User participation models in online forum and regularization methods for classifier combination

Speaker: Chun-hung Li (Assistant Professor at Hong Kong Baptist University)

In the first part of the talk, we will briefly go over our recent work on the modeling of user participation in online discussion forum. While topic detection and tracking based on document corpus has gained significant advances in recent years, the analysis of online discussion is often limited by its terseness. Using non-negative matrix factorization, user participation models can be developed to help us gain insights on the latent base topics of online discussions. The second part of the talk covers regularization methods for classifier combination. Majority voting in classifier combination treats all base classifiers equally without considering their performance differences. We will present a regularized classifier combination strategy that maximize the entropy of probability weights assigned to base classifiers subjected to the margin constraints of the ensemble classifier.