基于声誉与基于网络结构的用户聚类比较分析

用户聚类问题是在线用户行为分析的一个重要研究方向。基于在线评分系统,用户声誉反映的是用户对产品打分的准确程度,用户——产品二部分网络结构反映的是用户对产品的品味偏好。结合声誉度量算法,分别采用DBSCAN方法和基于模块度的贪婪算法从用户打分准确程度和品味偏好角度进行用户聚类,提出一种一致性度量指标来衡量根据用户声誉与根据网络结构得到的两种聚类结果之间的联系。两个实证数据集上的实验结果表明根据用户声誉与根据网络结构得到的两种聚类结果是不一致的,说明打分准确程度相似的用户的品味偏好并不相似。

Comparison of User Clustering Based on User Reputation and Network Structure

User clustering is an important research hotspot in online user behavior analysis.User reputation reflects the accuracy of ratings which users give to objects for online rating systems,while network structure of user-object bipartite networks reflects the user preference to the objects.Combined with the existing several reputation measurement algorithms,users were clustered with DBSCAN method and the greedy algorithm based on the modular degree from the user rating accuracy and taste preferences,respectively.A consistency metric was proposed to measure the association between two clustering results obtained based on the user reputation and the network structure.The experimental results based on the two empirical datasets show that the two clustering results based on the user reputation and the network structure are inconsistent,indicating that the taste preferences of users with similar rating accuracy are not similar.