The inborn human instinct to know what others think has contributed to the growing popularity of Sentiment Analysis. Sentiment in a text is mostly derived from opinion oriented words or lexicons. But the challenge is put forward by opinion oriented words which are domain specific. Various researchers have proposed methods to improve the polarity scores of these domain specific lexicons. Existing works utilize mainly single domain knowledge which is not sufficient to update a domain-specific lexicon. The proposed work attempts a domain adaptation model by building a polarity lexicon using knowledge from multiple related domains. The polarity lexicon thus built when tested on new domains provides fairly good classification results thus implementing true domain adaptation. The proposed approach has been tested on Amazon product reviews from ten related domains which include Printer, Scanner, MP3 Player, iPod, LCD TV etc. A significant improvement in accuracy ranging from 1 to 14.5 points on learned domains and 0.5 to 8 points across new domains over the baseline has been observed.
Manju Venugopalan and Dr. Deepa Gupta, “An Enhanced Polarity Lexicon by Learning-based Method Using Related Domain Knowledge”, International Journal of Information Processing and Management, vol. 6, no. 2, pp. 61–72, 2015.