Research on mixture language model-based document clustering

Language modeling with semantic smoothing is proposed as an effective way to improve the quality of document clustering. However, the existing semantic smoothing model is not effective for partitional clustering because it can not assign fit weight to "general" word in a collection. In this paper, inspired by mixture probability model, we put forward a mixture language model for document clustering. The new model can alleviate the effect of "general" word, simultaneously, it can integrate the context information and solve the polysemy problems in a document. Based the new model, an EM algorithm for partitional clustering is present. The experimental results show our algorithms are more effective than the previous methods to improve the cluster quality.