摘要
在大数据时代,信息包含大量值得被挖掘和分析的价值,对情感进行自动分类的需求成为了自然语言处理的热门领域之一.由于大型预训练语言模型参数较多,针对下游任务进行微调时需要大量有标注语料以及时间对模型进行训练.本文基于多任务联合训练的思想,提出了一种多任务属性感知情感分类模型.首先,该模型采用提示学习的策略将多属性文本拆解为多条单属性文本,并针对可用语料不足的问题使用多个提示拼接文本进行训练;其次,该模型设计了对属性进行分类的辅助任务模块,让模型能关注到文本中属性信息从而作出更准确的预测;最后,在四个常用的公开数据集上进行了实验,通过分析证明该模型能够有效提高属性级情感分类的性能.
In the era of big data,information contains a great deal of value worth mining and analysis,and the need for automatic classification of sentiment has become one of the hot fields of natural language processing.Due to the large pre-training language model with many parameters,it requires a large amount of annotated corpus and when fine-tuning for downstream tasks.Based on the idea of multi-task learning,this paper proposes an aspect-aware sentiment classification model.First,this model uses the prompt learning strategy to disassemble multi-aspect texts into multiple single-aspect texts,and uses multiple prompt concatenation texts for training to solve the problem of insufficient available corpus.Secondly,the auxiliary task module is designed to classify aspects,so that the model can pay attention to the aspect information in the text and make more accurate prediction.Finally,experiments are carried out on four commonly used public data sets,and the analysis proves that the model can effectively improve the performance of aspect level sentiment classification.
作者
刘欣怡
过弋
LIU Xinyi;GUO Yi(Department of Computer Science and Engineering,East China University of Science and Technology,Shanghai 200237,China;Business Intelligence and Visualization Research Center,National Engineering Laboratory for Big Data Distribution and Exchange Technologies,Shanghai 200436,China;Shanghai Engineering Research Center of Big Data&Internet Audience,Shanghai 200072,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2024年第7期1545-1551,共7页
Journal of Chinese Computer Systems
基金
上海市科学技术委员会科技计划项目(22DZ1204903,22511104800)资助.
