Two Stage Semantic Relation Extraction

Ontology is applied to a real-world web intelligent question answering system in PC troubleshooting domain as an improvement, so semantic relations need to be extracted to construct ontology semi-automatically. The paper mainly pays attention to "product-trouble" relation and "product-attribute" relation. A two stage semantic relation extraction method is proposed. In stage one, relation identification is executed to produce high-precision relations instance as the seed for next stage; stage two is the actual relation extraction process, the related terms which describe troubleshooting information and attribute information are extracted based on association rules, then the terms is clustered to find target semantic relation. The relation extraction integrated the advantage of pattern-based and clustering-based methods. The experimental results show it is superior to individual pattern-based and clustering-based methods in precision and recall.