Cost Evaluation Model Generation Method and Apparatus, and Cost Evaluation Method and Apparatus

Abstract

The present invention discloses a method and device for generating a cost evaluation model, as well as a cost evaluation method and device, including performing data enhancement and generation processing on data in each field in the cost evaluation of e-government informatization projects to obtain heterogeneous data in each field; after performing data matching and data alignment on the heterogeneous data in each field, using a weighted average algorithm for data fusion to generate a cost evaluation data set; decomposing the data in the cost evaluation data set into shared feature data and specific feature data, and using a feature generalization network for data generalization; dynamically fusing the shared feature data and specific feature data after data generalization to obtain a feature representation data set; using a pre-training method to construct and train a cost evaluation model for the e-government informatization project to obtain a trained cost evaluation model. Using this model for cost evaluation can improve the objectivity, accuracy, and credibility of the cost evaluation of e-government informatization projects.

Description

Technical Field

[0001] The present invention relates to the technical field of cost evaluation, and in particular, to a method and device for generating a cost evaluation model, as well as a method and device for cost evaluation. Background Art

[0002] The cost evaluation of e-government informatization projects (also known as the measurement of e-government informatization projects or the cost of e-government informatization projects) refers to the process of evaluating, accounting, and controlling the costs of various resources, equipment, manpower, and services required for a project during the implementation of an e-government informatization project. However, due to objective factors such as the personalized customization requirements of e-government informatization project construction, the fast product iteration speed, and the lack of a unified applicable reference basis, it has caused difficulties in data collection for the current cost evaluation of e-government informatization projects (the business processes and requirements of each e-government department are highly personalized, resulting in the need to customize corresponding data collection plans for each project during the data collection stage), a small amount of data retention (new functions and systems are constantly introduced, while old systems and data may be quickly phased out, lacking sufficient historical data as a reference during evaluation), and poor data quality (lacking unified standards and specifications, the data quality of different regions or departments varies, making it difficult to ensure the accuracy and reliability of the data). As a result, the cost evaluation of e-government informatization projects lacks objectivity, accuracy, and credibility. Summary of the Invention

[0003] The present invention provides a method and device for generating a cost evaluation model, as well as a method and device for cost evaluation. By using methods such as data generalization, fusion, and heterogeneity, a cost evaluation model for e-government informatization projects is constructed and trained, and this model is used for cost evaluation to improve the problems in the cost evaluation of e-government informatization projects, such as large data collection difficulties, small data retention, and poor data quality, and to enhance the objectivity, accuracy, and credibility of the cost evaluation of e-government informatization projects.

[0004] To achieve the above object, an embodiment of the present invention provides a method for generating a cost evaluation model, including:

[0005] Performing data enhancement and generation processing on the data in each field in the cost evaluation of e-government informatization projects to obtain heterogeneous data in each field;

[0006] After performing data matching and data alignment on the heterogeneous data in each field, using a weighted average algorithm to perform data fusion on the aligned heterogeneous data to generate the cost evaluation data set of the e-government informatization project;

[0007] Decomposing the data in the cost evaluation data set into shared feature data and specific feature data; using a feature generalization network to perform data generalization on the shared feature data and specific feature data;

[0008] Dynamically fuse the shared feature data and specific feature data after data generalization to obtain a feature representation data set; according to the feature representation data set, use a pre-training method to construct and train a cost evaluation model for the e-government informatization project to obtain a trained cost evaluation model.

[0009] As an improvement to the above solution, the use of a feature generalization network to perform data generalization on the shared feature data and specific feature data includes:

[0010] Use a feature generalization network to perform data generalization on the shared feature data and specific feature data;

[0011] Among them, the feature generalization network includes a feature network layer, a task network layer, and a feedback network layer;

[0012] Use the feature network layer to extract features from the shared feature data and specific feature data to obtain general features;

[0013] Based on the task network layer, map the general features to the corresponding e-government informatization projects to obtain a predicted mapping class probability distribution, and use a supervised classification loss function to measure the difference between the mapping class probability distribution and the true class of the e-government informatization project;

[0014] In the feedback network layer, use an auxiliary loss function to evaluate the general features and the mapping class probability distribution to update the parameters of the feature network layer and the task network layer.

[0015] As an improvement to the above solution, the decomposition of the data in the cost evaluation data set into shared feature data and specific feature data includes:

[0016] After using a tokenizer and the BPE algorithm to tokenize the data in the cost evaluation data set, perform feature extraction through a Word2Vec model to obtain the vector features of each word;

[0017] Use a self-attention mechanism to perform correlation calculations on the vector features to generate a feature matrix; perform weighted summation on the vocabulary in the cost evaluation data set according to the feature matrix to obtain a self-attention vector; decompose the self-attention vector into shared features and specific features;

[0018] Delete the shared features corresponding to the variance of the shared features less than a preset variance threshold, and delete the specific features corresponding to the variance of the specific features less than a preset variance threshold to obtain shared feature data and specific feature data.

