Research And Implementation Of Generative Intelligent Voice Customer Service Based On Deep Learning

In order to solve customers’ questions about products or services,all walks of life have customer service centers to provide customers with consulting services.With the development of the Internet era,the amount of data has exploded,and deep learning technology has gradually matured and begun to be widely used.In order to solve the problems of uneven business abilities and limited working hours of customer service staff,and reduce labor costs in the customer service industry,traditional voice customer service began to be transformed from manual to automated or semi-automated form,and a new type of intelligent voice customer service system came into being.In the field of product customer service,most of the current mainstream intelligent customer service systems are based on deep learning models to drive related business products.These models are based on recurrent neural networks,supplemented by the attention mechanism to improve semantic understanding,and have achieved significant results in related business scenarios,but there are still two problems.First of all,due to the sequential nature of recurrent neural networks,this type of model is slow in training and inference.Secondly,the current similar models have the problem of multiple occurrences of the same semantics but different contexts in business texts,which leads to misjudgments of semantics.In order to solve these two problems,this paper proposes a spatial feature fusion question answering network model based on depthwise separable convolution and attention mechanism.This paper uses deep separable convolution to replace the traditional recurrent neural network,reduce the amount of its parameters,and increase the training speed;the interaction of semantic global and local information is formed by multiple encoders,decoders and attention modules,which makes the mordel reach and exceed the performance of existing similar models without using a large-scale network structure.Addressing the problem of semantic misjudgment,this paper designs a spatial feature fusion algorithm to assign a learnable weight to each layer feature,which aims to adaptively filter out contradictory or useless information in features,form more effective features,and improve the accuracy of the model.Finally,this paper makes a full experimental comparison between the proposed model and other comparison models,and achieves an EM/F1 value of 76.8/84.8,which verifies the effectiveness of the algorithm in this paper.Based on the above model,this article implements an end-to-end generative intelligent voice customer service system in the field of product customer service,which can better complete automatic question and answer tasks and has a high accuracy rate.