Fine-Tuning Bert For Financial Sentiment Analysis

Abstract

Sentiment analysis is a prominent research area in Natural Language Processing. It involves deducing a statement's underlying meaning, context, purpose, and emotional tone using NLP techniques. Sentiment analysis has been especially desired in the finance industry to draw plausible conclusions from the huge array of daily financial news and make satisfactory predictions about the market. Several models such as Roberta, XLNet, and BERT have previously been used for this task, however, they require a significantly large training corpus and training time. In this paper, we propose a BERT model fine-tuned on financial text corpus significantly smaller than solutions proposed in previous literature and achieve accuracy on par with other models.