Research On Classification Method Of Call Center Recording Text Based On Deep Learning

The call center agents of the Network Information Center of colleges and universities create work orders synchronously when answering the call.The historical work orders reflect that incoming calls can be divided into a certain number of business types,and each business type contains high-frequency consulting problems.Therefore,a knowledge base can be constructed according to the type of business,in order to intelligently answer some incoming calls efficiently.Recording classification has become the basis for building an intelligent response call center.A large amount of recording data is usually converted into text to process.The existing recording data has a work order associated with it,so the work order can be used as supplementary information to improve the accuracy of recording text classification model.Based on the call center's recording text and related work order data,this paper improves the existing deep learning classification method,proposes a call center recording text classification algorithm which introduces a weighting factor to sentences features,then implements a call center recording text classification system based on the algorithm.Firstly,a call center recording text classification algorithm for weighted extracting sentences features is proposed.The information in the work order and the first sentence of the recording text usually reflect the subject of the call and contain more keyword information.Therefore,a weighting factor is introduced into the features of sentences.The recording text and the work order text are trained by the ELMo model to obtain the word vector,and then entered to the convolutional neural network to extract the sentence features.After concatenating the sentence features weighted by the above weighting factor,the context is entered to the gated recurrent unit neural network which equipped with the Attention mechanism for semantic feature extraction,and finally the classification results are obtained through the output layer.Secondly,based on the proposed algorithm,this paper designs and implements a call center recording text classification system.The system includes call function layer,classification model layer and application layer.The call function layer is based on the FreeSWITCH telephone softswitch platform,which provides call control,interactive voice response and automatic call distribution services.The classification model layer performs speech-to-text conversion on the recording,data preprocessing,training classification model and predicting using the model.The application layer can obtain user's configuration of agents,voice navigation and classification model hyperparameters,and provides a visual interface to display classification results,model evaluation indices and statistics results.The experimental results show that the classification method which combines work order information and weighted extracts text features has better performance.The recording text classification system based on this algorithm can implement call control function of the call center,also it can use the call recording to train classification model and predict recording labels.