Clustered Hierarchical Concept Based Semantic Closeness Between Two Concepts Using WordNet
Boddu Bhaskara Rao, V. Valli Kumari · 2014
The search engine needs relatedness to measure closeness between two concepts for determining optimal results in major applications like information retrieval, information integration and of many more in natural language processing tasks i.e. text classification, word sense disambiguation, matching problems in artificial intelligence etc,. The clustered hierarchical concept network helps to overcome the fuzzy variations in different levels of granularity in measures of closeness based on weights, frequency or distances but these measures are not considered since no method takes the actual context of the user intention, user query or context domain subject fields. Clustered hierarchical concept network has three steps: Elicitation: extract the concepts of user query using concept extraction algorithm and name the output as context domain. Construction: building hierarchical clusters based on context or concept domain with re lated concepts as nodes and relations as edges. Matching: determine the matching concepts like Least Common General Concept (LCGC) and Least Common Specific Concept (LCSC). Clustered hierarchical concept based semantic closeness has three features i.e., context domain, concept net and common concepts. These features are used to calculate the relatedness. The primary goal of hierarchical concept network is to include the semantic of the concept by including its three features. The extraction of concepts are not only related to individual concepts, but it is also an organizational structure of the concepts that are combined in the ontology i.e. WorNet. In this paper, we propose a method for computing semantic closeness of two concepts in which the holonyms, meronyms, instances of concepts are considered synthetically. By calculating test data, the experiment results show that the method can compute concepts closeness effectively. The human judgments on a set of concept pairs led our approach to be more effective and have shown one of the best performance than the measures based on concept vector.