结合句法结构和语义信息的方面情感三元组抽取
Aspect sentiment triple extraction combining syntactic structure and semantic information
石恽本 1苟刚1
作者信息
- 1. 贵州大学计算机科学与技术学院公共大数据国家重点实验室,贵州贵阳 550025
- 折叠
摘要
为解决先前方面情感三元组抽取方法中忽略句法结构和语义信息的问题,提出一种结合句法结构和语义信息的抽取模型.使用BERT预训练模型编码输入语句,同时编码句法结构特征.通过注意力层学习词对间的语义信息.将句法结构特征和语义信息输入图卷积网络,增强对单词间句法结构的学习.通过网格解码生成情感三元组.在lap14、res14、res15、res16数据集上的实验结果表明,该模型在精确率、召回率和F1值上相较其它基线模型有显著提升,有效提升方面情感三元组抽取效果.
Abstract
To address the issue of neglecting syntactic structure and semantic information in previous aspect-based sentiment trip-let extraction methods,an extraction model that combining syntactic structure and semantic information was proposed.The BERT pre-trained model was utilized to encode input sentences,and syntactic structure features were encoded.The semantic information between word pairs was learned through an attention layer.The syntactic structure features and semantic informa-tion were input into a graph convolutional network to enhance the learning of syntactic structure among words.Sentiment triplets were generated through grid decoding.Experimental results on the lap14,res14,res15,and res16 datasets demonstrate that the proposed model outperforms other baseline models in terms of precision,recall,and F1 score,effectively improving the effective-ness of aspect-based sentiment triplet extraction.
关键词
方面情感三元组/句法结构/语义信息/BERT预训练模型/注意力/图卷积网络/网格Key words
aspect-based sentiment triplet/syntactic structure/semantic information/BERT pre-trained model/attention/graph convolutional network/grid引用本文复制引用
基金项目
国家自然科学基金项目(62162010)
贵州省科技支撑计划基金项目(黔科合支撑[2022]一般267)
出版年
2024
NETL
NSTL
万方数据