Abstract: In the human-machine dialogue system, in order to understand the user query, should do intent classification and named entity recognition (NER), and NER can be thought the fine-grained information based on user intent. The difficulty for NER task is that the task includes two parts: entity type extraction and entity boundary extraction, and both parts need to be correctly identified. But in practical application scenarios, due to lack of domain knowledge, our model always fail on understanding user query. Work on music, radio, video, poetry and other related fields, and propose a method which adds entity type and entity boundary to Bert model. Experimental results demonstrate that our proposed model achieves 3% and 5% improvement on intent classification and NER by adding domain knowledge. Domain knowledge can help the model grab information, and improve the accuracy of the model, especially when lack of contextual information.

Key words: Dialog system, Intent classification, NER, Artificial intelligence