Design And Implementation Of The Net-Information Project Pre-Evaluation Model Based On Text Embedding

In recent years,Shaanxi Province's informatization construction has stepped into the stage of comprehensively deepening applications.In order to respond to the requirement of the informatization plan of Shaanxi Province,it is necessary to carry out unified life cycle management of government-invested Network Security and Informatization Projects(referred to as "Net-Information Projects"),so that there is a standardized management process for the Net-Information Projects.A complete life cycle of the Net-Information Projects includes seven phases: project declaration,project pre-evaluation,project approval,project implementation,project acceptance,project follow-up inspection,and project performance evaluation.The goal of this thesis is to solve the decision-making problem in the pre-evaluation phase of the Net-Information Projects.Project pre-evaluation is an important part of project life cycle management.Scientific and reasonable pre-evaluation provides important reference for subsequent approval decisions.Based on the Net-Information Projects pre-evaluation norm system proposed by the research group of "Research on performance evaluation method of Net-Information Projects",the thesis establishes a pre-evaluation model of Net-Information Projects.The proposed model can automatically pre-evaluate and grade the declared Net-Information Projects,which not only avoids the subjectivity and non-repeatability of artificial evaluation,but also saves labor costs and greatly improves the objectivity and fairness.The proposed Net-Information Projects pre-evaluation model is based on text embedding.The specific implementation process is described as follows.First,pre-process the original input texts(i.e.,Net-Information Project declarations),including Chinese words segmentation and stopping words removal.Second,embed the processed texts into a latent vector space so that the text feature vectors are obtained.Third,transform the task of predicting pre-evaluation scores into a regression problem and train a nonlinear regression model to predict the pre-evaluation scores.The main research contents of the thesis are summarized as follows.First,the thesis improves four existing text embedding algorithms.These existing text embedding algorithms belong to unsupervised learning algorithms.The text feature vectors obtained by these algorithms only save the context structural and semantic information,so they cannot help reach a high accuracy rate in predicting pre-evaluation scores.In order to improve the quality of text embedding,this thesis employs the Siamese Neural Network to optimize these four text embedding algorithms.Text labels are added when training the model,so using the Siamese Neural Network to achieve the second text embedding belongs to supervised learning.Obtained text embedding vectors captures the latent semantic information in the labels,which is helpful to improve the accuracy of the pre-evaluation task.The superiority of the improved text embedding algorithm based on the Siamese Neural Network is proved by conducting experiments.Second,the thesis proposes a Net-Information Project pre-evaluation model based on text embedding.The model uses the improved text embedding algorithm to generate optimized the text feature vectors.Then the thesis employs a nonlinear regression model based on neural networks to build the nonlinear relation between the text feature vectors and the preevaluation scores.The nonlinear regression model can output the predicted pre-evaluation scores according to the input text feature vectors.At last,the proposed Net-Information Project pre-evaluation model is employed to predict the pre-evaluation scores in the realworld dataset.The pre-evaluation results prove the feasibility and effectiveness of the NetInformation Project pre-evaluation model.