Joint Intent Detection Model for Task-oriented Human-computer Dialogue System using Asynchronous Training

How to accurately understand low-resource languages is the core of task-oriented human-computer dialogue system. Language understanding consists two sub-tasks, i.e., intent detection and slot filling. Intent still faces challenges due semantic ambiguity implicit intentions with users’ input. Moreover, separately modeling filling significantly decrease correctness relevance between questions answers. To address these issues, we propose a joint method using asynchronous training strategy. The proposed firstly encodes local text information extracted by CNN relationship among words emphasized attention structure. Later, model strategy either fusing hidden states layers, or adopting key fine-tune whole network, greatly increasing subtasks. accuracy achieved tested on an open-source airline travel dataset self-collected electricity service dataset, ATIS ECSF, are 97.49% 89.68%, respectively, which proves effectiveness learning training.