Abstract

Current methods rely on inaccurate domain-specific dictionaries for the difficult task of emotion extraction from financial documents. This research offers a new perspective on Financial Sentiment Analysis (FSA) by integrating monetary and non-monetary metrics for success. Using grammar-based linguistic analysis, the proposed Sentiment Analysis Engine (SAE) advances sentiment analysis to the phrase level within each sentence. It employs a heuristic to extract aggregate attitudes from texts and proposes a hierarchical sentiment classifier based on association rule mining to predict whether financial messages are positive, neutral, or negative. The preprocessing module includes tokenization, stop-word removal, duplication removal, and part-of-speech (POS) tagging. The BERT Model, incorporating a convolutional neural network layer and n-Encoders, is used for classification. SAE outperforms traditional bag-of-words methods, suggesting a link between text emotions and mood time series. Financial sentiment analysis has advanced with this model's impressive accuracy.