Automatic stuff relation extraction from scientific documents for natural product ontology construction

To extract Part-Whole relations, especially the stuff relation, from unstructured textual data is the challenging work This paper presents how to automatically extract the stuff relation from technical documents on the Web for supporting chemical industries. The research extracts the stuff relation without applying POS (Part-of-Speech) annotation. There are three problems of extracting the stuff relation: a) the identification of stuff relation without POS annotation problem, b) the chemical-formula-embedded name entity determination problem and c) the genus-species name entity determination problem. We propose using Naive Bayes to learn the stuff relation. The results from our proposed methodology are 87% precision and 61% recall.