Research And Implementation Of A Personalized Reading Recommender System For Forum

As the Internet develops, we see the explosive growth of information. As a result, the information overload problem becomes obviously. Recommender sys-tem emerged in the 90’s of last century effectively help people extract useful part from the vast information. Nowadays we see recommender system everywhere, such as Amazon in E-commerce field, Netflix in movie review field and Douban in music field etc. By means of collecting data of interaction and analyzing these data, recommender system depicts user and item using various models for the sake of intelligent recommendation. However text related recommender system are seldom considered. Even those few like Readwise and Wumii are not specified for time-sensitive articles like news and posts in forum, meanwhile they don’t take other factors like user relationship and feedback into consideration. On the other hand, although modern social network like Facebook attracts many attentions, forum as a traditional media is still one of the most important place where people communicate. As a result, we propose a solution for recommendation for this kind of articles. Our main work and contribution are:1. A customized reading recommender system for time-sensitive articles, in-cluding content-based and social-based parts. We take into consideration various factors like topic modeling, similar article merging, modeling of user’s interest, users’ relation mining and score variation along with time etc. We synthesize all these perspectives to make a recommendation.2. A innovative model for depicting variation of article’s score along with time, which accurately describe article’s feature.3. A new algorithm for group recommendation. The algorithm syntheses pro-files of all members in the group to create a virtual user, which turn the group recommendation problem into the traditional one.4. We implemented a prototype specified for the Nanjing University LilyBBS. Meanwhile we implement a incremental collaborative filtering algorithm for parallel computing, which enhance the scalability of our system in case of the big data problem. On the other hand, we take advantage of advanced technique like HTML5 and Cache to improve the efficiency of the system.