Data quality issues and duel purpose lexicon construction for mining emotions

Abstract

Mining emotions over the Internet has seen limited use even though it has important research implications in fields such as applied econometrics, the interdisciplinary study of happiness and well-being, and for various applications in customer relationship management, finance, marketing, human resources, and managerial science. A key ingredient to making progress in these areas is the development of an emotion specific lexicon, one that can capture intensity and select relevant sentiment laden texts from online sources. An approach to doing this is developed, issues relating to data quality are pointed out, and methods to overcome them are explained. Justifications for constructing the lexicon are given using state of the art empirical results and research. Then a 10 step algorithm that populates a lexicon using a hybrid procedure (thesaurus-corpus based) is developed. It captures sentiment no matter how it is expressed, and balances issues of speed, cost, and data quality. © 2009 Springer Berlin Heidelberg.