Scholarlink

Predicting stock movements through market data analysis is a popular area of research. Integrating technical indicators from stock prices with insights from news articles for sequential analysis remains challenging despite widespread use in stock prediction research. This paper presents a Partial Contextual Embedding BERT-LSTM (PCE-BERT-LSTM) approach to stock index prediction by integrating Bidirectional Encoder Representations from Transformers (BERT) with Long Short-Term Memory (LSTM) through partial contextual embedding. Furthermore, the model incorporates an additional architecture to leverage news characteristics for a better understanding of the importance of each news article. Comprehensive experiments in different sectors show our approach, combining BERT’s language understanding with LSTM’s sequence modeling, significantly outperforms baseline methods. Evaluation metrics such as accuracy, weighted F1 score, and macro F1 score affirm the effectiveness of our model. The study contributes to the evolving landscape of financial prediction, highlighting the potential of embedding methods for capturing nuanced information from textual sources.

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

Stock Market Index Movement Prediction Using Partial Contextual Embedding BERT-LSTM

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Youjia Liu,
  • Yasumasa Matsuda,
  • Zhijie Zhang

摘要

Predicting stock movements through market data analysis is a popular area of research. Integrating technical indicators from stock prices with insights from news articles for sequential analysis remains challenging despite widespread use in stock prediction research. This paper presents a Partial Contextual Embedding BERT-LSTM (PCE-BERT-LSTM) approach to stock index prediction by integrating Bidirectional Encoder Representations from Transformers (BERT) with Long Short-Term Memory (LSTM) through partial contextual embedding. Furthermore, the model incorporates an additional architecture to leverage news characteristics for a better understanding of the importance of each news article. Comprehensive experiments in different sectors show our approach, combining BERT’s language understanding with LSTM’s sequence modeling, significantly outperforms baseline methods. Evaluation metrics such as accuracy, weighted F1 score, and macro F1 score affirm the effectiveness of our model. The study contributes to the evolving landscape of financial prediction, highlighting the potential of embedding methods for capturing nuanced information from textual sources.