The prime textual sources used for ontology learning are a domain corpus and dynamic large text from web pages. The first source is limited and possibly outdated, while the second is uncertain. To overcome these shortcomings, a novel ontology learning methodology is proposed to utilize the different sources of text such as a corpus, web pages and the massive probabilistic knowledge base, Probase, for an effective automated construction of ontology. Specifically, to discover taxonomical relations among the concept of the ontology, a new web page based two-level semantic query formation methodology using the lexical syntactic patterns (LSP) and a novel scoring measure: Fitness built on Probase are proposed. Also, a syntactic and statistical measure called COS (Co-occurrence Strength) scoring, and Domain and Range-NTRD (Non-Taxonomical Relation Discovery) algorithms are proposed to accurately identify non-taxonomical relations(NTR) among concepts, using evidence from the corpus and web pages.
Taxonomy Extraction for Customer Service Knowledge Base Construction