[0019] As an improvement to the above solution, after performing data matching and data alignment on the heterogeneous data in each field, a weighted average algorithm is used to perform data fusion on the aligned heterogeneous data to generate the cost evaluation data set for the e-government informatization project, including:

[0020] Use a matching method of similarity and rules to match the heterogeneous data in each field, and filter out the generated low-quality data;

[0021] Use an embedding model to map the matched heterogeneous data in each field into the same low-dimensional vector space for data alignment;

[0022] Use a weighted average algorithm to perform weighted average calculation on all the aligned heterogeneous data to obtain the fused data, and perform smoothing and denoising processing on the fused data to generate the cost evaluation data set for the e-government informatization project.

[0023] As an improvement to the above solution, the dynamically fusing the shared feature data and the specific feature data after data generalization to obtain a feature representation data set, including:

[0024] Extract the general shared features in the shared feature data after data generalization, and obtain the weight vector for each field according to the general shared features;

[0025] Extract the feature representation for each field in the specific feature data after data generalization; perform fusion according to the weight vector and the feature representation for each field to obtain the feature representation data set.

[0026] As an improvement to the above solution, according to the feature representation data set, use a pre-training method to construct and train the cost evaluation model for the e-government informatization project to obtain a trained cost evaluation model, including:

[0027] Use the feature representation data set as input data and use a pre-training method to construct the cost evaluation model for the e-government informatization project;

[0028] Based on the cross-entropy loss function and the SGF optimization algorithm, use the feature representation data set to perform iterative training on the cost evaluation model to update the parameters of the cost evaluation model;

[0029] Use the cost evaluation data set to verify the updated cost evaluation model. If the verification result does not meet the expected effect, adjust the model parameters and hyperparameters according to the verification result;

[0030] Based on the cross-entropy loss function and the SGF optimization algorithm, the adjusted cost evaluation model is iteratively trained using the feature representation dataset, the parameters of the adjusted cost evaluation model are updated, and the updated cost evaluation model is verified using the cost evaluation dataset until the verification result reaches the expected effect, obtaining a trained cost evaluation model.

[0031] As an improvement of the above solution, in the cost evaluation of government affairs informatization projects, data enhancement and generation processing are performed on the data in each field to obtain heterogeneous data in each field, including:

[0032] Collect and sort various evaluation data affecting the cost of government affairs informatization projects, and classify the evaluation data by category to obtain the data in each field;

[0033] Perform data enhancement processing on the data in each field, and perform data generation processing on the data in each field in an adversarial manner according to the context information and the dependency relationship between the data in each field to obtain heterogeneous data in each field.

[0034] To achieve the above object, an embodiment of the present invention correspondingly provides a cost evaluation model generation device, including:

[0035] A heterogeneous data acquisition module, configured to perform data enhancement and generation processing on the data in each field in the cost evaluation of government affairs informatization projects to obtain heterogeneous data in each field;

[0036] An evaluation data generation module, configured to perform data matching and data alignment on the heterogeneous data in each field, and then perform data fusion on the aligned heterogeneous data using a weighted average algorithm to generate the cost evaluation dataset of the government affairs informatization project;

[0037] A feature data generalization module, configured to decompose the data in the cost evaluation dataset into shared feature data and specific feature data; perform data generalization on the shared feature data and specific feature data using a feature generalization network;

[0038] An evaluation model construction module, configured to dynamically fuse the data-generalized shared feature data and specific feature data to obtain a feature representation dataset; construct and train the cost evaluation model of the government affairs informatization project according to the feature representation dataset using a pre-training method to obtain a trained cost evaluation model.

[0039] To achieve the above object, an embodiment of the present invention further provides a cost evaluation method, including:

[0040] Input the evaluation data of the target e-government informatization project into the trained cost evaluation model, perform domain division and regional division on the evaluation data, determine the cost evaluation criteria for the evaluation data, and output the cost evaluation result of the target e-government informatization project;

[0041] Among them, the trained cost evaluation model is obtained by using the above-mentioned cost evaluation model generation method.

[0042] To achieve the above object, an embodiment of the present invention correspondingly provides a cost evaluation device, including:

[0043] An evaluation result acquisition module, configured to input the evaluation data of the target e-government informatization project into the trained cost evaluation model, perform domain division and regional division on the evaluation data, determine the cost evaluation criteria for the evaluation data, and output the cost evaluation result of the target e-government informatization project;

[0044] Among them, the trained cost evaluation model is obtained by using the above-mentioned cost evaluation model generation method.

[0045] Compared with the prior art, a cost evaluation model generation method and device and a cost evaluation method and device disclosed in an embodiment of the present invention perform data enhancement and generation processing on the data in each field in the cost evaluation of e-government informatization projects to obtain heterogeneous data in each field; after performing data matching and data alignment on the heterogeneous data in each field, a weighted average algorithm is used to perform data fusion on the aligned heterogeneous data to generate the cost evaluation data set of the e-government informatization project; the data in the cost evaluation data set is decomposed into shared feature data and specific feature data; a feature generalization network is used to perform data generalization on the shared feature data and specific feature data; the shared feature data and specific feature data after data generalization are dynamically fused to obtain a feature representation data set; according to the feature representation data set, a pre-training method is used to construct and train the cost evaluation model of the e-government informatization project to obtain a trained cost evaluation model. Using the trained cost evaluation model to perform cost evaluation according to the evaluation data of the target e-government informatization project can improve the problems in the cost evaluation of e-government informatization projects such as difficult data collection, data retention, and poor data quality, improve the objectivity, accuracy, and credibility of the cost evaluation of e-government informatization projects, and solve the problem that e-government informatization projects cannot be effectively evaluated in some regions due to the lack of current reference basis. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a schematic flowchart of a cost evaluation model generation method provided by an embodiment of the present invention;

[0047] Figure 2It is a block diagram for evaluating a data processing process provided by an embodiment of the present invention;

[0048] Figure 3 It is a schematic structural diagram of a device for generating a cost evaluation model provided by an embodiment of the present invention;

[0049] Figure 4 It is a schematic flowchart of a cost evaluation method provided by an embodiment of the present invention;

[0050] Figure 5 It is a schematic structural diagram of a cost evaluation device provided by an embodiment of the present invention. Detailed implementation manners

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention.

