Dialogue state tracking (DST) is a process to estimate the distribution of the dialogue states of each dialogue turn given the previous interaction history. It plays an important role in statistical dialogue management. The second Dialog State Tracking Challenge (DSTC2) provides a common test-bed for state tracking algorithms. For DSTC2, we proposed two statistical models and a rule-based model. The final submitted tracker (SJTU Tracker) is a combination of these models. The SJTU system significantly outperformed all the baselines and showed competitive performance in DSTC2.
The third DSTC (DSTC3) focuses on the challenge of extending the domain of a state tracker. The domain is the tourist information which is extended from the domain in DSTC2. For DSTC3, we proposed a novel framework to formulate rule-based models in a general way, which is called /Markov Bayesian Polynomial/(MBP) model. In the framework, a rule is considered as a special kind of polynomial function satisfying certain linear constraints. Under some particular definitions and assumptions, rule-based models can be seen as feasible solutions of an integer linear programming problem. Experiments showed that the proposed approach can not only achieve competitive performance compared to statistical approaches, but also have good generalization ability. It is one of the only two entries that outperformed all the four baselines in DSTC3.