A Collaborative Filtering Recommendation Algorithm For Library Circulation System

University library is the school documentation center, as well as one of the most important knowledge bases for teachers and students, which plays an important supporting role in the aspect of teaching and scientific research. Book circulation is the most fundamental but significant service content of the university library. However, the more collection library possess, the larger difficulty degree of seeking the certain related book will be. Along with the rapid development of computer technology and the continuous improvement of library automation system, it seems possible for the managers to have large amount of electronic book-borrowing information, which provides an original data for the library personalized service construction as well.How to make good use of the current data information to achieve better customer service and how to provide in-depth personalized services for different users are always issues worthy thinking about. Aiming at different borrowing habits, the individual recommendation service becomes one of the research hotspots in the field of digital library under new circumstances for its accurate and rapid conformity to users’preferences, and gains more and more attentions from the researchers.In this paper, on the basis of circulation records of Dalian University of Technology library, data was analyzed by data mining techniques. From records we can extract characteristic value of the user’s interest. Users clustering is made firstly to narrow down the scope of calculating the neighboring dataset. The categories of targeted users clustering are identified and the neighboring dataset is calculated according to Pearson Similarities. A certain amount of neighbors are selected and given weights to their similarities. Next, the degree of the interests shown by users to a certain type of books is identified, and a part of the borrowing ranking list is selected to make personalized recommendation. At last, a simplified experiment system based on the design mentioned above is realized to test and measure the experiment results. The findings include both the measurement indicators of recommended system and the acceptance of real users to this system. These analyses will be used to resolve the cold start problems among new users. In addition, a collaborative filtering recommendation algorithm based on cluster was put forward according to the characteristics of data, in order to realize the recommendation function and supply helpful reference to the construction of the recommendation system for library.