Knowledge-enhanced prototypical network with graph structure and semantic information interaction for low-shot joint spoken language understanding

Kunpeng Du, Xuan Zhang, Chen Gao, Weiyi Shang, Yubin Ma, Zhi Jin, LinYu Li

Abstract: Highlights•A joint modeling method for low-shot Spoken Language Understanding task.•Auxiliary knowledge is used to optimize prototypes and learn cross-domain features.•Labels prototypes interact via semantic space to enhance task-specific modeling.•Label interaction graph model dependencies between intent detection and slot filling.