A framework for information synthesis into sentiment indicators using text mining methods. Issue 15 (30th June 2022)

• Record Type: Journal Article
• Authors: Casarin, Roberto; Camargo, Jorge E.; Molina, German; ter Horst, Enrique
• Is Part Of: Communications in statistics. Volume 51:Issue 15(2022)
• Page Start: 5265
• Page End: 5283
• Publication Date: 2022-06-30
• Subjects: Information synthesis -- text mining -- sentiment indicators; 62C10 (Bayesian problems; characterization of Bayes procedures); Mathematical statistics -- Periodicals; Mathematics; Statistics; 519.2
• DOI: 10.1080/03610926.2020.1837878
• Languages: English
• ISSNs: 0361-0926

Abstract

Abstract: A formal statistical framework is proposed for synthesis of text information into sentiment indicators. Each text document is treated as an exchangeable collection of stems of words (tokens), and used in conjunction with a multinomial inverse regression approach to efficiently synthesize the information content in text documents. The proposed methodology is illustrated through the buildup of sentiment indicators using Twitter news outlet text information. These synthesizing indicators, quantitative in nature, can be built across disciplines to capture changes in the economic, financial, and social conditions, and also serve to reveal heterogeneity across countries, sectors, or markets. The proposed approach is computationally fast and allows for time variation in the indexes.