Scholarlink

<p>Text summarization is essential for extracting key information from large documents, especially in domains such as news, research, and patent analysis. Despite the potential of abstractive summarization, it is often hindered by issues such as factual inconsistencies, long-distance dependencies, hallucinations, and grammatical errors. To address these issues, the Convolutional BERT2BERT Attention Encoder-Decoder (CB2BAED) model is introduced as a novel abstractive text summarizer. The proposed model is optimized using Coyote and Badger Optimization (CBO) to enhance efficiency by fine-tuning hyperparameters for improved convergence and reduced computational overhead. It improves accuracy and performance while maintaining lower resource consumption. This model incorporates advanced text processing methods, including tokenization, lemmatization, stemming, stop-word removal, and SHapley Additive exPlanations (SHAP) based interpretability analysis. The proposed model uses a convolutional encoder, achieving 99% accuracy, 98% precision, 98% recall, and 97% F1-score on its token-importance classification task. The model achieves state-of-the-art ROUGE scores of 71.3% for ROUGE-1, 69.3% for ROUGE-2, and 71.3% for ROUGE-L, indicating significant improvements in summary quality. These results highlight the model’s significant advancement in abstractive summarization, surpassing current approaches in performance.</p>

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

SHAP with optimized convolutional BERT2BERT-attention encoder-decoder: a novel abstractive text summarizer

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Vinukonda Chandra Sekhar,
  • E. Sivasankar,
  • E. Muthu Kumaran

摘要

Text summarization is essential for extracting key information from large documents, especially in domains such as news, research, and patent analysis. Despite the potential of abstractive summarization, it is often hindered by issues such as factual inconsistencies, long-distance dependencies, hallucinations, and grammatical errors. To address these issues, the Convolutional BERT2BERT Attention Encoder-Decoder (CB2BAED) model is introduced as a novel abstractive text summarizer. The proposed model is optimized using Coyote and Badger Optimization (CBO) to enhance efficiency by fine-tuning hyperparameters for improved convergence and reduced computational overhead. It improves accuracy and performance while maintaining lower resource consumption. This model incorporates advanced text processing methods, including tokenization, lemmatization, stemming, stop-word removal, and SHapley Additive exPlanations (SHAP) based interpretability analysis. The proposed model uses a convolutional encoder, achieving 99% accuracy, 98% precision, 98% recall, and 97% F1-score on its token-importance classification task. The model achieves state-of-the-art ROUGE scores of 71.3% for ROUGE-1, 69.3% for ROUGE-2, and 71.3% for ROUGE-L, indicating significant improvements in summary quality. These results highlight the model’s significant advancement in abstractive summarization, surpassing current approaches in performance.