个性化推荐系统中用户多态聚类研究

Research on Users' Polymorphic Clustering in Personality Recommendation System

针对传统协同过滤算法依赖单一用户需求形态影响推荐效果的问题,提出一种基于用户多态聚类的数字图书馆个性化推荐方法。该方法以改进的海明距离计算候选邻居集,结合多态相似度进行二次聚类,预测用户的多态需求度并形成推荐。实验表明,使用多态聚类产生的推荐精确度上优于单一聚类产生的推荐。

Traditional collaborative filtering algorithm is usually dependent on single kind of user requirement to generate clustering and this may affect the accuracy of recommendation.In view of the problem,this paper proposes a personalized recommendation method in digital library based on users＇ polymorphic clustering.This method uses an improved Hamming distance to calculate candidate neighbors,then combines polymorphic similarity to cluster again,finally forecasts user＇s requirements degree and generates recommendation.The experiments show that recommendation based on polymorphic clustering is more accurate than the single＇s.