Research On Dialogue State Tracking Technology In Dialogue

In task-oriented dialogue systems,dialogue state tracking(DST)is an indispensable part.The dialogue state task tracking model is designed to predict and judge the user’s intentions and goals in each round of dialogue in the process of task-based dialogue.However,most of the current methods in the process of dealing with scenes and semantic slots do not fully dig out the internal associations between different scenes and different semantic slots,and only by splicing and combining different conversation topics and semantic slots.Representation between them,so that the model is still insufficient in the modeling of dialogue state.In addition,the existing dialogue state tracking models mostly use shared model parameters to learn dialogue state tracking knowledge in different scenarios and different semantic slots.However,in general,the knowledge information in different scenarios is sometimes not exactly the same..Therefore,the existing dialog state tracking model is particularly rigid and inflexible in the face of complex cross-domain DST tasks.In addition,as the dialogue data set becomes larger and larger,the dialogue context content becomes more and more.This problem makes it necessary for the DST model to find practical ways to help the model filter out the long and meaningless information in the historical text,and better grasp it.The core key content in the dialogue text.Based on the above challenges,in the final work of this article,we will discuss three aspects:dialogue history context modeling(GAC),effective combination of external knowledge and model(KnowAdaptDST),and joint modeling of domain and semantic slot(CS-DST).From a different perspective,we will explore better and more suitable representation methods for task-based dialogues from different aspects.These models have obtained the current optimal JointGoalACC results and TurnACC results on the WOZ 2.0 and MultiWoZ 2.0 data sets,reflecting the optimization of dialogue history context modeling,the introduction of external knowledge and the domain-slot relationship joint modeling method for improving the effect of the DST system model has a certain gain effect.