An alternative technique for populating Thai tourism ontology from texts based on machine learning

This paper proposes an alternative technique to perform ontology population by using natural language processing and machine learning techniques. This study conceptually considers the population task as classifying terms into ontological subcategories. The proposed technique adopts the recognition method named Conditional Random Fields (CRFs) to identify boundary of instances and define types of subconcepts to generate relationships between instance-of and related concept. Also, the lexico-syntactic pattern is used to identify the relationships between instances. The experiments are conducted on Thai language documents in the tourism domain. The experimental results showed that the instances extraction step provided 77.62% and 70.87% of precision and recall measures, respectively, and relationships extraction step yielded 82.67% and 72.61% of recall measures.