Research On Knowledge Base Retrieval Generative Intelligent Question Answering Method For Cloud ERP Ecological Community

Under the cloud service model,cloud ERP providers,various user groups in the supply chain,developers and software developers and other subjects,businesses,data and their relationships have formed a cloud ERP ecosystem.Relying on the ecosystem to create a cloud ERP ecological community that communicates,opens,shares and cooperates with all parties is an important content and trend in the development of cloud ERP.Among them,cloud ERP service consultation is a key link of ecological operation services,and an important guarantee for flexible and fast communication and sharing of all individuals and organizations in the ecological community.However,the existing consulting service methods have high requirements for users’ professional knowledge and low retrieval efficiency,and cannot provide services to various customers in a timely,convenient and effective manner.Considering that the cloud ERP ecological community has a lot of domain data in terms of time and space,it is an urgent need and main task to create professional intelligent customer service by combining large-scale customer service session data and knowledge base.To this end,this paper uses the customer service data and knowledge base in the Kingdee Cloud ERP ecological community as the data support,and focuses on the intelligent question answering technology for online consultation and technical services in the ecological community.The main contents are as follows:（1）Aiming at the problem that a single process is not conducive to the rapid solution of similar problems,the overall technical solution of knowledge base retrieval generative intelligent question answering for the cloud ERP ecological community is researched and proposed.（2）In view of the characteristics of high sensitivity,high noise,and strong professionalism of text data in the cloud ERP ecological community,data preprocessing has been completed,including data desensitization,cleaning and denoising,word segmentation,stop word removal,and text vectorization.（3）Aiming at the problems that the questions are not effectively classified and the question and answer process is not easy to control,a problem classification control method based on BERT-Text CNN is proposed,and through experiments on 6 different domain datasets,the problem classification in the cloud ERP domain is The F1 value reaches 97.35%,which verifies the effectiveness of the method.（4）Aiming at problems such as low accuracy and efficiency of question retrieval,a knowledge base retrieval question answering method based on BERT-CRF-SMI is proposed.The method includes a BERT-CRF-based question entity recognition model and a BERT-SMI question attribute similarity matching model.This method can accurately reply to the questions based on the knowledge base.Through the experiments on the Chinese knowledge base dataset,the F1 value of the knowledge base question and answer is 94.82%,which verifies the effectiveness of this method.（5）In view of the difficulty of answering questions outside the knowledge base and the low flexibility of question and answer,a generative question and answer model based on GPT2 is proposed,which effectively answers the questions outside the knowledge base and improves the flexibility of question and answer.The experimental results show that the BLEU value reaches 19.51%,which verifies the effectiveness of the method.（6）In order to verify and test the usability and interactivity of the generative question answering method for cloud ERP ecological community knowledge base retrieval,this paper designs and develops an intelligent question answering tool system,through which the questioner can realize question and answer consultation.