The Accuracy of Computer-Assisted Text Analysis Based on Standardized Sentiment Dictionaries
Irene Pollach, Lea Vindvad Hansen · Academy of Management Proceedings · 2019
Computer-assisted text analysis is often used for the measurement of constructs based on word frequencies in texts, typically measured with standardized collections of content dictionaries (e.g. LIWC, DICTION) developed for the analysis of language in general. The purpose of this paper is to compare the results of a standardized dictionary of positive and negative sentiment against the results obtained by a dictionary of positive and negative sentiment developed by the researchers for a business/management context. We apply these two dictionaries to a sample of 419 corporate documents and a sample of 2,939 news articles and qualitatively examine their suitability for this kind of data, based on high-frequency words. Our results indicate that the standardized dictionary, which was not built for a business context, does not interpret business terms correctly. Our own dictionary, meanwhile, contains high-frequency words with positive or negative sentiment that are not found in the standardized dictionary. Since the standardized dictionary produces inaccurate results in a business/management context, our results suggest that management scholars should not apply standardized dictionaries blindly but should at least examine high-frequency words for false positives.