Determining the semantic orientation of opinion words using typed dependencies for opinion word senses and SentiWordNet scores from online product reviews. (13th November 2019)

• Record Type: Journal Article
• Authors: Kumar, K.C. Ravi; Santosh, D. Teja; Vardhan, B. Vishnu
• Is Part Of: International journal of knowledge and web intelligence. Volume 6:Number 2(2019)
• Page Start: 89
• Page End: 105
• Publication Date: 2019-11-13
• Subjects: e-commerce products -- product reviews -- opinion mining -- opinion word -- seed term -- typed dependencies -- contextual clues -- opinion word sense -- sentiwordnet score -- opinion word semantic orientation; Information retrieval -- Periodicals; Web site development -- Periodicals; Internet programming -- Periodicals; 004.67805
• Languages: English
• ISSNs: 1755-8255

Abstract

Opinion words express the information regarding the like and dislike of a user on the target entities such as products and product aspects present in the online reviews. The polarised information collected from the reviews is analysed by calculating the orientation of the adjectives. The synonymy relation graph is a way to determine the orientation of the adjectives present in the product reviews dataset. It considers the minimum path length between the adjectives under analysis using WordNet synsets. The synonymy relation graph cannot determine the orientations of all the opinion words present in the dataset. In order to evaluate opinion orientation of all the adjectives from the dataset, the synonymy relation graph of WordNet is to be replaced with the SentiWordNet scores of the opinion words. These scores are provided to the opinion words by finding the contextual clues surrounding the opinion words to disambiguate their sense. The contextual clues are finalised based on the typed dependencies grammatical relations. The distance between the opinion word and the context insensitive seed term (good/bad) is computed by calculating the difference between these scores. This paper addresses advantages of using SentiWordNet scores. This improves the accuracy of the determined opinion word orientations.