Computing text-to-text semantic relatedness based on building and analyzing enriched concept graph

Abstract

This paper discusses about effective usage of key concepts in computing texts semantic relatedness. Thus, we present a novel method for computing texts semantic relatedness by using key concepts. Problem of appropriate semantic resource selection is very important in Semantic Relatedness algorithms. For this purpose, we proposed to use a collection of two semantic resource namely, WordNet, Wikipedia, so that provide more complete data source and accuracy for calculate the semantic relatedness. Result of this proposal is compute semantic relatedness between almost any concepts pair. In purposed method, text is modeled as a graph of semantic relatedness between concepts of text that are exploited from WordNet and Wikipedia. This graph is named Enriched Concepts Graph. Then key concepts are extracted by analyzing ECG. Finally, texts semantic relatedness is obtained semantically by comparing key concepts of texts together. We evaluated our approach and obtained a high correlation coefficient of 0.782 which outperformed all other existing state of art approaches. © 2012 Springer Science+Business Media B.V.