Scholarlink

<p>Aspect sentiment triplet extraction (ASTE) focuses on extracting triplets that include aspect terms, their linked sentiments, and the relevant opinion terms. Even though applications of linguistic knowledge have made good progress in ASTE, they generally lack sufficient mining of contextual cues which can lead to potential mismatch of word pairs. To overcome this challenge, we propose Aspect Sentiment Triplet Extraction via Integrating Contextual Semantic Relevance and Syntactic Relevance (CSSR), which seeks contextual cues associated with aspect and opinion terms through our designed Contextual Semantic Guidance Module. Subsequently, we introduce a Multilingual Relational Graph Attention Module to further explore the semantic and syntactic relevance between word pairs. In addition, we devise a relation-aware loss mechanism to strengthen the relational modeling capability of CSSR and employ a grid tagging scheme along with an effective inference strategy to extract triplets simultaneously. Extensive experiments on four benchmark datasets demonstrate the effectiveness and robustness of CSSR. Moreover, comparative experiments with the large language model Qwen2.5-14B-Instruct show that CSSR achieves superior performance while maintaining a favorable balance between inference efficiency and computational cost.</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 sentiment triplet extraction via integrating contextual semantic relevance and syntactic relevance

  • 期刊论文

  • 发表日期：

    2026年3月6日

查看全文
  • Xiaodong Zhu,
  • Long Jiang,
  • Ting Zhang,
  • Xiaoyao Liu

摘要

Aspect sentiment triplet extraction (ASTE) focuses on extracting triplets that include aspect terms, their linked sentiments, and the relevant opinion terms. Even though applications of linguistic knowledge have made good progress in ASTE, they generally lack sufficient mining of contextual cues which can lead to potential mismatch of word pairs. To overcome this challenge, we propose Aspect Sentiment Triplet Extraction via Integrating Contextual Semantic Relevance and Syntactic Relevance (CSSR), which seeks contextual cues associated with aspect and opinion terms through our designed Contextual Semantic Guidance Module. Subsequently, we introduce a Multilingual Relational Graph Attention Module to further explore the semantic and syntactic relevance between word pairs. In addition, we devise a relation-aware loss mechanism to strengthen the relational modeling capability of CSSR and employ a grid tagging scheme along with an effective inference strategy to extract triplets simultaneously. Extensive experiments on four benchmark datasets demonstrate the effectiveness and robustness of CSSR. Moreover, comparative experiments with the large language model Qwen2.5-14B-Instruct show that CSSR achieves superior performance while maintaining a favorable balance between inference efficiency and computational cost.