Detecting Implicit Expressions of Sentiment in Text Based on Commonsense Knowledge

Abstract

Sentiment analysis is one of the recent, highly dynamic fields in Natural Language Processing. Most existing approaches are based on word-level analysis of texts and are able to detect only explicit expressions of sentiment. In this paper, we present an approach towards automatically detecting emotions (as underlying components of sentiment) from contexts in which no clues of sentiment appear, based on commonsense knowledge. The resource we built towards this aim -- EmotiNet - is a knowledge base of concepts with associated affective value. Preliminary evaluations show that this approach is appropriate for the task of implicit emotion detection, thus improving the performance of sentiment detection and classification in text.