Key Technology Research On Knowledge Graph Reasoning Question And Answer For Telecom Operators

With the rapid development of mobile communication technology,telecom operators have introduced diversified services in response to the rising demand of users for communication services.Due to the complexity and diversity of operators’ services and their specialized fields,users often find it difficult to understand or handle telecom services,and therefore need guidance from human customer service.In order to reduce labor costs,major operators have introduced automatic Q&A(Question and Answer)functions,but when users turn to existing Q&A systems,they often find that the systems do not understand the user’s intention well,and when the questions are complex or have unclear semantic descriptions,the systems cannot answer the user’s questions correctly and lack reasoning capabilities.Therefore,it is important to implement an automatic Q&A system in the field of telecom operators and improve its ability to identify and reason about questions,which can reduce cost consumption and improve user experience.In recent years,intelligent Q&A solutions based on knowledge graph can make good use of abstract semantic information,and it has become a general trend to apply knowledge graph and knowledge inference techniques to Q&A systems.In this paper,according to the demand for intelligent Q&A systems in the field of telecom operators,the key technologies of reasoning Q&A for telecom operators’ knowledge graphs are studied.The main work and innovation points of this paper are as follows.(1)Proposes a knowledge representation inference method that incorporates logic rules.The method first applies an automatic rule mining algorithm to the knowledge graph to obtain rules with confidence level higher than a threshold,then it iteratively generates vector embeddings of queries by executing first-order logic operations from the subject entities using the relationships among the knowledge graph entities,injects a rule language in the logic operations,and finally reasoning about answers by calculating the distance between entities and queries in the vector space.The method combines the features of high accuracy and interpretability of rule inference and high efficiency and scalability of representation learning methods to introduce logic rules into the knowledge embedding representation.(2)A query graph-based knowledge Q&A approach is proposed.To reduce the cost of manual annotation as well as to make full use of semantic information,the question is decomposed into four element labels of topic,predicate,object,and query type to express user intent,and a two-way attention mechanism is used to jointly embed the labels and queries to achieve intent recognition of the question.On this basis,the semantics of the problem is parsed into a knowledge graph-based query graph,and the conditional constraints of the query graph are extended for the telecom operator domain to make it have better representation capability for complex problems.Finally,the model is trained by combining the policy gradient of reinforcement learning,and achieves good performance on a real operator dataset.(3)A Q&A system for telecom operators is designed and implemented by combining a knowledge representation reasoning method incorporating logic rules and a query graph-based knowledge Q&A method.The Q&A system uses the knowledge graph as a structured semantic representation framework to construct a knowledge graph of the operator domain by extracting knowledge from relevant data collected by crawlers and other means.After the system requirement analysis and overall architecture design,two key functions of intelligent search and knowledge Q&A are implemented.The system test results show that the Q&A system can understand the user’s intention and meet the user’s needs.