基于ERNIE-SKEP与BiGRU的APC情感分析

Sentiment Analysis of Airline Passenger Comments( APC)Based on ERNIE-SKEP and BiGRU

针对传统文本情感分类所获取词向量对语义理解不充分且依赖数据质量,以及主流的基于Transformer的预训练语言模型未考虑文本情感分类中情感知识等问题,提出ERNIE-SKEP-BiGRU航空公司旅客评价情感分类模型,基于持续学习语义理解框架ERNIE 2.0再次进行情感知识增强预训练(Sentiment Knowledge Enhanced Pre-training, SKEP),进一步引入双向门控循环单元(Bi-directional Gate Recurrent Unit, BiGRU)对情感特征进行深度提取。在航空公司旅客评价文本数据上进行多模型对比实验,并对送入模型的旅客评价数据进行emoji表情转义及简单数据增广以增强文本情感特征。结果表明提出的模型在二元和三元情感分类上准确率分别达到96.49%和87.12%,较传统模型有效的提高了航空公司旅客评价情感分类精度。

In response to the insufficient semantic understanding of word vectors obtained from traditional text sentiment classification and the dependence on data quality,as well as the mainstream Transformer based pre trained language models not considering emotional knowledge in text sentiment classification,an ERNIE-SKEP-BiGRU airline passenger evaluation sentiment classification model is proposed.Based on the continuous learning semantic understanding framework ERNIE 2.O,Sentiment Knowledge Enhanced Pre-training(SKEP)was conducted again,and Bi-directional Gate Recurrent Unit(BiGRU)was further introduces to deeply extract sentiment features.A multimodel comparison experiments were conducted on airline passenger comments text data,and the comments fed into the model were transcoded with emoji expressions and easy data augmentation to enhance the text sentiment features.The results show that the proposed model has an accuracy of 96.49%and 87.12%in binary and ternary sentiment classification,respectively,which effectively improves the accuracy of airline passenger evaluation sentiment classification compared with the traditional model.