The dialogue state tracking is an important component in the dialogue system, which interacts with external KB to retrieval informative results. Can we measure the similarity between user text and results of KB in the dialogue system to select related one ? or that does not make any sense?