Research on Co-Interactive Model Based on Knowledge Graph for Intent Detection and Slot Filling

Intent detection and slot filling tasks share common semantic features and are interdependent. The abundance of professional terminology in specific domains, which poses difficulties for entity recognition, subsequently impacts the performance of intent detection. To address this issue, this paper proposes a co-interactive model based on a knowledge graph (CIMKG) for intent detection and slot filling. The CIMKG model comprises three key components: (1) a knowledge graph-based shared encoder module that injects domain-specific expertise to enhance its semantic representation and solve the problem of entity recognition difficulties caused by professional terminology and then encodes short utterances; (2) a co-interactive module that explicitly establishes the relationship between intent detection and slot filling to address the inter-dependency of these processes; (3) two decoders that decode the intent detection and slot filling. The proposed CIMKG model has been validated using question–answer corpora from both the medical and architectural safety fields. The experimental results demonstrate that the proposed CIMKG model outperforms benchmark models.