An Improved Word Similarity Measure For Ontological Context
Devendra Kumar Tayal, Amita Jain, Anindita Roy, Megha Gupta · 2019
For various Natural Language Processing (NLP) use cases, it is desirable to know the significance of the text. Various methods based on path, corpus and knowledge measures are used to find out the similarity among words. Different word similarity approaches are analysed in this paper. The widely accepted approach is Wu and Palmer similarity measure. But it has a major disadvantage. It produces less similarity score for those pair of words which are in same hierarchy in the ontology, whereas according to the contextual meaning, these words are more connected and so should have more similarity score. This paper lists the shortcomings of Wu and Palmer formula and presents a remodelled formula to improve the scores of such pair of words. The remodelled formula uses logarithm bringing the depth of the words in the ontology under a uniform scale. wup- Wu and Palmer, LCS- Least Common Subsumer(the most clearly identified concept or word which is an ancestor of two words in the ontology), Simremodelled- Remodelled Wu and Palmer formula.