| 摘 要: 为了应对方面情感分析任务中标注数据稀缺的问题,提出了一种基于大语言模型数据增强与监督对比学习的联合训练方法。该方法首先利用大语言模型生成多样化的标注样本,以此扩展训练语料规模;随后通过构建正负样本对引入监督对比学习;最后结合多指令提示微调策略进一步提升方法的稳定性和泛化性。在两个常用的基准数据集上的实验结果表明,所提方法的总体准确率分别达到了90.98%和85.89%,优于其他基线模型,证实了该方案的有效性。 |
| 关键词: 方面情感分析 大语言模型 数据增强 监督对比学习 |
|
中图分类号: TP391.1
文献标识码: A
|
| 基金项目: 国家重点研发计划项目资助(2022YFB3105401) |
|
| Aspect-Based Sentiment Analysis Based on Large Language Model Data Augmentation and Supervised Contrastive Learning |
|
TIAN Shaocong, JI Zhongqing, ZHOU Renjie, ZHANG Wei
|
(School of Computer Science, Hangzhou Dianzi University, Hangzhou 310018, China)
231050075@hdu.edu.cn; 231050065@hdu.edu.cn; rjzhou@hdu.edu.cn; magherozhw@hdu.edu.cn
|
| Abstract: To address the issue of limited labeled data in aspec-t based sentiment analysis, this paper proposes a joint training method that integrates large language mode-l based data augmentation with supervised contrastive learning. The proposed method first leverages large language models to generate diverse labeled samples, thereby expanding the training corpus. It then introduces supervised contrastive learning by constructing positive and negative sample pairs. Finally, mult-i instruction prompt tuning is employed to further enhance the stability and generalization of the method. Experimental results on two commonly used benchmark datasets demonstrate accuracy rates of 90.98% and 85.89% , outperforming other baseline models and confirming the effectiveness of the proposed approach. |
| Keywords: aspec-t based sentiment analysis large language model data augmentation supervised contrastive learning |
