融合句法依存与BERT-Att-BiLSTM的属性分类  

Aspect Category Classification Integrated in Syntactic Dependency and

作　　者：包乾辉 文娟 石淑珍 董萌萍 刘雪[1] BAO Qianhui;WEN Juan;SHI Shuzhen;DONG Mengping;LIU Xue(College of Information and Electrical Engineering,China Agricultural University,Beijing 100083,China)

机构地区：[1]中国农业大学信息与电气工程学院,北京100083

出　　处：《北京邮电大学学报》2023年第4期123-128,共6页Journal of Beijing University of Posts and Telecommunications

基　　金：2021年现代农业产业技术体系北京市创新团队建设项目(BAIC04-2021)。

摘　　要：针对细粒度情感分析属性分类准确率低的问题,提出了一种融合句法依存关系和基于转换器的双向编码器-注意力机制-双向长短期记忆网络(BERT-Att-BiLSTM)的属性分类模型。该模型首先构建基于句法依存关系的目标信息提取层,进行属性-观点对提取;其次,在词嵌入层,使用BERT模块实现结合上下文动态特征的词向量预训练;然后,在特征提取层,融入Att的BiLSTM模块进行特征空间降维处理;最后,在分类层,通过激活函数输出属性-观点对的属性类别。实验结果表明,所提模型的精准度、召回率和F1值分别为85.25%、72.38%和77.06%,均优于其他模型,证明了所提模型的有效性。Current aspect category classification in fine-grained sentimental analysis suffers from low accuracy problems.An aspect category classification method is proposed integrates in syntactic dependency andbidirectional lencoderrepresentationsfrom Transformers-attention mechanism-bidirectional short-term memory network(BERT-Att-BiLSTM).Firstly,in the target information extraction layer,aspect-opinion pairs from comments is extracted by syntactic dependency.In the word embedding layer,BERT module is used to combined the dynamic features of context to achieve word vector representation.Then,in feature extraction layer,the BiLSTM module is integrated in attention mechanism to reduce the dimension of feature space.Finally,the aspect category is obtained at the classification layer through activation function.Experimental results show that the precision,recall rate,and F1 value of the proposed method reached 85.25%,72.38%and 77.06%,surpassed the other chosen models and proved its effectiveness.

关 键 词：属性抽取 句法依存关系 属性分类 基于转换器的双向编码器 注意力机制 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]