A Domain Independent Technique to generate Feature Opinion Pairs for Opinion Mining

摘要

The rapid growth of e-commerce and social media services in recent times have opened up many interesting Opinion mining research problems. Among others, Aspect/Feature based Opinion extraction have managed to grab researcher's attention. Towards this direction, many researchers have proposed techniques that are supervised and domain dependent for extracting opinions. In this paper, a novel technique that is unsupervised and domain independent is proposed for generating relevant Feature Opinion pairs with good accuracy. Technique employs grammatical relationships obtained by using typed dependency parsers to further refine the pairs extracted by Part of Speech taggers. It also focusses on words that are verbs and nouns which in some cases imply opinions, unlike other existing work which mainly focusses on adjectives and adverb expressions for Opinion analysis. The proposed technique was tested on 9 data sets of different domains. The result demonstrated a good percentage reduction in number of irrelevant Feature Opinion pairs and the relevancy of retained pairs was found to be considerably high.