Affective Computing and Sentiment Analysis

Abstract

An introduction to affective computing and sentiment analysis. \rCommon tasks of AC and SA are emotion recognition and polarity detection.\rAgreement detection is, given a pair of inputs, deciding whether they receive same sentiment labels.\rBinary sentiment classification or assignment of degrees of polarity to the input.\rTopic identification and separate the opinions associated with them.\rMultimodal fusion - identifying the sentiment inherent in multiple modes of information such as video audio text etc. There are two types of fusion techniques- feature level and decision level. \r\rGeneral categorization of techniques :\rKnowledge-based, Statistical methods, and Hybrid approaches\r\rKnowledge-based - limitations\rValidity depends on the extensivity of resources. \rTypicality of knowledge representation\r\rStatistical methods\rSVM and deep learning\r\rlimitation of statistical methods\rsemantically weak\r\rHybrid approaches\rexamples\rsignificance