Fig. 1: Schematic comparison of previous relation extraction (RE) methods to this work.
From: Structured information extraction from scientific text with large language models
The objective of each method is to extract entities (colored text) and their relationships from unstructured text. a An example multi-step pipeline approach first performs entity recognition, then intermediate processing such as coreference resolution, and finally classification of links between entities. b seq2seq approaches encode relationships as 2-tuples in the output sequence. Named entities and relationship links are tagged with special symbols (e.g., “@FORMULA@", “@N2F@"). c The method shown in this work outputs entities and their relationships as JSON documents or other hierarchical structures.
