基于ResNet-BiLSTM模型的电力客服工单分类研究

Research on power system customer service tickets classification based on ResNet⁃BiLSTM model

电力客服工单记录着用户在用电过程中的需求、建议和意见,对电力客服工单进行有效的分类,对提升电力系统运行质量和用户体验有着重要意义。对此基于提升工单分类准确度为目的,通过将残差卷积网络(Residual Convolutional Network,ResNet)与双向长短时记忆网络(Bilateral Long-Short-Term Memory network,BiLSTM)结合构造ResNet-BiLSTM模型来挖掘和学习客服工单中的深度语义信息。将ResNet-BiLSTM模型在真实客服工单数据集上进行实验验证,其分类准确度达到了90.8%。结果表明,相较于TextCNN、BiLSTM和ResNet三种模型,所提出的方法识别准确率分别提升了1.6%、6.2%和10.6%,能够很好地实现电力系统客服工单分类。

The data of power customer service tickets contains customer demands,suggestions,and opinions in the process of electricity consumption.Effective classification of power customer service tickets will be important to improve the operation quality of the power system and customer experience.In order to improve the accuracy of power customer service tickets classification,the Residual Convolutional Network(ResNet)and two⁃way BiLSTM(Bilateral Long⁃Short⁃Term Memory Network)are combined to learn the deep semantic information of customer service tickets.The ResNet⁃BiLSTM model was experimentally verified on the real customer service tickets data set,and its classification accuracy reached 90.8%.The results show that compared with TextCNN,BiLSTM and ResNet,the recognition accuracy of the proposed method is improved by 1.6%,6.2%,and 10.6%,respectively,which can give an outstanding performance on the classification of power customer service tickets.