Research Experience

  • (Ongoing) Low Rank Modeling of Signed Networks

    We explore the low rank structure of signed networks. By taking this point of view, many signed network analysis tasks, such as sign prediction or clustering, can be formulated as low-rank matrix completion problem. Solving the problem by matrix completion algorithms can improve the task results.

  • Link prediction on signed networks based on high order cycles, 2011

    We propose a cycle-based method to do sign prediction in signed social networks. The high order cycles reveal the balance quantity of networks, and also can be used as features in supervised learning methods.

  • Chinese common sense collection via game on social network, 2008 - 2009

    We design a community-based game and mechanisms to collect Chinese commonsenese with efficiency and sustainbility. Also, we analyze commonsense data, develop API and build analogy space for commonsense reasoning.