SentiCircles for Contextual and Conceptual Semantic Sentiment Analysis of Twitter

摘要: Lexicon-based approaches to Twitter sentiment analysis are gaining much popularity due their simplicity, domain independence, and relatively good performance. These rely on lexicons, where a collection of words marked with fixed polarities. However, words’ orientation (positive, neural, negative) and/or strengths could change depending context targeted entities. In this paper we present SentiCircle; novel lexicon-based approach that takes into account the contextual conceptual semantics when calculating strength in Twitter. We evaluate our three datasets using different lexicons. Results show significantly outperforms two lexicon baselines. competitive but inconclusive comparing state-of-art SentiStrength, vary from one dataset another. SentiCircle SentiStrength accuracy average, falls marginally behind F-measure.