Biased Parameter Estimation in LDA

Latent Dirichlet allocation (LDA) and other related topic models are increasingly popular tools for summarization, manifold discovery and other application in discrete data. However, LDA alone does not perform well in IR application. We alleviate it by biased parameter estimation, which makes the topics in LDA more independent than standard LDA. We introduced the background for the biased estimation at first. Then we detailed method for the bias estimation within LDA frame. We reported the result of the biased model in ad hoc IR experiment showing that the biased estimation outperforms the basic EM method. Finally, the influence of some model parameters was analysed briefly.