摘要
针对传统微博文本情感分析方法不能充分提取文本特征的缺陷,提出一种将ALBERT与基于注意力机制的双向长短时记忆网络相融合的ALBERT-BiLSTM-ATT微博文本情感分析模型。模型首先使用ALBERT预训练生成文本特征表示,然后将其输入基于注意力机制的双向长短时记忆网络模型BiLSTM-ATT中,并对微博文本特征进行训练,最后经过SoftMax函数层实现微博文本情感分析。在已公开发布的微博文本数据集“微博2018”上,与其他7种微博情感分析模型进行情感分析对比实验,对比实验结果表明ALBERT-BiLSTM-ATT模型在对比实验中的精确率、召回率和F1值皆更高。
Aiming at the defect that traditional sentiment analysis method for micro-blog text cannot fully extract text features,a micro-blog text sentiment analysis model ALBERT-BiLSTM-ATT,which was based on the combination of ALBERT and the BiLSTM-ATT,was proposed.Firstly,the model uses the ALBERT pre-training to generate text feature representation.And then inputs it into the BiLSTM-ATT model to train the micro-blog text feature.Finally,the SoftMax layer is used to finish micro-blog text sentiment analysis.The comparison experiment was carried out with other seven micro-blog sentiment analysis models on the published micro-blog text data set“Weibo 2018”,the experimental results showed that the precision,recall and F1 value of ALBERT-BiLSTM-ATT model were high in the comparative test.
作者
徐洪学
汪安祺
车伟伟
杜英魁
孙万有
王阳阳
XU Hongxue;WANG Anqi;CHE Weiwei;DU Yingkui;SUN Wanyou;WANG Yangyang(School of Information Engineering,Shenyang University,Shenyang 110044,China;School of Automation,Qingdao University,Qingdao 266071,China)
出处
《沈阳大学学报（自然科学版）》
CAS
2022年第2期112-118,133,共8页
Journal of Shenyang University：Natural Science
基金
国家自然科学基金资助项目(61873338)。
