Research And Implementation Of Government Procurement Consultation Question Answering System Based On Knowledge Graph

With the rapid development of China’s economy,government procurement is playing an increasingly important role in economic activities.Providing effective and stable government procurement consulting services for all procurement parties has become an indispensable part.However,the fields involved in government procurement consulting are complex and policy oriented,and the traditional search engine approach is difficult to meet users’ higher use needs.The emergence of question answering system based on knowledge map provides a good idea to solve this problem.By building a knowledge map in the field of government procurement,it is possible to link relevant projects,personnel,processes and policy documents in the field of government procurement.On this basis,a question and answer system can be developed to provide users with accurate and efficient consulting services.It can be seen that it is of great research significance and application value to conduct knowledge mapping research and implement question and answer system around the field of government procurement.This paper carries out research work from the following three aspects:(1)Construction of knowledge map in the field of government procurement: use crawler technology to collect the original data of government procurement websites,and then use data cleaning methods to deal with the problems in the original data.Through the analysis of government procurement data types,8 entity types and 11 relationship types are extracted.Then,the Synonyms Chinese synonym tool is used to analyze the similarity of the cleaned data,integrate the data from different data sources but with high similarity,and combine the extracted entities and relationship types to build a triple format csv file,then import it into the map database Neo4 j,and then build a knowledge map database in the field of government procurement.(2)Research on the core model of consultation and Q&A: In the research on named entity recognition,BMES four digit sequence tagging method was used to build the government procurement entity dataset,and the structure and parameter settings of the Bi LSTM-CRF algorithm model were introduced.In the experiment,compared with the benchmark model CRF,the F1 value of the government procurement entity dataset on the Bi LSTM-CRF model was 3.16% higher than the benchmark model.The experimental results show that it is feasible to combine the two to obtain question entities.In the research of intention recognition,Word2 Vec model is selected to train the word vector and used as the input of Text CNN model to form Text CNN-Word2Vec model.The experimental results show that the classification effect of Text CNN-Word2 Vec model on the self built government procurement question and answer dataset is significantly improved compared with other text classification models.On the basis of the above two studies,the matching of government procurement Q&A results is achieved through the combination of semantic analysis and rule template matching.(3)Implementation of the consultation question and answer system: based on the constructed knowledge map database in the government procurement field and the core model of consultation question and answer,a consultation question and answer system in the government procurement field has been implemented.The We Chat applet is used to build the user interaction interface,the server is developed based on Python,and the Flask framework is used to realize the knowledge base management,question and answer pair management,answer template configuration,question and answer response and other functions.The system can give fast and accurate answers to government procurement consulting questions,and can be used for consulting services in the field of government procurement.