Case-Based Reasoning for Personalized Recommender on User Preference through Dynamic Clustering

Abstract

The existence of user data sparsity is inevitable for that the user interacts only with items of interest in the recommender system, and the abundance of cyber resources increases the scale of data makes the system data more sparse eventually, they seriously affect recommender system performance. Many researches have highlighted such problems for applying supplements to those vacancies explicitly or implicitly, to balance computing complexity and the system efficiency. CBR-recommender is proposed to learn user preference from what user really interacts to keep the integrity and virginity of such personalized information, accompanied with Covering algorithm to partition the features of rare items into some specific domains, implement high personalized requirements. Our experiments results indicates that the new system has shorter running time in the research of large-scale recommendation for computing user preference dynamically, and can perform better results to meet the individual needs with high user satisfaction.