Sentic Parser: A Graph-Based Approach to Concept Extraction for Sentiment Analysis
Erik Cambria, Rui Mao, Sooji Han, Qian Liu
Abstract
Concept-level sentiment analysis improves on standard word-level opinion mining by leveraging the power of multiword expressions, linguistic objects formed by two or more words that behave like ‘semantic atoms’ by displaying formal or functional idiosyncratic properties with respect to free word combinations. The extraction of meaningful multiword expressions from text, however, is not an easy task, as it goes beyond simple n-gram modeling. In the context of sentiment analysis, such meaningful concepts are represented by those multiword expressions with high connotative, rather than denotative, information, i.e., combination of words that convey a certain degree of subjectivity (positive or negative polarity) rather than objectivity (neutral polarity). In this work, we propose a morphology-aware concept parser for the efficient extraction and generalization of affective multiword expressions from English text. The same methodology can potentially be applied to other knowledge bases, as well as different languages and multiple modalities.