[0052] It should be noted that the terms "including" and "specific" in the present invention and any of their variations are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0053] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of a method for generating a cost evaluation model provided by an embodiment of the present invention. The method for generating a cost evaluation model includes:

[0054] S1. Perform data enhancement and generation processing on the data in each field in the cost evaluation of government affairs informatization projects to obtain heterogeneous data in each field;

[0055] S2. After performing data matching and data alignment on the heterogeneous data in each field, use the weighted average algorithm to perform data fusion on the aligned heterogeneous data to generate the cost evaluation data set of the government affairs informatization project;

[0056] S3. Decompose the data in the cost evaluation data set into shared feature data and specific feature data; use a feature generalization network to perform data generalization on the shared feature data and specific feature data;

[0057] S4. Dynamically fuse the shared feature data and specific feature data after data generalization to obtain a feature representation data set; according to the feature representation data set, use a pre-training method to construct and train the cost evaluation model of the e-government informatization project to obtain a trained cost evaluation model.

[0058] Exemplarily, collect and sort out various evaluation data affecting the cost of e-government informatization projects, perform data enhancement and generation on the cost evaluation data of e-government informatization projects in different fields (development, procurement, service, and basis), so as to obtain high-quality and diverse data, and then realize the expansion of the data set; then perform data matching and data alignment on the data, and adopt the method of knowledge fusion to fuse the heterogeneous data of the cost evaluation of e-government informatization projects from different fields, different formats, and different types to form a comprehensive and complete cost evaluation data set of e-government informatization projects; decompose the features in the data set into two parts: shared features and specific features by means of feature decoupling, and use a feature generalization network to perform feature generalization on the shared feature data set and specific feature data set after heterogeneous fusion, so that the data has generalization ability for the e-government informatization project field; finally, integrate all the shared features and heterogeneous features by means of ensemble learning, and use a pre-training method to construct and train the cost evaluation model of the e-government informatization project, so that the model has generalization and recognition ability, and finally improve the objectivity, accuracy, and credibility of the current cost evaluation of e-government informatization projects. And solve the problem that some regions are unable to effectively evaluate e-government informatization projects due to the lack of current reference basis.

[0059] Specifically, in the step S3, the use of the feature generalization network to perform data generalization on the shared feature data and specific feature data includes:

[0060] Use a feature generalization network to perform data generalization on the shared feature data and specific feature data;

[0061] Among them, the feature generalization network includes a feature network layer, a task network layer, and a feedback network layer;

[0062] Use the feature network layer to extract features from the shared feature data and specific feature data to obtain general features;

[0063] Based on the task network layer, map the general features to the corresponding e-government informatization projects to obtain the predicted mapping class probability distribution, and use a supervised classification loss function to measure the difference between the mapping class probability distribution and the true class of the e-government informatization project;

[0064] In the feedback network layer, use an auxiliary loss function to evaluate the general features and the mapping class probability distribution to update the parameters of the feature network layer and the task network layer.

[0065] Exemplarily, the generalization ability of the shared feature data and specific feature data for unknown targets or tasks is achieved by constructing a feature generalization network, which consists of three parts, including a feature network layer responsible for extracting features from the input data and using these features for subsequent tasks; a task network layer that generalizes data based on the features extracted by the feature network. It provides a supervised classification loss to guide the training of the feature network; a feedback network layer that learns an auxiliary loss function, which aims to encourage the feedback network to extract enhanced losses and provide additional feedback to the feature network, thus supporting more complex domain generalization tasks. For example, a pre-trained CNN is used as the input end of the feature generalization network, and the first four convolutional blocks are frozen, and only the remaining convolutional blocks and the average pooling layer are updated to improve computational efficiency. This module is mainly responsible for extracting general features from the input shared features and specific features so as to be applicable to different e-government informatization projects. One or more fully connected layers are connected after the feature network layer as the task network layer to map the output of the feature network layer to specific categories of e-government informatization projects. At the same time, a supervised classification loss function is set during the mapping process to measure the difference loss between the output and the actual e-government informatization project. By minimizing the difference loss, the accuracy of mapping data to the correct category can be improved. That is, for each training sample, first, general features are extracted from the shared features and specific features in the feature network layer, then the task network layer maps these features to the corresponding e-government informatization project categories and outputs the predicted probability, and the supervised classification loss is used to measure the difference between this predicted probability distribution and the true category. The feedback network layer introduces an auxiliary evaluation loss function through a multi-layer perceptron to evaluate the quality of the output features of the feature network layer and the task network layer, and then optimizes the generalization ability of the features through continuous iteration. Specifically, the quality of the features processed by the feature network layer and the task network layer can be evaluated by comparing with and without using the auxiliary evaluation loss function to generate a meta-loss, and then this meta-loss is used to update the parameters of the feature network layer and the task network layer, thereby guiding the feature network layer and the task network layer to generate features more conducive to generalization.

