摘要
为实现对即时通讯软件使用过程中话题分组信息的情感分析,采用基于BERT的预训练语言模型的微调方法,支持中文语义下的情感分析。在此基础上,提出一种跨领域模型迁移方法以进一步提高模型的准确率,实验结果表明,采用该方法相对于BERT基础方法平均准确率提升2.1%,对比RNN模型的各项指标也具有优势。该方法有效解决了即时通讯分组文本的情感分类问题。
Sentimental analysis of text is a challenging task of NLP.In order to realize sentimental analysis of important predictive information in the use process,this paper adopts a fine-tuning method based on the BERT language model to meet the sentimental analysis of Chinese demand.On this basis,we propose a cross-domain model migration method further improves the accuracy of the model.The experimental results show that the average accuracy of the method compared with the BERT basic method is increased by 2.1%.Compared with RNN,the indicators of the model also have advantages.This method effectively solves the emotional classification problem of instant messaging group text.
作者
吴迪
WU Di(College of Information and Electronics Engineering, Liming Vocational University, Quanzhou, Fujian 362000, China)
出处
《闽江学院学报》
2021年第5期42-49,共8页
Journal of Minjiang University
基金
黎明职业大学规划项目(KHL19014)。
