Uncovering heterogeneous interactions in online commercial networks.

With the rapid development of Internet, the research on online commercial networks has become crucial for filtering out irrelevant information for users and predicting their future interest. The common methods for understanding such typical user-item networks are mainly projecting them to unipartite ones with only positive ratings, which may result in losing a large amount of information. In this paper, we propose a novel approach to construct a signed unipartite network with heterogeneous interactions (i.e. positive or negative) between users from the original bipartite network. Based on the signed similarity, we carry out the percolation analysis on this signed unipartite network, which reveals a phase transition phenomenon. The statistical features of the giant component consisting of the positive and negative interactions are investigated respectively. Finally, the roles of the negative links and weak ties are revealed by adding them back to the giant component. This work not only deepens our understanding of the online commercial networks, but also has potential applications in the design of recommendation algorithms.