Domain Specific Ontology Extractor For Indian Languages

We present a k-partite graph learning algorithm for ontology extraction from unstructured text. The algorithm divides the initial set of terms into different partitions based on information content of the terms and then constructs ontology by detecting subsumption relation between terms in different partitions. This approach not only reduces the amount of computation required for ontology construction but also provides an additional level of term filtering. The experiments are conducted for Hindi and English and the performance is evaluated by comparing resulting ontology with manually constructed ontology for Health domain. We observe that our approach significantly improves the precision. The proposed approach does not require sophisticated NLP tools such as NER and parser and can be easily adopted for any language.