[0066] It is worth noting that domain generalization is an important concept in machine learning and computer vision. It refers to the ability of a model to successfully apply the knowledge or patterns learned from one or more source domains to a target domain that was not seen during training, even if there are distribution differences between these domains. Simply put, domain generalization hopes that the model can "draw inferences from one instance", not only perform well on a specific dataset or in a specific environment, but also maintain stable performance across different environments or datasets.

[0067] Specifically, in step S3, the decomposition of the data in the cost evaluation dataset into shared feature data and specific feature data includes:

[0068] After tokenizing the data in the cost evaluation dataset using a tokenizer and the BPE algorithm, feature extraction is performed through the Word2Vec model to obtain the vector features of each word;

[0069] Using the self-attention mechanism, correlation calculations are performed on the vector features to generate a feature matrix; the words in the cost evaluation dataset are weighted and summed according to the feature matrix to obtain a self-attention vector; the self-attention vector is decomposed into shared features and specific features;

[0070] Delete the shared features corresponding to the variance of the shared features being less than a preset variance threshold, and delete the specific features corresponding to the variance of the specific features being less than a preset variance threshold to obtain shared feature data and specific feature data.

[0071] Exemplarily, by learning a function to map feature vectors into a set with different functional parts, so that each part only contains information related to itself, that is, decomposing the data features in the cost evaluation dataset into shared features and specific features. For example, use SentencePiece, a customized tokenizer for feature extraction, to tokenize the text, and at the same time use the byte-level Byte Pair Encoding (BPE) algorithm to ensure that the information after tokenization is not lost. Use the Word2Vec model to extract features from the cost evaluation dataset to obtain the vector features of each word. Use the self-attention mechanism to process the extracted vector features to perform correlation calculations on the data features in the field of government affairs informatization projects, and then generate a feature matrix. According to the feature matrix, perform weighted summation on the vocabulary in the cost evaluation dataset to generate a self-attention vector, which contains all the information in the field of government affairs informatization projects, that is, decomposing the data in the field of government affairs informatization projects after heterogeneous fusion into multiple sub-feature vectors. Finally, through the UndoBias method, learn a general weight for each field of all government affairs informatization projects and the weight increment for each field based on this , then the field weight of each field is expressed as , using maximum margin classification to learn the combined relationship of the above weights, the shared features and specific features can be decomposed. Screen by calculating the variance of each feature. Features with too small variance indicate that the samples have basically no difference in this feature, and may have the same value for most values or the same value for the entire feature. Such features have little effect on subsequent steps. The embodiment of the present invention adopts the VarianceThreshold method, and deletes all features less than the threshold by setting a threshold

[0072] Specifically, in the step S2, it includes:

[0073] S21, using the matching method of similarity and rules to match the heterogeneous data in each field, and filtering out the generated low-quality data;

[0074] S22, using an embedding model to map the heterogeneous data in each field after matching into the same low-dimensional vector space for data alignment;

[0075] S23, using the weighted average algorithm to perform weighted average calculation on all the aligned heterogeneous data to obtain the fused data, and performing smoothing and denoising processing on the fused data to generate the cost evaluation dataset of the government affairs informatization project

[0076] Exemplarily, data matching and data alignment are respectively performed on heterogeneous data in each field to ensure correct association and integration during the fusion process. Then, the heterogeneous data in different fields are subjected to average weighting processing through the weighted average algorithm to achieve the fusion of heterogeneous data in each field, forming a comprehensive and complete cost evaluation data set for e-government informatization projects. For example, on the one hand, a similarity-based matching method (cosine similarity) is adopted, that is, the semantic similarity of texts is compared by measuring the cosine value of the angle between two text vectors in a multi-dimensional space, thereby realizing data alignment within the field. On the other hand, a rule-based matching method is adopted, that is, a predefined rule set is used to identify, classify, and filter data. Data matching is performed on the evaluation basis in the data set, such as document data such as management measures and implementation rules, and low-quality data generated by data operations is filtered. Data alignment aims to correspond and integrate the data within the field after data operations to ensure data consistency and accuracy. In this embodiment, the TransE model is used to train the heterogeneous data in each field after data matching, so that the heterogeneous data in each field are mapped to the same low-dimensional vector space, and then a similarity threshold (which can be set as needed) is set. When the similarity of the data exceeds the similarity threshold, it can be considered that they may represent the same object or content, thereby realizing data alignment within each field. In this embodiment, the weighted average algorithm is used to fuse the data, that is, the data in different data sources or data sets are weighted and averaged according to a certain weight to obtain the fused data result. First, an adaptive algorithm is used to dynamically adjust the weight. This method can adjust the weight in real time according to the quality of the data itself or the fusion effect to achieve a better fusion effect. Secondly, weighted average calculation is performed on all data sets in each field, and the formula is:

[0077]

[0078] wherein, is the weight of the th data source, is the data value of the th data source, is the number of data sources, and the fused data result is obtained. Finally, the fused data are respectively subjected to smoothing processing (removing noise and outliers in the fusion result) and denoising (further reducing the noise level in the fusion result through filtering technology) to improve the usability and accuracy of the data, and the fusion of heterogeneous data between different fields is formed into a comprehensive and complete cost evaluation data set for e-government informatization projects.

