Intent Recognition and Slot Filling Joint Model for Question-Answering of Hazardous Chemical Knowledge

A question-answering system that provides accurate knowledge about hazardous chemicals is essential for their production and management. The system’s ability to accurately recognize intent and slot in the relevant information in the question sentence is critical to ensuring semantic accuracy in the answers. In response to a field-specific issue that slot information in hazardous chemical questions may impede accurate intent recognition, this paper proposes the CJIRSF-Bert model. This model leverages the location and intent information of slots to strengthen the relationship between key slots and intents. To further enhance intent recognition accuracy, we create a hazardous chemical question dataset through manual labeling. Experimental verification demonstrates that the CJIRSF-Bert model achieves a semantic frame accuracy rate of 91.6% on the dataset, surpassing the Joint-Bert model’s performance by 1.4%. This model effectively addresses the challenges of intent recognition and slot filling in hazardous chemical question-answering systems.