| 引用本文: | 杨 艺1 ,黄镜月1 ,贺品尧1 ,荣 婷2.基于人工与 ChatGPT 标注的推文情感分析对比研究(J/M/D/N,J:杂志，M：书，D：论文，N：报纸).期刊名称,2025，42（4）：95-101 |
| CHEN X. Adap tive slidingmode contr ol for discrete2ti me multi2inputmulti2 out put systems[ J ]. Aut omatica, 2006, 42(6): 4272-435 |
|
| 摘要: |
| 目的 针对特定推文情感分析任务中标注数据的困难和由于标注不准确带来的分类结果不尽如人意问题,提出
一种机器标注数据的方法来研究深度学习模型对人工标注和机器标注推文数据情感分类的性能表现差异。 方法 研
究中,对于统一的标签体系,分别对推文数据进行人工标注和运用 ChatGPT 模型接口标注,再采用 BERT-TextCNN 深
度学习混合模型,对经过人工标注和 ChatGPT 标注的数据集进行情感分类。 结果 实验结果表明:人工标注数据集在
整体性能上表现出更高的准确性和可信度,但是在某些推文数据上, ChatGPT 大模型以其比人更丰富的知识储备,可
以生成比人更客观科学的可解释性标注,在情感分类结果上呈现出一定的优势,人工标注和机器标注方法各具优劣;
由此可以得出对于文本情感分类任务,机器标注是一种可行的标注方法。 结论 在实际应用场景中,可以根据任务需
求灵活选择和结合两种标注方法,充分利用两者之间的优势,以达到更佳的分析性能和效果。 |
| 关键词: 人工标注 ChatGPT 标注 推文 情感分析 BERT-TextCNN |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Comparative Study on Sentiment Analysis of Tweets Based on Manual and ChatGPT Annotation |
|
YANG Yi 1 HUANG Jingyue 1 HE Pinyao 1 RONG Ting2
|
|
1. School of Artificial Intelligence Chongqing Technology and Business University Chongqing 400067 China
2. Chongqing Center for Research and Consultancy of Cyber Public Opinion and Ideological Development in Universities
Chongqing Technology and Business University Chongqing 400067 China
|
| Abstract: |
| Objective This study addresses the challenges of annotating data for specific tweet sentiment analysis tasks and
the issues arising from inaccurate annotations leading to unsatisfactory classification results. A method for machine
annotation of data was proposed to investigate the performance differences in sentiment classification of tweets annotated by
human annotators and the ChatGPT model. Methods In this study for a unified labeling system tweet data was
annotated both manually and using the ChatGPT model interface followed by sentiment classification using a BERT-
TextCNN hybrid deep learning model on both datasets. Results Experimental results indicated that the manually
annotated dataset exhibited higher overall accuracy and reliability. However for certain tweet data the ChatGPT model
with its richer knowledge base can produce more objective and scientifically interpretable annotations showing certain
advantages in sentiment classification results. Both human and machine annotation methods have their strengths and
weaknesses. Therefore it can be concluded that machine annotation is a feasible labeling method for text sentiment
classification tasks. Conclusion In practical applications it is advisable to flexibly choose and combine both annotation
methods based on task requirements and fully leverage the strengths of these two methods to achieve better analytical
performance and outcomes. |
| Key words: manual annotation ChatGPT annotation tweets sentiment analysis BERT-TextCNN |