[0079] Specifically, in the step S4, the dynamic fusion of the shared feature data and specific feature data after data generalization to obtain a feature representation data set includes:

[0080] Extract the general shared features in the shared feature data after data generalization, and obtain the weight vector for each domain according to the general shared features;

[0081] Extract the feature representation for each domain in the specific feature data after data generalization; fuse according to the weight vector and feature representation for each domain to obtain a feature representation dataset.

[0082] Exemplarily, by designing a reasonable feature layer and selecting a suitable extractor, construct an ensemble learning network for feature fusion; the ensemble learning network used in this embodiment includes the following three parts: the shared CNN feature extractor is responsible for extracting general features from the input data; the specific feature extraction layer, for each source domain, there is a corresponding feature extraction layer responsible for feature representation for unseen targets; the domain weight branch, which receives the output of the CNN feature extractor and learns a weight vector for measuring the contribution degree of the target feature representation of each source domain; use the weight output by the domain weight branch to fuse the features from different specific branches. Specifically, each domain feature is multiplied by the corresponding weight and then aggregated (such as weighted average). Since it is based on the weights of the learned shared features and specific features and can automatically adjust the importance of the features of different types of e-government informatization project data, the feature representation of unknown targets can be constructed.

[0083] Specifically, in the step S4, constructing and training the cost evaluation model of the e-government informatization project by using the pre-training method according to the feature representation dataset to obtain a trained cost evaluation model includes:

[0084] Use the feature representation dataset as input data and construct the cost evaluation model of the e-government informatization project by using the pre-training method;

[0085] Based on the cross-entropy loss function and the SGF optimization algorithm, use the feature representation dataset to iteratively train the cost evaluation model and update the parameters of the cost evaluation model;

[0086] Use the cost evaluation dataset to verify the updated cost evaluation model. If the verification result does not meet the expected effect, adjust the model parameters and hyperparameters according to the verification result;

[0087] Then, based on the cross-entropy loss function and the SGF optimization algorithm, use the feature representation dataset to iteratively train the adjusted cost evaluation model, update the parameters of the adjusted cost evaluation model, and use the cost evaluation dataset to verify the updated cost evaluation model until the verification result meets the expected effect to obtain a trained cost evaluation model.

[0088] Exemplarily, the fused feature representation is used as input data and input into the Transformer architecture, and a feature fusion layer and an embedding layer are set therein for combining the received fused features with the output of the embedding layer; the Cross-Entropy Loss function is selected as the loss function to measure the difference between the content generated by the model and the real content; then, the cost evaluation data set is used for training, the gradient is calculated through the backpropagation algorithm, and the SGF optimization algorithm is used to update the model parameters to minimize the loss function; finally, heterogeneous data is used as the validation set to validate the model, the model parameters and hyperparameters are adjusted according to the validation results, and through the iterative training and optimization process until the content generated by the model reaches the expected result.

[0089] In the embodiment of the present invention, by designing the network structure and adopting the method of ensemble learning, the shared features and specific feature data after feature generalization are dynamically fused, and the data set is applied to the training of the model to improve the adaptive evaluation ability of the model, so as to realize the effective evaluation of the cost evaluation of government affairs informatization projects under the conditions of difficult data collection, large data retention, and poor data quality.

[0090] Specifically, in the step S1, it includes:

[0091] S11, collecting and sorting out various evaluation data affecting the cost of government affairs informatization projects, and classifying the evaluation data by category to obtain the data in each field;

[0092] S12, performing data augmentation processing on the data in each field, and performing data generation processing on the data in each field in an adversarial manner according to the context information and the dependency relationship between the data in each field to obtain heterogeneous data in each field.

[0093] Exemplarily, Figure 2 as shown, Figure 2It is a block diagram for evaluating the data processing process provided by an embodiment of the present invention. It collects and organizes various types of evaluation data (development, procurement, service, basis) that affect the cost of e-government informatization projects, classifies the evaluation data by category to obtain data for each field. The evaluation data used in this embodiment are the main factors affecting the cost evaluation of e-government informatization projects, including the content of informatization projects (software development, system operation and maintenance, project operation, hardware acquisition, customized equipment, etc.), the e-government project cost guidance manuals of each province and city, the salary guidance prices of the human resources market in each region over the years, the cost evaluation methods, etc. At the same time, the existing e-government informatization project cost evaluation data are divided into fields. E-government informatization projects usually cover multiple fields, such as the construction of e-government platforms, smart city systems, public security management systems, etc. Each field has its specific project characteristics, cost composition, and evaluation criteria. Through field division, the data in these different fields can be classified to facilitate more targeted data processing and analysis in the follow-up. Data enhancement is performed on the data for each field to increase the randomness and variability of the field data; TF-IDF is used to extract keywords or key phrases (which may represent the theme or context of the dataset) from the dataset, and then the obtained context information is converted into numerical vectors through Word2Vec and added back to the original dataset; GAN (Generative Adversarial Networks) is a deep learning model composed of two neural networks - a generator and a discriminator. These two networks compete with each other during the training process to form a competitive relationship, thereby improving the quality of the generated data. In this example, SeqGAN is selected as the model for data generation. First, the enhanced dataset is input into the generator, and the generator will generate simulated data (false samples) for the adversarial network according to the input data; then the dataset and the simulated data are input into the discriminator together, and the strategy of alternating maximum and minimum optimization is adopted for the iteration of SqeGAN to make the generator and the discriminator compete with each other, and finally a high-quality dataset is generated.

