Comparison between Knowledge Network System and Collaborative Filtering in Recommender System.
ICCCM(2020)
摘要
Recommender system is an effective tool to avoid amount of useless information given large-scale websites, which has achieved commercial success. Existing recommendation methods can be roughly divided into the content based filtering and collaborative filtering, and the latter one actually takes advantage of collective intelligence and is particularly popular because of its flexibility and independence of the item content description. Despite its popularity, there are at least two disadvantages with the collaborative filtering, respectively about the diversity of interest points and cold start. In this paper, a knowledge network system is introduced into the recommender system, which can be formed by collective intelligence in an evolutional way. Also using the collective intelligence, the proposed method is able to tackle or relieve the above disadvantages of the collaborative filtering. Specifically, it is able to recommend to users the closely related items based on an inherent interest structure and push items with diverse interest points. And by including a multi-round voting mechanism in the evolutional collective intelligence algorithm, the accumulation requirement of new records is reduced. Two more benefits with the proposed method are respectively the higher recommendation accuracy also based on the above mentioned multi-round voting mechanism, and the generalizable knowledge structure automatically extracted in the evolutional process. Explanative comparison with the collaborative filtering is made, and finally experimental comparison witnesses the efficacy of the proposed method.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络