Prototypical Networks with Dual Attention and Regularization for Few-Shot Relation Classification
Bei Liu, Wan Tao, Sanming Liu · 2023
Relation extraction is a vital subtask of information extraction. It is also an important component for constructing the knowledge graphs. The purpose of relation extraction is to find out the semantic relationships between the entities from the natural language so as to discover the connections between them. Most existing models for relation extraction heavily rely on annotation data which is difficult to obtain. Few-shot learning can overcome the problem of data sparsity effectively. Prototypical networks are important models for dealing with few-shot learning problems because of their simplicity and efficiency. In this paper we proposed a novel model (PNAR) that integrates the dual attention and regularization into prototypical networks for few-shot relation extraction. The experimental results verify that our model outperforms the several baseline model.