Online Education Customer Service Data Mining And Design Of A Chatbot

The online education customer service is getting more and more attention.The customer service is an important way for online education providers to understand users' needs and experience.It is also an important reference source for online education providers to timely obtain user's feedbacks and improve their products.In practice,there are two problems in the customer services:The online education provider has accumulated lots of customer service data.But these data have not been fully explored and utilized.And at present,customer service is conducted by human labors and thus suffers from problems such as unstable servicing quality and high cost.It is urgent to have automated customer services to overcome this problem.The progress of natural language processing technology,especially the deep learning natural language processing technology,brings hope to the resolution of the above problems.This paper applies natural language processing technologies in online education customer services to solve the above problems.The contributions are as follows:(1)We conducted data mining on customer service data and made the basic preparations for the design of chatbot,including:For the online customer service data,we used machine learning algorithms to cluster the dialogue texts and found that class selection is one of the most frequently used customer service tasks.For the telephone customer service data,we developed programs to evaluate the performance of six Chinese commercial speech recognition systems.We found that the error rates of systems are usually high.The best one is Alibaba's system,with error rate 49.55%.The other systems have higher error rates,which are over 60%.We designed a customer service dialogue text classification system based on machine learning and deep learning methods.We further classified the customer service data and analyzed the basic rules and characteristics of the class selection dialogue.The evaluation on the customer service dialogue dataset showed that the accuracy of the classifier based on deep learning got 99.16%.The conclusions have been adopted by the related company and applied in practice.(2)We designed and generated 30,000 groups of customer service dialogue corpus of class selection and trained a chatbot model based on this corpus.We first analyzed the users' class selection scenarios and sub-tasks based on the collected customer service dialogue data.We then generated the corpus and trained the chatbot model based on the end-to-end memory networks.The evaluation results show that our chatbot achieves similar(even higher in some tasks)dialog accuracy to similar English systems.Our study proved that the end-to-end memory network has excellent performance in task-oriented Chinese chatbot systems.Finally,we interconnected the chatbot model with a web interface,and implemented a prototype of a realistic chatbot system.We evaluated the system response delay and demonstrated the practicality of the entire system.Our work in customer service data mining helped companies find the most urgent needs of users and provided suggestions for making related decisions.In terms of customer service chatbot,we completed the class selection chatbot prototype,which demonstrated the feasibility of a intelligent customer service system.