Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
380462 | Engineering Applications of Artificial Intelligence | 2014 | 24 Pages |
The challenge of measuring semantic similarity between words is to find a method that can simulate the thinking process of human. The use of computers to quantify and compare semantic similarities has become an important area of research in various fields, including artificial intelligence, knowledge management, information retrieval and natural language processing. The development of efficient measures for the computation of concept similarity is fundamental for computational semantics. Several computational measures rely on knowledge resources to quantify semantic similarity, such as the WordNet « is a » taxonomy. Several of these measures are based on taxonomical parameters to achieve the best expression possible for the semantics of content. This paper presents a new measure for quantifying the degree of the semantic similarity between concepts and words based on the WordNet hierarchy and using a number of topological parameters related to the “is a” taxonomy. Our proposal combines, in a complementary way, the hyponyms and depth parameters. This measure takes the problem of fine granularity into account. It is argued, however, that WordNet sense distinctions are highly fine-grained even for humans. We, therefore, propose a new method to quantify the hyponyms subgraph of a given concept based on depth distribution. Common nouns datasets (RG65, MC30 and AG203), medical terms dataset (MED38) and verbs dataset (YP130) formed by word pairs are used in the assessment. We start by calculating semantic similarities and then compute the correlation coefficient between human judgement and computational measures. The results demonstrate that, compared to other currently available computational methods, the measure presented in this study yields into better levels of performance. Compared to several measures, it shows good accuracy covering all the pairwises of the verbs dataset YP130.