一种基于数据分布驱动的小样本无梯度学习的情感分类方法  

A Data Distribution-driven Sentiment Classification Method Based on Few-shot Gradient-less Learning

作　　者：李宛蓉 邢凯[2,3] LI Wanrong;XING Kai(School of Software Engineering,University of Science and Technology of China,Jiangsu 215123,China;School of Computer Science and Technology,University of Science and Technology of China,Anhui 230026,China;Suzhou Institute for Advanced Research,University of Science and Technology of China,Jiangsu 215123,China)

机构地区：[1]中国科学技术大学软件学院,江苏215123 [2]中国科学技术大学计算机科学与技术学院,安徽230026 [3]中国科学技术大学苏州高等研究院,江苏215123

出　　处：《电子技术（上海）》2024年第3期61-65,共5页Electronic Technology

基　　金：江苏重点研发项目(BE2020665)。

摘　　要：阐述一种基于数据分布驱动的小样本正向无梯度学习方法。该方法基于最优传输理论和格兰杰因果关系,在数据分布空间连续映射过程中,设计能稳定提取有效信息的特征通道,并运用资本资产定价理论,生成全局的高维最优收益结构表征,进而利用多头注意力机制生成情感分类模型。将该方法应用于情感分类问题,在该领域的通用英文数据集上,可将BERT模型的准确率提升7.07%,在中文数据集上可将其提升2.23%。将该方法应用于当前达到SOTA性能的情感分类模型LCF-ATEPC上,准确率平均提升0.6%左右。This paper describes a few-shot gradient-less learning method driven by data distribution, which designs feature channels that can stably extract effective information during the continuous mapping of data distribution space through optimal transmission theory and Granger causality, and applies the capital asset pricing theory to generate a global high-dimensional optimal revunue structure representations, finally generates a sentiment classification model by using Multihead Self-attention mechanism. We apply the method to the sentiment classification problem, and can improve the accuracy of the BERT model by 7.07% on the general English dataset and 2.23% on the Chinese dataset in this domain. Applying the method to LCF-ATEPC, an emotion classification model that achieves SOTA performance, the accuracy rate has increased by about 0.6% on average.

关 键 词：计算机技术 智能计算 无梯度学习 情感分类 注意力机制 资本资产定价理论 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]