| Informatization construction is the backbone of promoting social development and improving the level of the national economy.In order to further strengthen the informatization construction of Shaanxi Province and improve the level of government affairs informatization,it is urgent to carry out the intelligent management of the whole life cycle of the network security and informatization projects in Shaanxi Province,so as to standardize the management process of the project.The project management process is divided into project declaration,project pre-assessment,expert review and other stages.Focusing on the pre-assessment stage of the project,the traditional method only uses manual assessment,resulting in a greater impact of human subjectivity on the assessment results.The thesis aims to achieve automatic scoring through a natural language understandingbased project pre-assessment system,thereby reducing the impact of human factors on the assessment results.Project pre-evaluation models are studied based on text embedding and network representation learning methods in this thesis.Based on this model,a project pre-assessment system is designed and implemented.The system uses the model for scoring,and the scoring results serve as the reference for expert review.Experts conduct the final review.The project pre-assessment model focuses on the content of the project declaration form.Understanding the text content requires two aspects: on the one hand,the meaning of the words in the text,and on the other hand,the semantic information of the context in the text.Both are important factors that affect project pre-assessment.The main research contents of the thesis are as follows:(1)An item pre-evaluation model based on local semantic features is studied in this thesis.The model focuses on the impact of the meaning of words in the text on project preevaluation.The model uses CNN and Bi LSTM models to extract text features in parallel,and then performs feature fusion.The thesis adopts a personalized feature fusion method.First,vector splicing is performed,and then the final text feature vector is obtained through the convolutional neural network.Finally,the neural network-based regression model is used for scoring.According to the first-level indicators in the project pre-assessment indicators,the project application form is divided into four parts,and four data sets are established for evaluation respectively.The validity of the project pre-assessment model based on local semantic features is proved by experiments.(2)An item pre-evaluation model based on global semantic features is studied in this thesis.The model focuses on the contextual semantic information of the text.The model processes the preprocessed text into segments,and then obtains the text embedding representation through BERT.In order to further optimize the text feature vector,the final text feature vector is obtained through the Bi LSTM+Attention model,and the model finally uses the neural network-based regression model to score.Compared with other algorithms,the effectiveness of Bi LSTM model in optimizing text feature vector is proved by experiments,and the superiority of project pre-evaluation model based on global semantic features is proved.(3)Based on the research of the project pre-assessment model,the thesis designs and implements the project pre-assessment system.The system has functions such as user declaration,system scoring and expert review.System scoring is done by invoking the project pre-assessment model.The project pre-evaluation model consists of a project preevaluation model based on local semantic features and an project pre-evaluation model based on global semantic features,and the final score is the average of the pre-evaluation scores of the two models.The system score will be used as a reference for the expert review,and the expert will give the final review opinion.The project pre-assessment system has complete functions and good performance,and can serve users safely and efficiently. |
