On The Study And Implementation Of Recommendation System For University Library

With the development of the network and the explosion of information, people usethe traditional information retrieval technology, in which to search information andselected in accordance with their interests is becoming more and more difficult.Collaborative filtering based recommendation technology, due to its many merits likecontent free, has been applied in many areas, and achieved great success ine-commerce. However, there is few cases once reported about the application of thistechnique on university libraries, and most published works are content-based orsome kind of hybrid recommendation algorithm. University library are known ofprofessional collection and readers, and publications on newly created or derivedfields emerge all alone. It is becoming more and more difficult for us to find suitablebooks, so recommendation system is highly demanded here.In order to help the teachers and students find the books that they are interestedmore conveniently and accurately, saving the time of retrieving books, we work tobuild a recommendation system for our university library using CF technique. Basedon five millions of usage records from Huaqiao University, we present qualitative aswell as quantitative analysis on the necessity of building recommendation system inuniversities. Different algorithms are compared, and user-based CF is preferred. Alsoit is noticed that models trained with partial data could produce similar accuracyperformance as models built on whole observations. In other words, we need onlyrefer to users of the same college (or department) to construct recommendation list.This discovery not only reflects the typical behavior characteristics of universitylibrary’s users, but also can be used to construct a more economic and flexiblesolution which (1) saves considerable computing resources,(2) allows for in-timeupdating, and (3) customizes updating solution for different college/department.Based on the real data, this discovered that building recommender model with localdata is not only efficient but effective, which is believed as valuable finding forindustrial and academic colleagues. At the same time, we use the local data to build the recommend system, and design a book recommendation system based on theuniversity library. The system not only realized the recommended function, but alsothrough the other auxiliary function to improve the cold-start problems and providesome basic services.