The Research Of Personalized Recommendation Algorithm Based On Web Library Resources

With the continuous development of the web, the library resource on the internet is becoming richer and richer.The demand for the users who are interested in the book is very small. Also it is difficult to find books which they are interested from the ocean of the books. If some books are found by the user, they may be not necessary for users. This not only wastes the time, but also wastes the money if the books are needed to buy from the network.To solve the problems above, the research of personalized recommendation algorithm based on web library resources was born. Using the datas of the reader’s behavior, the algorithm realized the personalize recommendation for the reader through analyzing the similarity between users or items. The using of collaborative filtering algorithm is most successful, including user-based collaborative filtering algorithm and item-based collaborative filtering algorithm. But there are some problems existing in the traditional collaborative filtering algorithm. There are more and more readers and books in the book e-commerce site, so the input matrix of the algorithm named user-item rating matrix become more and more sparse. Also, there are some problems existing in the both algorithm.The traditional algorithms are not considered that the active users and the popular items make negative effects on them.The thises is carried on the detailed analisis on the data set. And according to the relationship between the active users and popular items, the traditional algorithms are improved by the author. The results show that the improvement based on the rigid punishment improves the ability to recommend new items and the improvement based on the soft punishment improves the accuracy recommendation and the ability to recomment new items. The thiese is comprehensive compared the two algorithm and analysized the advantages and weaknesses of the both. For the shortcomings of the algorithm, the thiese puts forward an algorithm based on the combination of the both. The algorithm not only considers the correlation of the users, but also the items. Also the sparse matrix is made to be corresponding compressed by the algorithm. The final recommendation result will be got with the advantages of the both. The results are get by the experiment based on the Bookcrossing dataset, which show that in the case of the matrix is sparse and the matrix scale is large, the algorithm improved the prediction precision more effectively, corresponding with the user-based and the item-based collaborative filtering algorithm.