摘要
电信诈骗案件的增加对社会安全和经济稳定造成严重威胁。传统的诈骗案件分类方法主要依赖于人工经验,分类效率低、准确性不高。本文提出了一种基于双向Trnsformer编码器表征模型(Bidirectional Encoder Representations from Transformers,BERT)与潜在狄利克雷分配模型(Latent Dirichlet Allocation,LDA)主题建模的特征融合模型,用于电信诈骗文本笔录的特征提取以及分类。该模型结合了BERT模型的深度语义理解能力与LDA主题建模的文本主题分析能力,通过特征融合技术,将两者提取的特征进行有效整合,从而更全面地捕捉电信诈骗文本笔录的关键信息。实验结果表明,该模型的分类准确率达95.24%,F1-score为95.04%,显著优于GLM-4模型;在12类诈骗案件中如刷单返利、冒充电商客服等,均表现出色,分类效果稳定,数据依赖性较强。融合BERT与LDA的模型能有效捕捉文本语义与主题特征,为电信诈骗案件智能化分类提供了高效解决方案,对提升警务工作效率具有重要实践价值。
The increase in telecom fraud cases poses a severe threat to social security and economic stability.Traditional methods for classifying fraud cases primarily rely on human experience and are confronted with issues such as low classification efficiency and poor accuracy.This paper proposes a feature fusion model based on the pre-trained large-scale model BERT and LDA topic modeling for feature extraction and classification of telecom fraud text transcripts.This model combines the deep semantic understanding capability of the BERT model with the text topic analysis capability of LDA topic modeling.Through feature fusion techniques,it effectively integrates the features extracted by both methods,thereby more comprehensively capturing key information in telecom fraud text transcripts.Experimental results demonstrate that the model achieves a classification accuracy of 95.24%and an F1-score of 95.04%,significantly outperforming the GLM-4 model.The model performs excellently across 12 categories of fraud cases(such as click farming rebates,impersonating e-commerce customer service,etc.),with stable classification performance and strong data dependency.The experimental results indicate that the model fusing BERT and LDA can effectively capture textual semantic and thematic features,providing an efficient solution for intelligent classification of telecom fraud cases and offering significant practical value in enhancing police work efficiency.
作者
王江涛
WANG Jiangtao(School of Information and Cyber Security,People's Public Security University of China,Beijing 100038,China)
出处
《智能计算机与应用》
2025年第12期69-73,共5页
Intelligent Computer and Applications
