基于目标驱动与注意力机制的挪用检测方法

Misappropriated Call Detection via Target-driven and Attention Mechanism

随着信息时代的发展,语音通信体量变得越来越庞大,而极少量线路挪用的行为易藏匿于正常业务中,造成运营商监管的盲区。针对以上问题,提出基于目标驱动与注意力机制的挪用检测方法,将挪用检测问题转变为求解文本序列之间语义相似度问题。采用注意力机制设计孪生神经网络,在此基础上基于目标驱动方法对所提出的神经网络进行联合训练,得到适用于语音通信场景的语义相似度求解模型。之后设计在线挪用检测算法,甄别出未报备通信行为。通过在实际生产数据上的实验,验证了所提方法的有效性和准确性。

With the development of the information age,the volume of voice communication has become larger and larger,and a very small amount of line embezzlement is easy to hide in normal services,causing blind spots for supervision.In order to solve the above problems,proposes a method of misappropriated call detection based on target-driven and attention mechanism,and transforms the problem of misappropriated call detection into a problem of semantic similarity between text sequences.The attention mechanism is used to design the twin neural network,and on this basis,the proposed neural network is jointly trained based on the target-driven method,and a semantic similarity solving model suitable for voice communication scenarios is obtained.After that,an online misappropriated call detection algorithm is designed to screen out unreported communication behaviors.Through experiments on actual production data,the effectiveness and accuracy of the proposed method are verified.