Research And Design Of Personalized Recommendation System In University Library

With the rapid development of information technology and the Internet,it's difficult for college readers to find books that are of great interest or valuable from massive books based on traditional search-based services.Application of data mining technology and personalized recommendation technology,according to the reader's own information needs,will actively recommend books to readers.This active service approach not only improve the service level of university library,which makes a more comprehensive and humanized development for university library,but also can explore the potential information needs of readers to improve the library borrowing rate and maximize the usage of library resources.On the basis of analyzing various recommendation algorithms and the demand of personalized recommendation in university library,a personalized recommendation system for university library was designed in this paper based on the borrowing history accumulated in the Interlib library cluster automation management system of Wuhan University of Technology.The overall framework of the system,the work process was described in detail.The main work is as follows:1.The book feature model was established based on CLC(Chinese Library Classification).A measurement of book similarity was proposed based on the intercommunity and individuality of books in CLC tree.The reader preference model was built based on the book classification number,which is called Model-1.And Model-2 was built based on the tittle keywords.2.The selection of the initial centroid of k-means clustering stage based on the hybrid cluster algorithm of SOM and k-means was optimized.The comparison experiment between original clustering and optimized clustering was conducted based on Model-1.3.An improved association rule algorithm based on matrix-vector and full-join was proposed by analyzing association rule and Apriori.The experimental comparison was carried out on different transaction databases.4.The different recommendation strategies in the system framework were described in detail.For borrowing data of different academies,a recommendation list is generated based on borrowing records and association rules by applying improved association rules.In terms of the reader's preference feature,the readers are clustered based on Model-1.Then,combined with book feature model and Model-2,book recommendation list is generated by applying collaborative filtering recommendation and content-based recommendation.In terms of book that are newly imported into library,a recommendation list of new book is generated based on book feature model by applying content-based recommendation.In terms of new reader,the recommendation strategy is non-personalized recommendation,which is recommending via borrowing ranking list of academy that reader is associated with.5.A prototype system was built.The performance of prototype system was evaluated by using both prediction accuracy and user satisfaction.The experiments show the feasibility of this university library personalized recommendation system.