[0094] More specifically, the data enhancement processing for the data in each field includes:

[0095] Perform data enhancement processing of text simplification, text retelling, and noise injection on the data in each field.

[0096] Exemplarily, the source data is processed by some open-source text simplification models and manual editing methods to simplify the original complex text into an easier-to-understand version, and then added to the dataset to increase the data dimension; by rephrasing the content of the original text, new text instances are generated, thereby increasing the corpus size and implicit information in a specific scenario without adding extra information, thus enriching the connotation and extension of the dataset; noise injection, as a data augmentation method, has significant effects in enhancing the robustness of the model, preventing overfitting, improving generalization ability, and enhancing the stability of the model. In this embodiment, Gaussian noise is selected as the injected noise, and a random noise matrix matching the shape of the dataset is generated through a tool (such as the random number generation library numpy in Python), and then the generated noise matrix is added to the dataset.

[0097] A method for generating a cost evaluation model disclosed in an embodiment of the present invention includes: performing data augmentation and generation processing on the data in each field in the cost evaluation of e-government informatization projects to obtain heterogeneous data in each field; after performing data matching and data alignment on the heterogeneous data in each field, using a weighted average algorithm to perform data fusion on the aligned heterogeneous data to generate the cost evaluation dataset of the e-government informatization project; decomposing the data in the cost evaluation dataset into shared feature data and specific feature data; using a feature generalization network to perform data generalization on the shared feature data and specific feature data; dynamically fusing the data-generalized shared feature data and specific feature data to obtain a feature representation dataset; according to the feature representation dataset, using a pre-training method to construct and train the cost evaluation model of the e-government informatization project to obtain a trained cost evaluation model. Using the trained cost evaluation model to perform cost evaluation according to the evaluation data of the target e-government informatization project can improve the problems in the cost evaluation of e-government informatization projects, such as difficult data collection, low data retention, and poor data quality, improve the objectivity, accuracy, and credibility of the cost evaluation of e-government informatization projects, and solve the problem that e-government informatization projects cannot be effectively evaluated in some regions due to the lack of current reference basis.

[0098] Figure 3 FIG. is a schematic structural diagram of a cost evaluation model generation device provided by an embodiment of the present invention. The cost evaluation model generation device 10 includes:

[0099] A heterogeneous data acquisition module 11, configured to perform data augmentation and generation processing on the data in each field in the cost evaluation of e-government informatization projects to obtain heterogeneous data in each field;

[0100] The evaluation data generation module 12 is used to perform data matching and data alignment on heterogeneous data in each field, and then perform data fusion on the aligned heterogeneous data using the weighted average algorithm to generate the cost evaluation data set of the government affairs informatization project;

[0101] The feature data generalization module 13 is used to decompose the data in the cost evaluation data set into shared feature data and specific feature data; and use the feature generalization network to perform data generalization on the shared feature data and specific feature data;

[0102] The evaluation model construction module 14 is used to dynamically fuse the shared feature data and specific feature data after data generalization to obtain a feature representation data set; and construct and train the cost evaluation model of the government affairs informatization project using the pre-training method based on the feature representation data set to obtain a trained cost evaluation model.

[0103] Specifically, the using the feature generalization network to perform data generalization on the shared feature data and specific feature data includes:

[0104] Using the feature generalization network to perform data generalization on the shared feature data and specific feature data,

[0105] Wherein, the feature generalization network includes a feature network layer, a task network layer and a feedback network layer;

[0106] Using the feature network layer to perform feature extraction on the shared feature data and specific feature data to obtain general features;

[0107] Based on the task network layer, mapping the general features to the corresponding government affairs informatization projects to obtain a predicted mapping class probability distribution, and using a supervised classification loss function to measure the difference between the mapping class probability distribution and the true class of the government affairs informatization project;

[0108] In the feedback network layer, using an auxiliary loss function to evaluate the general features and the mapping class probability distribution to update the parameters of the feature network layer and the task network layer.

[0109] Specifically, the decomposing the data in the cost evaluation data set into shared feature data and specific feature data includes:

[0110] After using a tokenizer and the BPE algorithm to tokenize the data in the cost evaluation data set, performing feature extraction through the Word2Vec model to obtain the vector features of each word;

[0111] Using the self-attention mechanism, perform correlation calculations on the vector features to generate a feature matrix; perform weighted summation on the vocabulary in the cost evaluation dataset according to the feature matrix to obtain a self-attention vector; decompose the self-attention vector into shared features and specific features;

[0112] Delete the shared features corresponding to the variance of the shared features less than the preset variance threshold, and delete the specific features corresponding to the variance of the specific features less than the preset variance threshold, to obtain shared feature data and specific feature data.

