Coreference Resolution for Measuring Sentiment in Financial News
33 Pages Posted: 15 Jul 2022
Date Written: July 2, 2022
Abstract
We adopt a machine learning-based algorithm to identify direct and indirect company mentions in newspaper articles. We identify the importance of properly matching text segments to their respective companies when estimating human-perceived newspaper sentiment. Previous research has developed methods for measuring text sentiment and has revealed its relation to company returns. We further establish that knowing whether a text segment refers to a particular company is as important for estimating sentiment as the choice of the sentiment measure. Incorporating our methods into finance sentiment measures will significantly improve how accurately news sentiment both explains current and predicts future returns.
Keywords: Textual analysis, News media, Language, Bag-of-words, Coreference resolution, Sentiment analysis, Word lists
JEL Classification: G10, G12, G17, G41
Suggested Citation: Suggested Citation