Financial sentiment analysis: techniques and applications

• Main Author: Du, Zidong
• Other Authors: Erik Cambria
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/182420
• Institution: Nanyang Technological University

Financial Sentiment Analysis (FSA) is an important domain application of sentiment analysis, with research categorized into technique-driven and application-driven tracks. The former explores human-annotated datasets to enhance FSA task performance, while the latter, gaining more attention recently, leverages FSA for financial market applications. This thesis conducted an in-depth review and introduced frameworks that clarify FSA's scope and its interplay with investor and market sentiments. It examined various FSA techniques ranging from lexicon-based, learning-based to Large Language Models (LLMs) and proposed FinSenticNet, a domain-specific lexicon that captures concept-level financial expressions, and the knowledge-enabled transformer models for Targeted Aspect-based FSA, integrating multiple lexical knowledge sources to refine model training. Further, it evaluated LLMs' reasoning capabilities in performing FSA. Lastly, it enhanced financial forecasting by proposing to incorporate company relationships through a dynamic dual-graph neural network, and presenting a novel explainable prediction using contrastive learning framework, showing superior efficacy in benchmark tests.