Design And Implementation Of E-commerce Customer Service Dialogue System

With the development of mobile Internet technology,online shopping is widely popular among the crowd due to its low price and convenience.The e-commerce platform should not only provide customers with rich products,but also provide high-quality customer service to ensure that people can have a good shopping experience.The introduction of customer service robots customized according to e-commerce business can effectively improve the reception efficiency of manual customer service,reduce a lot of repetitive work,and promote order conversion.The essence of the customer service robot is a dialogue system.The system administrator usually needs to configure the question and answer in the system in advance.When a customer consults,the system feeds back the best answer to the customer through steps such as semantic understanding,similarity matching,and answer retrieval.This type of retrieval question and answer guarantees the quality of answers to a certain extent,but due to the limitation of the size of the knowledge base,it is impossible to answer all customer questions.In order to improve the robot response coverage,a supplementary response module is added to this system.This module can reply the answer automatically generated by the model when the customer question can not hit the knowledge point.Focusing on the response capabilities and related functions of the dialogue system,the main work of this paper is as follows.This paper first analyzes the main demand points for customer service robots in the e-commerce industry,including: the need to realize automatic answering of business knowledge,configure system preset knowledge,and configure custom knowledge.In order to meet the above points,the system uses a lot of e-commerce data as model training data when constructing,so as to obtain the word vector model used to represent the text in the system,and then through short text cosine similarity calculation and similarity ranking The algorithm finally outputs the knowledge point answer with the highest similarity.This method can guarantee the response accuracy and coverage.At the same time,a generative question and answer module was added to the system,which was used to respond to customer questions that did not reach the similarity threshold.The model is mainly based on the Seq2 Seq network structure,and its encoderdecoder structure can well obtain the semantic information between the question and answer pairs to obtain the generated answers.By using LSTM as the cell structure,the gradient disappearance problem of RNN can be avoided.In addition,by adding the Attention mechanism to the Decoder structure of the model,it is possible to assign weights to the output data to obtain a better answer generation effect.Finally,the test window function is added to this system.This function can simulate customer questioning scenarios and respond to feedback results,which is convenient for system administrators to adjust the knowledge base in time.The final result of this paper is an e-commerce customer service dialogue system that can be delivered.The value of the system lies in helping the e-commerce customer service to improve the work efficiency and improve the response effect through the customized customer service robot,and realize the 24-hour reception.After evaluation,the system function tests all passed;the accuracy rate of the knowledge base response module was 65.37%,and the accuracy rate of the supplementary response was 56.14%.The system has great market application value.