一种半监督学习的金融新闻文本分类算法  

A semi-supervised learning financial news classification algorithm

作　　者：张晓龙 支龙 高剑 苗仲辰 林越峰 项雅丽 熊贇[1,2] ZHANG Xiaolong;ZHI Long;GAO Jian;MIAO Zhongchen;LIN Yuefeng;XIANG Yali;XIONG Yun(School of Computer Science and Technology,Fudan University,Shanghai 200438,China;Shanghai Key Laboratory of Data Science,Shanghai 200438,China;Shanghai Financial Futures Information Technology Co.,Ltd.,Shanghai 200120,China)

机构地区：[1]复旦大学计算机科学技术学院,上海210438 [2]上海市数据科学重点实验室,上海200438 [3]上海金融期货信息技术有限公司,上海200120

出　　处：《大数据》2022年第2期134-144,共11页Big Data Research

基　　金：国家自然科学基金资助项目(No.U1636207,No.U1936213)。

摘　　要：对金融文本进行分类是一项常见的用于识别金融风险的任务。传统的金融新闻文本分类方法需要大量的已知类别文本来训练分类器,然而标注金融新闻文本标签不仅需要专业的金融背景知识,而且耗时耗力。为了减少对已知类别文本的依赖,提出了一个基于半监督学习的金融文本分类算法,该算法采用有监督学习和无监督学习的一致性训练方式,以更好地利用未知类别的文本数据;针对金融领域文本引入无监督数据增强方法,即对特定任务使用特定目标的数据增强方法,以产生更有效的数据。在多个金融文本数据集上开展的实验证明,相比其他文本分类算法,提出的算法在有效性上有明显提升。Classifying financial texts is a common task for identifying financial risks.Traditional financial news classification requires a large number of labeled texts to train the classifier.However,labeling financial news requires not only professional financial background knowledge,but also time-consuming and labor-intensive.In order to reduce the dependence on labeled text,a semi-supervised learning financial text classification algorithm-SSF(semi-supervised learning financial news classification algorithm)was proposed,which uses a consistent training method of supervised learning and unsupervised learning to improve the use of unlabeled data.And unsupervised data augmentation for financial texts was introduced,that is,use specific target data augmentation methods for specific tasks to generate more effective data.Experiments on multiple financial news data sets were conducted to verify that the proposed SSF algorithm has a significant improvement in effectiveness compared with other text classification algorithms.

关 键 词：自然语言处理 文本分类 半监督学习 金融 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]