计算机工程与应用 ›› 2024, Vol. 60 ›› Issue (16): 159-167.DOI: 10.3778/j.issn.1002-8331.2305-0139

• 模式识别与人工智能 • 上一篇    下一篇

融合多窗口特征的词对标记情感三元组抽取

林杰,刘建华,陈林颖,郑智雄,孙水华   

  1. 1.福建工程学院 计算机科学与数学学院,福州 350118
    2.福建省大数据挖掘与应用技术重点实验室,福州 350118
  • 出版日期:2024-08-15 发布日期:2024-08-15

Word-Pair Tagging Sentiment Triplet Extraction of Fusing Multi-Window Features

LIN Jie, LIU Jianhua, CHEN Linying, ZHENG Zhixiong, SUN Shuihua   

  1. 1.College of Information Science and Engineering, Fujian University of Technology, Fuzhou 350118, China
    2.Fujian Provincial Key Laboratory of Big Data Mining and Applications, Fuzhou 350118, China
  • Online:2024-08-15 Published:2024-08-15

摘要: 方面情感三元组抽取旨在从句子中抽取方面词、意见词和对应的情感极性。针对目前研究未充分挖掘局部上下文语义信息,缺乏对局部范围内的方面意见词对关联学习,以及遭受错误传播等问题,提出一种融合多窗口特征的词对标记情感三元组抽取模型。该模型利用BERT对句子信息进行处理,获取句子编码特征,采用多窗口特征学习机制学习局部范围内的情感特征关联,并挖掘句子包含的潜在语义信息,使用多头注意力图转换模块将所学习到的特征聚合成标记分布概率,利用改进的词对标记方案标记句子并解码得到三元组。在SemEval-ASTE的四个基准数据集上进行实验分析,相比GTS-BERT模型,所提模型在三元组抽取任务上F1分值分别提高了2.33、6.57、2.97、4.84个百分点。实验结果表明,所提模型可以有效学习局部语义信息,准确标记方面意见跨度,较为精确地提取情感三元组。

关键词: 方面情感三元组, 情感极性, 特征学习, 多头注意力, 词对标记方案

Abstract: Aspect sentiment triplet extraction aims to extract aspect words, opinion words and corresponding sentiment polarity from sentences. Aiming at the problems such as insufficient semantic information mining in local context, lack of word-pair association learning in local scope, and suffering from error propagation, a word-pair tagging sentiment triplet extraction of fusing multi-window features is proposed. In this model, BERT is used to process sentence information and obtain sentence coding features, and multi-window feature learning mechanism is used to learn local affective feature associations and dig the potential semantic information contained in sentences. Then the learned features are aggregated into label distribution probabilities by using multi-head attention graph conversion module. Finally, an improved word-pair tagging scheme is used to tag sentences and decode them to obtain triplet. The experimental analysis on four reference datasets of SemEval-ASTE shows that compared with the GTS-BERT model, the F1 scores of the proposed model for triplet extraction are improved by 2.33, 6.57, 2.97 and 4.84?percentage points, respectively. The experimental results show that the proposed model can effectively learn local semantic information, accurately label opinion span, and accurately extract sentiment triplet.

Key words: aspect sentiment triplet, sentiment polarity, feature learning, multi-head attention, word-pair tagging scheme