Labeled Link Analysis for Extracting User Characteristics in E-commerce Activity Network

This paper proposes an approach to characterize user activity in e-commerce, especially in Internet auctions. It is considered that users are connected to each other through their transactions. The connectivity makes an extensive complex network where users\´ intentions and behaviors are reflected in the network structure. That is to say, the network is composed of nodes as users and links as transactions between two users. Moreover, the links have one label that indicates a category type of transacted goods such as "book", "fashion", "music" among others. In order to characterize user activities in the network, a modified HITS algorithm is proposed. The results of the analysis using real data collected from an Internet auction site show characteristics of each user. Since the method takes advantage of network structures, evaluation values for the user characteristics that have been obtained illustrate not only the tendency of category types of goods that the users have transacted but also the tendency of relationships with other users. The characteristics are influenced by both direct and indirect connections among users. Our final goal is to construct a system that can personalize and profile users for marketing in e-commerce