文章摘要
黄建民,李强,王雪绒,李聪聪.基于BERT融合多模块的方面级情感分析[J].井冈山大学自然版,2021,42(6):64-68
基于BERT融合多模块的方面级情感分析
ASPECT-LEVEL SENTIMENT ANALYSIS BASED ON BERT FUSION MULTI-MODULE
投稿时间:2021-09-15  修订日期:2021-10-16
DOI:10.3669/j.issn.1674-8085.2021.06.012
中文关键词: BERT  序列标注  微调  情感分析
英文关键词: BERT  sequence annotation  fine-tuning  emotion analysis
基金项目:甘肃省自然科学基金项目(17JR5RA177)
作者单位
黄建民 兰州财经大学信息工程学院, 甘肃, 兰州 730020 
李强 兰州财经大学信息工程学院, 甘肃, 兰州 730020
兰州财经大学电子商务综合重点实验室, 甘肃, 兰州 730020 
王雪绒 兰州财经大学信息工程学院, 甘肃, 兰州 730020 
李聪聪 兰州财经大学信息工程学院, 甘肃, 兰州 730020 
摘要点击次数: 1398
全文下载次数: 2415
中文摘要:
      现有的BERT模型大多关注初始层到中间层的语法信息,而更高层的语义信息往往被忽略。由于判断句子情感是需要语义的,本研究在BERT模型的基础上加入并行聚合和层次聚合两个模块,分别用于方面抽取(AspectExtraction, AE)和方面情感分类(Aspect Sentiment Classification, ASC)。同时选择条件随机场(Conditional Random Fields,CRF)作为序列标记任务,从而提取到更多的语义信息。在SemEval 2014、SemEval 2016数据集上的实验结果显示,微调后的模型准确率和F1值均优于对比模型,证实了该模型的有效性。
英文摘要:
      The existing BERT models mostly focus on the syntactic information from the initial layer to the middle layer, while the higher level semantic information is often ignored. Because of the semantics requirement in judging sentence emotion, two modules as parallel aggregation and hierarchical aggregation were added on the basis of BERT model, which were applied in Aspect Extraction (AE) and Aspect Sentiment Classification (ASC), respectively. At the same time, Conditional Random Fields (CRF) was selected as sequence marking task to extract more semantic information. Experimental results on SemEval 2014 and SemEval 2016 data sets showed that the accuracy and F1 values of the proposed model were better than those of the comparison model, which confirmed the validity of the proposed model.
查看全文   查看/发表评论  下载PDF阅读器
关闭