Previous Automated Extraction Research
- Summary: A comprehensive survey of causality extraction studies including benchmark datasets, three major technical techniques for extraction and open challenges in the field.
- The studies described focus on broad representations (Wikipedia, WSJ) of causality in natural language (see Tables 1 & 2), such as:
- He derives great joy and happiness from cycling - Financial stress is one of the main causes of divorceThough two datasets do cover more specialized (but not statistical) biomedical relationships such as the adverse effects of drugs, or protein, gene, and RNA relationships
- Domain specific pre-trained language models combined with graph models are a promising approach to causality extraction.
- Extraction of drug-gene, drug-disease, and gene-disease named relationships from scientific text and assembly into a knowledge graph
- Work from the INDRA lab has focused on large scale extraction of biological causal mechanisms and more recently mechanisms governing processes such as agricultural production, food security and migration