| Abstract: |
Worldwide online sales already surpassed US$1 trillion since 2012. Accurately locate the customers from the Internet is undoubtedly important for e-commerce. An effective way to do so is to analyze the information from microblogs, such as Weibo. Microbloggers with common interests usually form communities. By identifying the community interests, we can locate our customers more precisely. Unfortunately, existing approaches for community interests identification poses two problems: (1) Ignore the minorities; (2) Unable to cope with the changing interests of active members effectively. Consider a Hiking Community. Intuitively, KFC is irrelevant to hiking. We should not push KFC advertisements to this community. Yet, if a few members (i.e., minorities) always eat KFC after hiking, we may conclude that these minorities are interested in KFC and should in fact receive KFC news. Consider another community, a Game Community where the members suddenly interested in a new game for a short period of time. Existing techniques can hardly capture such dynamically happened events effectively as almost all approaches are offline basic. Unable to identify the new interests means missing some marketing opportunities. To overcome these problems, we propose INCOMIRS – a real-time system based on a discriminative undirected probabilistic graphical model. |
