Research And Application Of Domain Ontology Construction For Service Chat Robot

With the rapid development of Internet commerce,more and more businesses provide online shopping services,which brings a lot of online customer service work.Replacing manual customer service with chat robots has become a growing choice for businesses.How to improve the work quality of customer service chat bots has become a hot issue for businesses.Compared with ordinary chat bots,the knowledge faced by customer service chats is mostly limited to a specific business area,which brings convenience to the use of ontology to store their knowledge.This thesis takes the customer service of overseas purchase field as an example,firstly analyzes the merchant website and the history of manual customer service,studies the semi-automatic construction method of the ontology in the customer service domain,and then use the constructed ontology library to extract customer intention,build a multi round answer retrieval model based on customer intention constraint and deep learning.Finally,we use the built model to implement the customer service chat robot system.This thesis started the research in the following aspects:(1)Research the semi-automatic construction method of the ontology in the customer service domain,and construct the customer service domain ontology of overseas purchase.In the process of constructing the ontology,this thesis first manually extracts information from the merchant website,constructs the ontology conceptual class and concept relationship;secondly,crawls the standard expression of the instance from the merchant website according to the concept class using the crawler,and uses the similarity of the word related ontology improved synonym expansion algorithm,improved the accuracy of the judgment of synonyms,and the method has been used to expand the synonymous expression of instances in the history of manual customer service.Finally,integrated and constructed the domain ontology database.(2)Constructed a multi-round answer retrieval model based on customer intent constraints and deep learning.This thesis first uses a hierarchical recurrent neural network to construct a deep retrieval chat model,and using the intention vector constraint model in order to solve the problem that model easily returns questions that deviate from customer intent.In the construction of the intention vector,this thesis use semantic analysis to extract the customer intention representated by ontology,and construct the customer intention vector at each moment by updating intention network.Finally,the experiment shows the effectiveness of the model.(3)Based on the previous research,this thesis designed and built a customer service chat robot system.After acquiring the customer's problem,the system firstly extracts the intention based on the ontology library,and then obtains the responses respectively based on the ontology library reasoning and the deep retrieval model.Finally,the system compares the responses and selects one of them to return to the user.The thesis finally shows the dialogue effect of the customer service chat robot.The performance of customer service chatbots was evaluated through experiments.The results show that the customer service chat robot constructed in this thesis has reached the expected requirements.