用户集聚系数对协同过滤算法的影响研究

Effect of User Clustering Coefficient on Collaborative Filtering Recommender Systems

本文研究了用户-产品二部分网络中用户集聚系数对协同过滤算法的影响。用户集聚系数是度量目标用户的所有邻居用户的特点或者兴趣爱好相同程度的一个统计量,文章将其引入协同过滤算法的相似性计算中,并提出一种改进的算法。数值模拟显示,引入用户集聚系数统计属性的改进算法相比于CF准确性可以提高12.0%,当推荐列表的长度为50时推荐列表多样性可以达到0.649,相比于经典的CF算法提高18.2%。该工作表明用户集聚系数对推荐算法具有非常大的影响,体现了个性化推荐以用户兴趣的度量为核心的基本思想。

In the paper, the effect of user clustering coefficient in the user-object bipartite on collaborative filtering（CF） is considered. User clustering coefficient is a statistical property to evaluate the similarity of all the neighbor users＇ interests and characteristics of the target users, which is introduced to the similarity of CF to propose an improved algorithm. The numerical results indicate that, the improved algorithmic accuracy, measured by the average ranking score, can be improved by 12.0% , and correspondingly, the diversity could be improved by 18.2% and reach 0. 649 in the optimal case when the recommendation list equals to 50. Furthermore, this work highlights the effect of user clustering coefficient on CF recommender systems, which reflects the core of personalized recommendation is to estimate the users tastes more reasonably.