Design And Implementation Of Intelligent Inspection System For Customer Service Speech Based On Machine Learning

Customer service speech quality inspection is a very important quality control link in hotline service operation.Traditional Speech quality inspection of customer service system derives recordings by manually applying the assessment standard form and mainly by the way of sampling,which owns low quality inspection efficiency and low coverage.This dissertation designs and implements the Speech quality inspection of the customer service system,using the Convolutional Neural Network model in machine learning to calculate text similarity,evaluate the deviation rate and give a mark to the quality inspection,which attain high degree of automation,wide coverage,and greatly improve the accuracy of quality inspection.This dissertation analyzes the current situation of customer service speech quality inspection,deeply studies the speech text data of customer service department of Hebei Sibosi Innovation Technology Co.,Ltd.,designs the overall framework through demand analysis,and realizes the intelligent quality inspection system of customer service speech based on machine learning.The main research contents are as follows:(1)Data acquisition.The data came from the speech data of customer service of Hebei Sibosi Innovation Technology Co.,Ltd.,over the years.(2)Data preprocessing.Firstly,the acquired original text was cleaned,and the word segmentation tool Hanlp was used for Chinese word segmentation and word stopping.(3)Model selection.Two kinds of text similarity models were proposed,one was text similarity model based on document vector,the other was text similarity model based on CNN,using company customer service data as data set,calculating text similarity of two models,combining with quality inspection standard to obtain deviation rate.The deviation rate based on CNN text similarity model was 0.05-0.06,and the deviation rate of text similarity model based on document vector was 0.12-0.13.Experiments showed that the CNN text similarity model had low deviation rate and high correct rate,so the model was selected for quality inspection.(4)Model training.Word2 Vec algorithm was used to train the word vector and establish the word vector model.After the processing of Word2 Vec algorithm,the semantic extension matrix was generated as the input of CNN,two completely identical CNN hierarchicalcombination models were established,the full connection layer generated the high-level semantics,and the Sigmoid activation function of the activation layer was finally used for the output,so as to identify the probability of belonging to a certain class,and then the probability was used to judge the similarity of text.The innovation of this dissertation is that it introduces the CNN model to calculate the text similarity into the customer service speech intelligent quality inspection system.Based on the B/S architecture,the customer service speech intelligent quality inspection system is designed and implemented.The system mainly completes the functions of automatic quality inspection score,speech mail management and so on.Through the application,the request of enterprise for the speech quality inspection service can be better satisfied.