Financial Sentiment Analysis on Stock Using NLTK, Transformer, and Deep Learning

Financial sentiment analysis has emerged as a critical tool for predicting stock market movements, where traditional quantitative models often fail to capture the nuanced impact of news and social media sentiment. this paper presents a novel distilbert-gru hybrid architecture that synergizes transformer-based contextual understanding with sequential pattern recognition capabilities of gated recurrent units. our model processes financial news headlines through a twostage architecture: first extracting deep contextual embeddings using the distilled bert framework (6 transformer layers, 768-dimensional hidden states), then capturing temporal dependencies through a bidirectional gru layer (256 units with 0.2 recurrent dropout). the classification head employs a 128-unit dense layer with relu activation and 0.3 dropout before final softmax prediction. evaluated on a comprehensive dataset of 85,000 annotated financial headlines from reuters, bloomberg, and yahoo finance (2018-2023), our model achieves 93% accuracy in ternary sentiment classification (positive/neutral/negative), outperforming baseline models by 4-11 percentage points. comparative analysis reveals the hybrid architecture's particular strength in handling financial jargon and implicit sentiment cues, with precision-recall curves showing 0.94 auc for positive sentiment detection. the system demonstrates robust performance across market conditions, maintaining 91% accuracy during highvolatility periods. beyond accuracy metrics, we quantify the model's practical utility through a simulated trading experiment, where sentimentbased signals generated 18.7% annualized returns versus 12.4% for a baseline moving-average strategy. this research makes three key contributions: (1) demonstration of optimal transformer-rnn hybridization for financial nlp, (2) empirical validation of sentiment-return relationships across market capitalizations, and (3) open-source release of our annotated financial sentiment corpus. the work bridges critical gaps in real-time analysis capability and financial domain adaptation, while identifying promising directions for multimodal sentiment integration and explainable ai applications in algorithmic trading.