The Designation And Implementation Of Hybrid Recommendation System For Personalized Data

Personalization technology has been involved in all areas of life, including the university library. Library management system has some general problems, such as lack of user ratings for books and few feature of book. These problems make it difficult to run some recommendation algorithm in library scene. Currently, collaborative filtering recommendation algorithm is mostly used in library scene, and this way tends to recommend popular books. Generally speaking, this recommendation algorithm’s precision is acceptable and coverage is poor. For example, some books had not been borrowed will never be recommended. It is a waste of book resources. Currently, single recommendation has made some progress to improve the accuracy and coverage, but the speed of progress is very slow. Meanwhile the hybrid recommendation algorithm provides a new idea for solving this problem. Hybrid recommendation research now is very hot in recommendation area.In this paper, the source data is from Beijing University of Posts and library management system. First, recommendation target needs to be clear and the target is improving the recommendation accuracy and coverage. Then, according to the recommendation target and current scene, latent factor model and content-based recommendation algorithm are chosen to research the hybrid recommendation. About the lack of user ratings for books problem, this paper proposes an interest model for calculating interest level based on borrowing-times and time context using readers’historical data in library management system. About the few feature of book problem, this paper proposes a method to combine the characteristics of virtual books (social tagging of book) with physical books. But the social tagging itself has some problems, such as fuzzy tag problem and tag redundancy problem. For the fuzzy tag problem, this paper proposes a fuzzy tag recognition method based on information entropy thought. For the tag redundancy problem, this paper proposes a tag clustering method based on clustering algorithm idea to solve the redundancy problem.Finally, experiments are done to verify the hybrid recommendation strategy, interest model and tag processing algorithm. The experimental analysis shows hybrid strategy proposed in this paper has good precision and coverage and Content-based algorithm can optimize the precision of LFM. Meanwhile, the interest model has a good effect about simulating user’s interest, and the model has the versatility in library recommendation scene. About the tag processing algorithms, fuzzy tag identification method has good effect, but the tag redundancy issues need to be studied further. On the basis of solving the above problems, a book recommendation system has been implemented, which can be used in campus already.