Extracting ISA Relations of Concepts from Books viaWeakly Supervised Learning

Relation extraction is an important task in natural language processing. In recent years, researchers have gradually discovered that relation extraction can be used for teaching and learning assessment. Among various dependencies, the ISA is a relation that accounts for a relatively large proportion and is of great significance for many related tasks. However, there is currently little research on the ISA relation extraction between cross-chapter concepts for few-shot in textbooks. In this paper, we propose a distant supervised few-shot relation extraction model based on example construction and self-attention optimization. We use the example construction method in contrastive learning to expand the samples and self-attention mechanism to weight the sentences in the sentence bag to extract the relations of the concepts. Experimental results show that the model we constructed performs better than the previous methods.