Scholarlink

<p>Aspect-based sentiment analysis (ABSA) has emerged as a vital subfield in natural language processing (NLP), aiming to identify sentiment polarities associated with specific target expressions within a sentence. However, existing methods often focus solely on semantic content, neglecting the syntactic structure that underlies contextual dependencies. This paper proposes a novel structural-aware sentiment classification framework, SS-Transformer, specifically designed for ABSA in academic texts. By incorporating part-of-speech (POS) sequences and attention fusion mechanisms—namely post-attention, first-attention, and attention-over-attention—we construct a joint representation that encodes both semantic and syntactic features. Our SS-Transformer model is specifically tailored to improve the sentiment analysis of academic texts by considering their structural complexity. Experimental results on the SemEval-2014 dataset demonstrate that our SS-Transformer significantly outperforms traditional models including SVM, TD-LSTM, and Transformer baselines, achieving an accuracy of 80.86% and an F1 score of 0.7105 on the restaurant dataset. Interpretability analyses using SHAP values and t-SNE visualizations further highlight the effectiveness of incorporating structural information into ABSA models.</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

Aspect-based sentiment analysis (ABSA) for academic linguistics articles

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月11日

查看全文
  • Haiyun Wang

摘要

Aspect-based sentiment analysis (ABSA) has emerged as a vital subfield in natural language processing (NLP), aiming to identify sentiment polarities associated with specific target expressions within a sentence. However, existing methods often focus solely on semantic content, neglecting the syntactic structure that underlies contextual dependencies. This paper proposes a novel structural-aware sentiment classification framework, SS-Transformer, specifically designed for ABSA in academic texts. By incorporating part-of-speech (POS) sequences and attention fusion mechanisms—namely post-attention, first-attention, and attention-over-attention—we construct a joint representation that encodes both semantic and syntactic features. Our SS-Transformer model is specifically tailored to improve the sentiment analysis of academic texts by considering their structural complexity. Experimental results on the SemEval-2014 dataset demonstrate that our SS-Transformer significantly outperforms traditional models including SVM, TD-LSTM, and Transformer baselines, achieving an accuracy of 80.86% and an F1 score of 0.7105 on the restaurant dataset. Interpretability analyses using SHAP values and t-SNE visualizations further highlight the effectiveness of incorporating structural information into ABSA models.