Design And Implementation Of Information Project Pre-evaluation Model Based On Natural Language Understanding

In order to respond to the call of information construction in Shaanxi Province and further promote the reform of information modernization,there are emergency needs for unified life cycle management and performance evaluation for the information projects declared by the network security and information project management platform in Shaanxi Province.The complete life cycle of the information project is divided into six stages: declaration,preevaluation,project establishment,implementation,acceptance and post-evaluation.Based on the evaluation system provided by the project "Research on performance Evaluation methods of Network Security and Informatization Project",this thesis proposes a preevaluation model of information project which should be evaluated in advance.The model uses automatic pre-evaluation method to solve the problems of subjectivity and nonrepeatability of manual pre-evaluation.The pre-evaluation of information project is divided into two stages: structure pre-evaluation and content pre-evaluation.The information project first completes the structure preevaluation,and then carries on the content pre-evaluation.After getting the pre-evaluation score of the information project,the model proposed in this thesis can judge whether the project can enter the next step of the process.The main research results of this thesis are as follows:(1)Pre-evaluation of information project structure.The main task of the structure preevaluation model is to complete the matching of the directory structure of the informatization project documents to judge the completeness of the project documents.When the relevant departments receive the project proposal application,they shall conduct a unified inspection of the directory structure of the declared project according to the official project template,and require the directories to be identical or semantically similar.Based on Bi-GRU network and attention mechanism,the structure pre-evaluation model extracts the features of the project structure text and maps the text class data into feature vector space.The experimental results show that the structure pre-evaluation model can match the project structure text efficiently.(2)Pre-evaluation of information project content.The information project document is divided into four chapters according to the first-level catalog.And each chapter is an independent text paragraph,pre-evaluated by the content pre-evaluated model.The content pre-evaluation model first uses the BERT to extract the text semantic information directly,and then carries on the secondary feature extraction of the text feature vector based on the Bi-GRU network.The optimized text feature vector is input into a nonlinear regression model based on neural network,and the pre-evaluation score of the project is obtained.Experiments show that Bi-GRU network is better than other neural networks(CNN,LSTM,etc.)in the secondary embedded representation of text features extracted by BERT,which proves the effectiveness and feasibility of the content pre-evaluation model.(3)The application of information project pre-evaluation model.The pre-evaluation model of information project is composed of structure pre-evaluation model and content preevaluation model.According to the score range,the project chapter content data sets are classified as excellent,good,medium and poor.The average classification accuracy of the pre-evaluation model proposed in this thesis is as high as 98.6% on the four project content data sets.The experimental results show that the model proposed in this thesis has practical application significance in the subject of "Research on performance Evaluation methods of Network Security and Informatization Project",and can complete the pre-evaluation task of network security and informatization projects.