A Survey of TensorFlow in Financial Sentiment Analysis

Financial sentiment analysis (FSA) is a core research area at the intersection of natural language processing and finance. This field aims to extract market signals from textual data in order to support investment decisions. This survey provides a comprehensive technical survey of the technical evolution of FSA and classifies existing methods into three main stages. These stages include early lexicon and rule based methods, Transformer based deep learning baselines, and recent Large Language Models techniques based on instruction tuning and preference alignment. This survey focuses on seven representative studies. These studies include FinBERT, FinBERT FOMC, FinGPT, Instruct FinGPT, FinLlama, InstructGPT, and Direct Preference Optimization. The survey also reviews their corresponding reference works. In addition, this survey critically analyzes the shift in evaluation methodology from traditional NLP metrics to financial utility metrics. Recent studies move away from traditional NLP metrics such as accuracy. They instead adopt financial utility metrics such as the Sharpe ratio and volatility. This change aims to better assess the practical investment value of sentiment models. The appendix proposes an end-to-end pipeline based on TensorFlow and Keras. The pipeline includes data processing, LoRA based fine tuning, and model deployment. This design aims to bridge the gap between theoretical research and practical quantitative trading systems.