FinBERT-RCNN-ATTACK:金融文本情感分析模型
FinBERT-RCNN-ATTACK:Emotional Analysis Model of Financial Text
段魏诚 1薛涛1
作者信息
- 1. 西安工程大学 计算机科学学院,陕西 西安 710600
- 折叠
摘要
金融文本包含投资者的情绪及公众对相关事件的态度.近年来,自然语言处理已广泛应用于金融领域,对金融文本数据进行情感分析可以得到丰富的投资价值和监管参考价值.然而由于金融词汇具有专业性和特殊性,现有的通用情感分析模型不适合金融领域情感分析任务,精确度有待提高,且现有模型易受到对抗样本的干扰导致模型结果出错.为了解决这些问题,提出了一个FinBERT-RCNN-ATTACK模型.利用在金融语料库预训练的FinBERT模型进行词嵌入处理,提取语义特征,将提取到的特征引入RCNN模型进一步挖掘上下文的关键特征,并且在模型中引入对抗训练,即在嵌入阶段添加扰动,提高模型的鲁棒性和泛化性.实验结果表明,在金融领域数据集上,提出的模型优于其他情感分析模型,精准度提升了3%～35%.
Abstract
The financial text contains investor sentiment and public attitudes towards the events.In recent years,natural language processing has been widely used in financial field,and the emotional analysis of financial text data can get rich investment value and regu-latory reference value.However,due to the professionalism and particularity of financial vocabulary,the existing general emotional analysis model is not suitable for the emotional analysis task in the financial field,and the accuracy needs to be improved,and the existing model is vulnerable to the interference of antagonistic samples,leading to the wrong model results.In order to solve these problems,we proposed a FinBERT-RCNN-ATTACK model.The FinBERT model pre-trained in the financial corpus is used for word embedding processing to extract semantic features,and the extracted features are introduced into the RCNN model to further excavate the key features of the context.In addition,adversarial training is introduced into the model,that is,disturbance is added in the embedding stage to improve the robustness and generalization of the model.The experimental results show that the proposed model is better than the other e-motional analysis models,and the accuracy is improved by 3%～35%.
关键词
金融文本/情感分析/FinBERT/循环卷积神经网络/对抗训练Key words
financial text/emotional analysis/FinBERT/recurrent convolutional neural networks/adversarial training引用本文复制引用
基金项目
陕西省技术创新引导专项计划资助项目(2020CGXNG-012)
出版年
2024
国家科技期刊平台
NSTL
万方数据