[0113] Specifically, the evaluation data generation module 12 is used for:

[0114] Adopt a similarity and rule matching method to match the heterogeneous data in each field, and filter out the generated low-quality data;

[0115] Adopt an embedding model to map the matched heterogeneous data in each field into the same low-dimensional vector space for data alignment;

[0116] Adopt a weighted average algorithm to perform weighted average calculation on all the aligned heterogeneous data, obtain the fused data, and perform smoothing and denoising processing on the fused data to generate the cost evaluation dataset of the government affairs informatization project.

[0117] Specifically, the dynamic fusion of the shared feature data and specific feature data after data generalization to obtain a feature representation dataset includes:

[0118] Extract the general shared features in the shared feature data after data generalization, and obtain the weight vector of each field according to the general shared features;

[0119] Extract the feature representation of each field in the specific feature data after data generalization; perform fusion according to the weight vector and feature representation of each field to obtain a feature representation dataset.

[0120] Specifically, according to the feature representation dataset, adopt a pre-training method to construct and train the cost evaluation model of the government affairs informatization project to obtain a trained cost evaluation model, including:

[0121] Use the feature representation dataset as input data and adopt a pre-training method to construct the cost evaluation model of the government affairs informatization project;

[0122] Based on the cross-entropy loss function and the SGF optimization algorithm, use the feature representation dataset to iteratively train the cost evaluation model and update the parameters of the cost evaluation model;

[0123] Validate the updated cost evaluation model using the cost evaluation data set. If the validation result does not meet the expected effect, adjust the model parameters and hyperparameters according to the validation result;

[0124] Then, based on the cross-entropy loss function and the SGF optimization algorithm, use the feature representation data set to iteratively train the adjusted cost evaluation model, update the parameters of the adjusted cost evaluation model, and use the cost evaluation data set to validate the updated cost evaluation model until the validation result meets the expected effect, obtaining a trained cost evaluation model.

[0125] Specifically, the heterogeneous data acquisition module 11 is used for:

[0126] Collect and sort out various evaluation data affecting the cost of e-government informatization projects, classify the evaluation data by category to obtain data for each field;

[0127] Perform data augmentation processing on the data for each field, and generate data for each field in an adversarial manner according to the context information and the dependency relationship between the data for each field, obtaining heterogeneous data for each field.

[0128] The cost evaluation model generation device provided by the embodiments of the present invention can implement all the processes of the cost evaluation model generation method in the above embodiments. The functions of the various modules in the device and the achieved technical effects are respectively the same as the functions and the achieved technical effects of the cost evaluation model generation method in the above embodiments, and will not be elaborated here.

[0129] Figure 4 It is a schematic flowchart of a cost evaluation method provided by an embodiment of the present invention. The cost evaluation method includes:

[0130] S101, input the evaluation data of the target e-government informatization project into the trained cost evaluation model, perform field division and regional division on the evaluation data, determine the cost evaluation criteria for the evaluation data, and output the cost evaluation result of the target e-government informatization project;

[0131] Among them, the trained cost evaluation model is obtained by using the cost evaluation model generation method as described above.

[0132] Exemplarily, the trained cost evaluation model has integrated the generalization ability of e-government informatization project data. Therefore, it can conduct cost evaluation for e-government informatization projects under the circumstances of difficult data collection, low data retention, and poor data quality. Input the content of an informatization project in a certain city into the model, which includes software part, operation and maintenance, operation, and personalized equipment, etc., and let the model conduct cost evaluation. First, since the city has not set its own exclusive cost evaluation standard for its region, the model will refer to the cost evaluation standards of other cities in the province where the city is located to generalize and form the cost evaluation standard for the city and conduct evaluation based on this. Then, the software part will be evaluated according to the estimation method (pre-estimation method or estimation method) adopted for the software cost in the city in the past. If there is no past data of the city, it will be generalized to the commonly used estimation method in the province for evaluation; for the costs of operation and maintenance and operation, they will be evaluated according to the wage guidance price of the human resources market in the city in the current year or the previous year and combined with the calculation method of the labor rate in the e-government project cost guidance book. If the wage guidance price of the human resources market in the city in recent years is lacking, then it will be generalized according to the fluctuation range of the wage guidance price of this position in the province in recent years to obtain the cost; the hardware cost will be evaluated according to the similarity degree of equipment parameters. Because the equipment models of different manufacturers are different, but the functions of their equipment can be compared by comparing parameters, the model will also give an evaluation of the hardware cost according to the similarity degree of parameters; for some personalized equipment, the model will evaluate according to the content of some of its modules and combine with equipment with similar functions in other provinces and cities, and then obtain a complete and appropriate cost evaluation report for the informatization project.

[0133] A cost evaluation method disclosed in an embodiment of the present invention inputs the evaluation data of a target e-government informatization project into a trained cost evaluation model, conducts domain division and regional division on the evaluation data, determines the cost evaluation standard of the evaluation data, and outputs the cost evaluation result of the target e-government informatization project; wherein, the trained cost evaluation model is obtained by using the cost evaluation model generation method as described above. Using the trained cost evaluation model to conduct cost evaluation according to the evaluation data of the target e-government informatization project can improve the problems of e-government informatization project cost evaluation in aspects such as difficult data collection, low data retention, and poor data quality, improve the objectivity, accuracy, and credibility of e-government informatization project cost evaluation, and solve the problem that some regions cannot effectively evaluate e-government informatization projects due to the lack of current reference basis.

