Granular Insights into Financial Discourse: Fine-Grained Opinion Analysis of Expert Texts

Jury

• Véronique MORICEAU, MdC IRIT – Université Paul Sabatier – Toulouse 3
• Serge DAROLLES, Prof. Université Paris Dauphine – PSL
• Philippe GILLET, MdC RITM – Faculté Jean Monnet – Université Paris-Saclay
• Elena CABRIO, Prof. Université Côte d’Azur, Inria, CNRS, I3S
• Bich-Liên DOAN Prof. CentraleSupelec, PhD supervisor
• Dirk SCHUMACHER Dr, Natixis, PhD supervisor
• Patrick PAROUBEK Dr/HDR, LISN, CNRS, Université Paris-Saclay, PhD supervisor

Abstract

This PhD research is about analyzing automatically texts written by financial professionals. As companies listed on stock markets have to communicate about their business activities every quarter, we leverage the recent progress of Artificial Intelligence to analyze the texts they publish. We want to know which topics business leaders are the most worried about and which topics they are optimistic / pessimistic about, referred to repectively as “aspect” and “polarity”. Currently, the most effective technology in Computation Linguistics to program computers to “understand” human languages are neural networks models, often called “black-box” algorithms. Thanks to examples labeled by humans and the effective design of algorithms, the computer learns how to extract the intended aspects and classify them into predefined polarities. The performance evaluation is enabled by reference annotated labels. My experiments show that information labeled by algorithms can also help to improve the accuracy in predicting the right opinions.