Optimization Design Of Intelligent Customer Service System For Customized Equipment Enterprises

Under the trend of growing company scale and increasingly customized features of equipment products,the traditional information interaction model that uses mail,telephone,fax,and paper documents as information media is lagging behind.Due to the diversity of questions asked by customers,companies need to arrange internal professionals to answer questions raised by customers.This backward information interaction mode will make customers feel that the process is cumbersome and slow,and at the same time will often interrupt the work rhythm of internal staff.In order to alleviate this situation,the customized equipment manufacturing enterprise A has developed an intelligent customer service.However,due to the limited system design level,the existing intelligent customer service system is cumbersome to use,and the user's satisfaction with the system's answers is very low.After in-depth analysis,The reasons for this result are:1)The questions are not classified and the questions and answers are not set permissions,resulting in a global traversal in the question database for each question of the user,which is inefficient and poorly effective;because no permis sions are set,the answer reply is confusing and may Leaking business secrets.2)The question preprocessing is too simple,there is no domain professional word labeling,updated stop word list,etc.for the actual information interaction of Enterprise A.For domain professional words such as "drawing plan","drill vent","match "Side board","Rotation amplitude","Knurling line",the word segmentation effect is confusing.3)The way to match user questions with questions in the question library is too simple.Therefore,it is of great significance to optimize the enterprise A's intelligent customer service system.In order to solve the contradiction between the company's growing scale and the backward information interaction mode,on the basis of in-depth analysis of the reasons that caused the original intelligent customer service system to be cumbersome to use and the answer returned to chaos,and the user's satisfaction with the system was low,Enterprise A decided to optimize its Intelligent customer ser vice system.Combining the actual situation of Enterprise A,the following optimization ideas are proposed:1)Data preprocessing combined with information characteristics.2)Classify questions by combining domain characteristics.3)Use the cosine similarity algorithm for sentence matching.The purpose of this research is to improve the accuracy of answer return and user experience in the intelligent customer service system.In order to achieve the goal,this article decided to re-optimize the overall design of the system.With the development of the customized equipment manufacturing industry,the amount of relevant data has surged,providing sufficient original data for the training of question classification models.In recent years,natural language pro cessing technology has also been maturing.The above development provides methods and conditions for the study of question classification and sentence similarity calculation in this paper.This study mainly made the following work:1)On the basis of analyzing historical interactive corpus and conducting on-the-spot investigations,a demand analysis was conducted for A enterprise intelligent customer service system,and the system architecture,functional structure number and database were designed.2)Detailed design of the core functions of the system.Three machine learning methods are used to classify the question sentences in order to narrow the scope of the next question similarity matching,and then use the cosine similarity algorithm to calculate the sentence similarity in the corresponding range,and verify the effectiveness of the algorithm.In the question classification part,the accuracy of the three machine learning methods of Logistic regression,BP neural network and support vector machine SVM are respectively,the running time is 67.6%,83.5% and 72%,and the average classification time of each sentence is 1.3S,0.9S and 1.6S.In the current number of training sets,BP neural network works best.In the similarity calculation part,the accuracy rates calculated based on TF-IDF similarity,keyword weight similarity,and cosine similarity are 51.6%,56.6%,and 82.4%,respectively,under the same database and test set,The effect based on cosine similarity calculation is better than that based on TF-IDF similarity calculation and keyword weighted similarity calculation.3)Tested the effect of system optimization and showed some user interfaces.After optimizing the system,the accuracy of answer return was increased from 43.2% to 80.6%,the average running time was not much different,and the average evaluation star was increased from 1.2 to 3.5 stars,achieving the expected optimization effect.This research is of great significance to the optimization of A's intelligent customer service system.