Scholarlink

This paper presents a Bidirectional Long Short-Term Memory-based Twin Support Vector Machine (BiLSTM-TSVM) for efficient and robust classification of time-series and text data. BiLSTM captures bidirectional temporal dependencies for deep feature extraction, while TSVM constructs two nonparallel hyperplanes for effective classification. By integrating BiLSTM’s sequence modeling with TSVM’s simplicity and generalization, the proposed method effectively handles complex sequential data. Experiments on multiple benchmark datasets (IMDB, SST, UCI-HAR, TREC, AG News, and ECG) show that BiLSTM-TSVM consistently outperforms LSTM, BiLSTM, and LSTM-TSVM in accuracy, stability, and generalization, highlighting its effectiveness for diverse classification tasks.

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

Bidirectional Long Short-Term Memory-Based Twin Support Vector Machine

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jingyu Lin,
  • Zhenhua Yao,
  • Huajuan Huang

摘要

This paper presents a Bidirectional Long Short-Term Memory-based Twin Support Vector Machine (BiLSTM-TSVM) for efficient and robust classification of time-series and text data. BiLSTM captures bidirectional temporal dependencies for deep feature extraction, while TSVM constructs two nonparallel hyperplanes for effective classification. By integrating BiLSTM’s sequence modeling with TSVM’s simplicity and generalization, the proposed method effectively handles complex sequential data. Experiments on multiple benchmark datasets (IMDB, SST, UCI-HAR, TREC, AG News, and ECG) show that BiLSTM-TSVM consistently outperforms LSTM, BiLSTM, and LSTM-TSVM in accuracy, stability, and generalization, highlighting its effectiveness for diverse classification tasks.