Application Research Of Personalized Recommendation For University Library

University library,with the merit of vast volume,wide coverage and high-speed update,provides a wide stage for students to acquiring information and knowledge.However,along with these characters comes the problem of difficulty for students to find their desired information from gigantic repository quickly and accurately.This problem finally results in poor user experience and low book utilization.The practice of personalized recommendation technology in electronic commercial has provided abundant reference experience for us to apply personalized recommendation to university library management system.By understanding domestic and international researching situation of book recommendation,comparing advantages and disadvantages of different technologies,combining the actual situation of university library,this dissertation selects collaborative filtering as key technology for book recommendation of university library.It is supported by clustering technology which can alleviate the problem of sparsity,a serious problem in collaborative filtering system that affects recommendation quality.At the end of the dissertation,personalized recommendations are produced based on hybrid recommendation strategy.The main contents of the dissertation are as follows:(1)This dissertation first divides users who borrowed same category of book into same experimental group.Individuals of each experimental group are clustered according to their major backgrounds and borrowing behaviors.Clustering results,in the form of small group of users,will replace individual users to be involved in subsequent recommendation.With this approach,the similarity between new users' borrowing records are improved and data sparsity are reduced.(2)The dissertation then applies collaborative filtering to new users of each experimental group,and evaluates the results by analyzing whether new users are correlated with the results reasonably.(3)One individual might belongs to several clusters and each cluster has one recommendation result.The combination of these results is recommended to the individual as the personalized recommendation.