exists the inevitable sparsely, cold start and low accuracy issues in the recommendation technique.This thesis proposed a new personalized recommendation algorithm based on rough sets, in order to solve the problem of sparsely and improve the accuracy. The recommendation algorithm can be divided into two steps. The first step is a solution for sparsely problem based on improved ROUSTIDA algorithm. The second step is the online score prediction based on rough set. In the process of Sparse processing in the first step, we designed the solution based on improved ROUSTIDA algorithms to process the sparse matrix, generating a dense matrix, thereby solving the problem of sparsely. In the process, firstly, the clustering method is used to reduce the data set to find the similar users to the target user and to find related candidate items. A new improved ROUSTIDA algorithm based on restrictions similar relationship is proposed to solve the drawbacks of traditional ROUSTIDA algorithm, and then the sparse matrix can be filled. The algorithm method is based on the expansion of basic rough set model to solve the problem of defining similar relationship simply and the complex structure of the original algorithm. In the second step, the recommendation process, the proposed method is implemented based on the dense matrix which has been treated. A new rating prediction based on rough set is designed. The process begins with the use of an improved approximate calculation method to determine the most similar Candidate items to the target item. After that is the rating prediction process based on rough set. The prediction method uses equivalence principle of division in rough set, combining item-based collaborative filtering. The user sets is divided by equivalence classes step by step. Finally, the lower approximation of the user sets can be found, and then a matching set of the final score can be found according to the value of the target user and user constraints. The score can be obtained by calculating mean of the matching sets and the items which have the higher scores will be recommended to the target user.The core issue of recommended problem studied in this thesis is the user-item relationship. From the perspective of mathematical, this thesis does the research on how to analyze the user-item relationship matrix efficiently combined with rough sets. An efficient and novel personalized recommendation algorithm is proposed in this thesis. Experimental results show the proposed algorithm can efficiently solve the problem of personalized recommendation.