CodeIE: Large Code Generation Models are Better Few-Shot Information Extractors

37594

Knowledge Triples