Abstract: Aiming at  the problem of few-shot relation extraction,  we proposed an induction network based on attention mechanism. Firstly, we used  dynamic routing algorithm in induction network to learn the class representation. Secondly, we proposed instance-level attention mechanism to  adjust support set and obtain high-level information between support set and query set samples, thereby obtaining  the support set samples that were more relevant to the query instances. The proposed  model effectively solved  the problem of how to extract relationships when the training data was insufficient. The experiment was conducted  on the few-shot relation extraction FewRel dataset, and the experimental results showed an  accuracy rate of (88.38±0.27)% in the 5-way 5-shot case,  (89.91±0.33)% in the 5-way 10-shot case, (77.92±0.44)% in the  10-way 5-shot case,  (81.21±0.39)% in the  10-way 10-shot case. The  experimental  results show that the model can adapt to tasks and outperforms other comparative  models, achieving better results than comparative  models in few-shot relation extraction.

Key words: relation extraction, few-shot learning, induction network, natural language processing, long short term memory network