Add temporal information to dependency structure language model for topic detection and tracking

The dependency structure language model was proposed to overcome the limitation of unigram and bigram models in topic detection and tracking (TDT). But its structure is based on mathematical models, which may has problems to express information. In this paper a new approach of topic tracking of Chinese news articles is presented which improves the existing ones with temporal information. The technique is implemented in a framework of dependency structure language model (DSLM). The experiments show remarkable improvement to existing approaches.