Towards an annotation schema of financial discourse based on functional discourse analysis

Abstract

The aim of this presentation is to explore the structure of economic opinion news texts and its implications for sentiment analysis. The Lingmotif sentiment analysis tool has achieved great results in detecting polarity in texts but further improvements can be made by studying textual structure. The Functional Discourse Units model is used to identify discourse segments issued by the opinion holder for correct polarity assignment. The scheme for text segments consists of five layers: polarity, discursive functions, aspect, entity, and opinion holder. A corpus of economic opinion news columns from three major English-language newspapers is annotated to generate a comprehensive dataset for improving sentiment analysis and to provide insights into textual linguistics such as discourse markers and news values. Preliminary results offer numerous opportunities for improving sentiment analysis and provide a great panoramic view on textual linguistics.