SOMA: Semantic Orientation inference using Memetic Algorithm

Abstract

One of the substantial tasks of opinion mining is to find semantic orientation and intensity of opinion words and phrases. This research tries to find numeric values for sentiment words and phrases by introducing a novel algorithm. The opinion about an object or its aspects is often expressed through sentiment phrases, and having a measure of quantification for them is essential for processing sentiments. In simple terms, semantic orientation or opinion intensity is a building block of opinion mining. This paper tries to (i) identify the sentiment phrases, and (ii) by means of a memetic algorithm, assign scores to each phrase and build a sentiment lexicon. This score shows the place of the phrase on a spectrum, ranging from very negative to very positive (0 to 10). The proposed method assigns real numbers to sentiment phrases and these scores show the intensity of each sentiment phrase. Three datasets were created and used in this paper: Movie, Music and Camera datasets which consist of reviews in each category. The intensity and polarity of words are calculated for each database, and compared to each other. The results show that (i) some words are not as positive or negative as previously thought; (ii) what is the effect of using adverbs, such as “very” and “not”; and (iii) sentiment phrases in different contexts have different intensities.