INTENTION DETECTION BASED ON BERT-BILSTM IN TASK-ORIENTED DIALOGUE SYSTEM

As an important module of intelligent dialogue system, intention detection has become an important research direction of man-machine dialogue at present. However, in the task-oriented dialogue, there are still problems that users' intention detection accuracy is low because of users' non-standard expression and excessive implied intentions. In order to solve the above problem, this paper proposes to use BERT launched by Google as a pre-training model, and use BiLSTM to build intention detection model of the task-oriented human-machine dialogue. With the Cambridge University Restaurant Reservation Corpus as the dataset, the accuracy of the intention detection model can reach 92.39% finally. Which provides a feasible solution for detecting users' intention in task-oriented man-machine dialogue system.