Levels and Tasks of Sentiment Analysis

Abstract

Research in the domain of medical sentiment analysis is very active. For this reason, researchers propose, evaluate and compare different approaches, feature sets and tools, consider different levels of analysis and address a multitude of related tasks. The overarching aim of these efforts is to increase performance of sentiment analysis results and to find solutions to open challenges. Typically, sentiment analysis can be realised by using lexicon-based methods or machine learning or a combination of both (aka hybrid methods). Lexicon-based approaches make use of sentiment lexicons that link terms with their sentiment polarity often by a numerical score indicating the strength and polarity of the sentiment. However, sentiment lexicons do not contain domain specific terms or often do not consider sentiment depending on a specific domain or context a term is used. In contrast, machine learning based approaches require labelled data which is—when created by human annotators—a labour-intensive task. The purpose of this section is to provide an overview on the different levels of sentiment analysis approaches (document-, sentence-, aspect-level) and the various tasks that can be addressed (e.g. subjectivity analysis, polarity analysis, emotion analysis).