A novel explicit context based approach for Computing Semantic Relatedness

Abstract

Computing semantic relatedness (SR) of words is a main functionality of large amounts of language applications. Explicit Semantic Analysis (ESA) is successful in computing semantic relatedness. ESA is an approach to explicitly represent the meaning of any text as a weighted vector of Wikipedia-based concepts and calculate semantic relatedness between terms or documents based on comparing the corresponding vectors. However, ESA method generates the same vector for an ambiguous word and does not consider the given context of the word-pair. In this paper, we propose an improved method that first specifies the given context of word-pairs by Wikipedia-based concept network named wikinet then represents any word as a weighted vector of Wikipedia-based concepts and compares vectors by cosine metric. Empirical Evaluation on WordSimilarity-353 dataset shows that the proposed method provides consistent improvements in correlation of computed relatedness scores with human judgments compared to the existing methods. Keywords: computing semantic relatedness, wikinet, Wikipedia, word sense disambiguation, explicit semantic analysis.