一种基于BERT微调-TextCNN的电信网络诈骗案情文本分类设计

Research on text classification of telecom network fraud cases based on BERT fine tuning TextCNN

为了有效遏制电信网络诈骗案件高发多发态势,公安机关在持续实行高压严打政策的同时,还需注重打防结合,以防为先,突出精准宣传。电信网络诈骗类型多样,各具特点。通过归纳总结特征进行诈骗类型分类,可以达到对受骗者诈骗类型进行预测的目的,以帮助公安机关精准宣传。目前,警务实践中通过人工标注的方法过于依赖标注人员个人经验,进而耗费一定警力资源。本文采用了BERT模型与卷积神经网络CNN模型相结合的BERT微调-TextCNN模型。首先,利用BERT微调生成包含上下文信息的动态词向量,然后通过TextCNN提取文本局部特征,最后通过全连接层传入Softmax进行多分类。实验结果表明,在诈骗案情文本分类研究中,相比于TextCNN和BERT微调,BERT微调-TextCNN在准确率上分别提升了7.71%和6.3%,效果显著。借助BERT微调-TextCNN模型快速准确地对诈骗案情文本进行分类,让警务人员快速掌握受骗人被骗类型从而进行精准宣传,可以优化警力资源配置,节省警务成本。

In order to effectively curb the high incidence of telecom network fraud cases,while continuing to maintain a high-pressure and crackdown policy,the public security organs must also pay attention to the combination of strike and prevention,put prevention first,and highlight precise publicity.Telecom network fraud types are diverse and have their own characteristics.By summarizing the characteristics and classifying the fraud types,the fraud types of the victims can be predicted so as to help the public security organs conduct accurate publicity.The current manual annotation method in police practice relies too much on the personal experience of the annotator,which consumes a certain amount of police resources.This paper proposes the BERT-FineTuning-TextCNN model,which combines the BERT model and CNN model.First,the self-attention mechanism in BERT is used to obtain the semantic relationship between words,then text features are extracted through TextCNN,and finally multi-classification is performed through the fully connected layer.Experimental results show that in the study of text classification of fraud cases,compared with TextCNN and BERT-FineTuning,BERT-FineTuning-TextCNN has improved the accuracy by 7.71%and 6.3%respectively,and the effect has been significantly improved.The BERT-FineTuning-TextCNN model can be used to quickly and accurately classify fraud case texts,allowing police officers to quickly understand the type of deceived people and conduct precise publicity,which can optimize the allocation of police resources and save police costs.