Scholarlink

Conventional stock price prediction models, predominantly reliant on historical univariate time-series data, often fail to capture the complex dynamics driven by exogenous factors such as market sentiment. To address this limitation, this paper introduces a novel hybrid deep learning framework, FinBERT-CNN-BiLSTM, designed to synergistically integrate quantitative price data with qualitative sentiment analysis. Our methodology first employs a Convolutional Neural Network (CNN) to extract salient local patterns from price series. Concurrently, the FinBERT model quantifies sentiment from financial news and social media. These heterogeneous data streams are then fused and fed into a Bidirectional Long Short-Term Memory (BiLSTM) network to model long-range temporal dependencies. We conducted a rigorous empirical validation using Apple Inc. (AAPL) FY2020 data. The proposed FinBERT-CNN-BiLSTM architecture demonstrated statistically significant outperformance against a spectrum of baseline deep learning models. Specifically, when augmented with sentiment features, the model achieved exceptional predictive accuracy. These results not only confirm the viability of our composite model but also provide compelling evidence for the indispensability of market sentiment as a predictive feature. This research contributes a robust and sophisticated framework for financial forecasting that effectively bridges the gap between quantitative analysis and unstructured textual data.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Composite Deep Learning Approach for Stock Price Prediction by Fusing Market Sentiment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lan Yangliu

摘要

Conventional stock price prediction models, predominantly reliant on historical univariate time-series data, often fail to capture the complex dynamics driven by exogenous factors such as market sentiment. To address this limitation, this paper introduces a novel hybrid deep learning framework, FinBERT-CNN-BiLSTM, designed to synergistically integrate quantitative price data with qualitative sentiment analysis. Our methodology first employs a Convolutional Neural Network (CNN) to extract salient local patterns from price series. Concurrently, the FinBERT model quantifies sentiment from financial news and social media. These heterogeneous data streams are then fused and fed into a Bidirectional Long Short-Term Memory (BiLSTM) network to model long-range temporal dependencies. We conducted a rigorous empirical validation using Apple Inc. (AAPL) FY2020 data. The proposed FinBERT-CNN-BiLSTM architecture demonstrated statistically significant outperformance against a spectrum of baseline deep learning models. Specifically, when augmented with sentiment features, the model achieved exceptional predictive accuracy. These results not only confirm the viability of our composite model but also provide compelling evidence for the indispensability of market sentiment as a predictive feature. This research contributes a robust and sophisticated framework for financial forecasting that effectively bridges the gap between quantitative analysis and unstructured textual data.