Advancing domain-specific sentiment analysis, a framework for robust performance in financial contexts: financial sentiment analytics model (MarketSentix)

Sentiment analysis has gained significant popularity across various sectors in recent times. However, domain-specific sentiment analysis encounters unique challenges attributed to the target domain. Consequently, domain ontology and requirements play crucial role to perform sentiment analysis in a specific domain such as financial markets. Despite various studies as financial markets sentiment analysis in this domain, there are different issues in terms of domain ontology reflected as semantic and syntactic characteristics of the text that need to be addressed. Hence, the need for broad comparative research with the aim of developing a framework for an optimized method in this domain is essential. Towards this goal, in this study we develop a multi-aspect comparative framework, that aims to address limitations in the existing methods and enables us to develop an optimized method for sentiment analysis, which is tailored for this domain. By implementing the proposed method, we construct a domain-specific sentiment analysis model on a diverse corpus of financial markets that surpasses existing counterparts. The model's performance is rigorously evaluated by machine learning as well as statistical measures demonstrating its robustness and efficacy for sentiment analysis in the domain. Given the rise of generative pre-trained transformers, we also measure the models’ performance against GPT 3.5, which leads to promising results in terms of both accuracy and efficiency.