Design And Implementation Of A Power Spectrum Question Answering System

With the rapid growth of China's economy,the business data of electric power enterprises is growing more and more rapidly.The scope and depth of enterprise services for customers are becoming more and more complex.The way of enterprise services for customers,whether traditional services or remote services,needs to consume a lot of service resources to answer questions and handle business with customers.It is necessary to change the existing business model,use artificial intelligence technology to provide customers with comprehensive and diversified services,use the intelligent response model to answer questions for customers,and update the business service model in time.Firstly,based on the data of power field,this thesis designs the knowledge base by artificial modeling,including ontology design and relationship design,extracts business entities from unstructured data by Python program.secondly,this system constructs entity recognition model to identify business entities.Two kinds of neural networks are used in the coding layer:bilstm and CNN.In the decoding layer,the conditional random field(CRF)algorithm is used to ensure the validity of label location.Thirdly,the system classifies the prediction entities based on the machine learning classification algorithm of SK learn library.This thesis trains Bayes classifier,ridge regression classifier,MLP classifier and random forest classifier successively.By analyzing and comparing the performance of each classifier to each label,we choose the best classifier as the entity classification of the system for entity recognition model prediction.Finally,the development of the front-end of the system is completed under the framework of Django The drawing of the front-end page and the calling of the interface,and the background development based on the flask framework.The evaluation of entity recognition model and classifier model in this system is based on confusion matrix.The performance of each classification label in their respective accuracy,accuracy,recall and F1 value is analyzed,and the optimal model is combined.Finally,the similarity between candidate answers and questions is calculated by similarity algorithm,and the answer with the highest similarity is returned to the app for display.