| 摘 要: 方面情感三元组提取(Aspect Sentiment Triplet Extraction,ASTE)是方面级情感分析领域的一项关键任务,目的是提取出句子中给定的方面词、观点词及对应的情感极性。传统的ASTE方法因未充分考虑语义信息和语法结构之间的交互作用,导致模型性能受限。针对这一问题,文章提出了一种基于语义增强的双编码器三元组提取方法。首先,使用基于BERT(Bidirectional Encoder Representation from Transformers)的基本编码器提取单词的上下文信息。其次,基于GloVe词向量和Amazon特定评论词典,使用由门控循环单元(Gate Recurrent Unit,GRU)网络和图卷积网络(Graph Convolutional Network,GCN)结合构建的特定编码器提取深层语义信息,在此过程中通过引入图注意力机制实现基本语义特征与深层语义特征的交互融合。最后,使用边界驱动表填充(Boundary-Driven Table-Filling,BDTF)方法进行三元组提取。在4个公开的数据集上的实验验证表明,所提模型可以高效地捕获并利用句子的深层语义信息,实现较准确的方面情感三元组提取。 |
| 关键词: 情感分析;方面情感三元组提取;双编码器;图卷积网络;图注意力 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 国家自然科学基金项目(61806118);陕西科技大学博士科研启动基金项目(2020BJ-30) |
|
| Aspect Sentiment Triplet Extraction Based on Semantic Enhancement Dual Encoders |
|
GAO Yiyi, ZHANG Pengwei, CHEN Jingxia
|
(College of Electronic Inf ormation and Artif icial Intelligence, Shaanxi University of Science & Technology, X'i an 710021, China)
221612167@sust.edu.cn; zhangpengwei@sust.edu.cn; chenjingxia@sust.edu.cn
|
| Abstract: Aspect Sentiment Triplet Extraction (ASTE) is a key task in the field of aspect level sentiment analysis. The aim of ASTE is to extract the given aspect word, opinion word and the corresponding emotion polarity in a sentence. Traditional ASTE methods often ignore the interaction between semantic information and syntactic structure, resulting in limited model performance. To solve this problem, this paper proposes a method of sentiment triplet extraction based on semantic enhancement dual encoders. First, the basic BERT-based encoder is used to extract word context information. Secondly, based on GloVe word vectors and Amazon specific comment dictionaries, a specific encoder constructed by combining Gate Recurrent Unit (GRU) network and Graph Convolutional Network (GCN) is used to extract deep semantic information. In this process, graph attention mechanism is introduced to achieve the interactive fusion of basic and deep semantic features. Finally, the Boundary Driven Table Filling (BDTF) method is used for triplet extraction. Experimental verification on four publicly available datasets shows that the proposed model can efficiently capture and utilize deep semantic information of sentences, achieving accurate extraction of aspect sentiment triplets. |
| Keywords: sentiment analysis; ASTE; dual encoders; graph convolutional network; graph attention |
