MULTI-TEACHER KNOWLEDGE BASED QUESTION ANSWERING MODEL FOR DYNAMIC KNOWLEDGE BASE

Aiming at the lack of continuous learning ability of existing knowledge base question answering methods and difficulty on dynamic knowledge bases, we proposed a question answering model based on multi-teacher distillation and incremental learning dynamic knowledge base. The model could guarantee the memory ability of historical knowledge while learning new knowledge, and realized continuous learning. In addition, inspired by multi-teacher distillation, this paper designed a multi-teacher framework to further optimize the effect of knowledge distillation by using multiple different teacher models. Experimental results show that the model achieves accuracy rates of 91.02, 72.65, and 73.82 on three standard datasets.