[0134] Figure 5 It is a schematic structural diagram of a cost evaluation device provided by an embodiment of the present invention. The cost evaluation device 100 includes:

[0135] An evaluation result acquisition module 101 is configured to input the evaluation data of a target government affairs informatization project into a trained cost evaluation model, perform domain division and regional division on the evaluation data, determine the cost evaluation criteria of the evaluation data, and output the cost evaluation result of the target government affairs informatization project;

[0136] Wherein, the trained cost evaluation model is obtained by using the cost evaluation model generation method as described above.

[0137] The cost evaluation device provided by an embodiment of the present invention can implement all processes of the cost evaluation model generation method in the above embodiment. The functions and achieved technical effects of each module in the device are respectively the same as those of the cost evaluation method in the above embodiment, and will not be elaborated here.

[0138] The above is the preferred implementation manner of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A method for generating a cost evaluation model, characterized in that, Including: Performing data enhancement and generation processing on the data in each field of the cost assessment of e-government informatization projects to obtain heterogeneous data in each field; After performing data matching and data alignment on the heterogeneous data in each field, using a weighted average algorithm to perform data fusion on the aligned heterogeneous data to generate the cost assessment data set of the e-government informatization project; Decomposing the data in the cost assessment data set into shared feature data and specific feature data; Using a feature generalization network to perform data generalization on the shared feature data and specific feature data; Performing dynamic fusion on the data-generalized shared feature data and specific feature data to obtain a feature representation data set; According to the feature representation data set, using a pre-training method to construct and train the cost assessment model of the e-government informatization project to obtain a trained cost assessment model; Among them, the decomposing the data in the cost assessment data set into shared feature data and specific feature data includes: After performing word segmentation on the data in the cost assessment data set using a word segmenter and the BPE algorithm, performing feature extraction through a Word2Vec model to obtain the vector feature of each word; Using a self-attention mechanism to perform correlation calculation on the vector features to generate a feature matrix; Performing weighted summation on the vocabulary in the cost assessment data set according to the feature matrix to obtain a self-attention vector; Decomposing the self-attention vector into shared features and specific features; Deleting the shared features corresponding to the variance of the shared features less than a preset variance threshold, and deleting the specific features corresponding to the variance of the specific features less than the preset variance threshold to obtain shared feature data and specific feature data.

2. The method for generating a cost evaluation model according to claim 1, wherein, The using a feature generalization network to perform data generalization on the shared feature data and specific feature data includes: Using a feature generalization network to perform data generalization on the shared feature data and specific feature data; Among them, the feature generalization network includes a feature network layer, a task network layer, and a feedback network layer; Using the feature network layer to perform feature extraction on the shared feature data and specific feature data to obtain general features; Based on the task network layer, mapping the general features to the corresponding e-government informatization projects to obtain the predicted mapping category probability distribution, and using a supervised classification loss function to measure the difference between the mapping category probability distribution and the true category of the e-government informatization project; In the feedback network layer, using an auxiliary loss function to evaluate the general features and the mapping category probability distribution to update the parameters of the feature network layer and the task network layer.

3. The method for generating a cost evaluation model according to claim 1, wherein The after performing data matching and data alignment on the heterogeneous data in each field, using a weighted average algorithm to perform data fusion on the aligned heterogeneous data to generate the cost assessment data set of the e-government informatization project includes: Using a matching method of similarity and rules to match the heterogeneous data in each field, and filtering out the generated low-quality data; Using an embedding model to map the matched heterogeneous data in each field into the same low-dimensional vector space for data alignment; Use the weighted average algorithm to perform weighted average calculation on all heterogeneous data after alignment to obtain the fused data, and perform smoothing and denoising processing on the fused data to generate the cost evaluation data set of the e-government informatization project.

4. The method for generating a cost evaluation model according to claim 1, wherein Dynamically fuse the shared feature data and specific feature data after data generalization to obtain a feature representation data set, including: Extract the general shared features in the shared feature data after data generalization, and obtain the weight vector of each domain according to the general shared features; Extract the feature representation of each domain in the specific feature data after data generalization; perform fusion according to the weight vector and feature representation of each domain to obtain a feature representation data set.

5. The method for generating a cost evaluation model according to claim 1, wherein According to the feature representation data set, use the pre-training method to construct and train the cost evaluation model of the e-government informatization project to obtain a trained cost evaluation model, including: Use the feature representation data set as input data and use the pre-training method to construct the cost evaluation model of the e-government informatization project; Based on the cross-entropy loss function and the SGF optimization algorithm, use the feature representation data set to iteratively train the cost evaluation model and update the parameters of the cost evaluation model; Use the cost evaluation data set to verify the updated cost evaluation model. If the verification result does not meet the expected effect, adjust the model parameters and hyperparameters according to the verification result; Then, based on the cross-entropy loss function and the SGF optimization algorithm, use the feature representation data set to iteratively train the adjusted cost evaluation model, update the parameters of the adjusted cost evaluation model, and use the cost evaluation data set to verify the updated cost evaluation model until the verification result meets the expected effect to obtain a trained cost evaluation model.

6. The method for generating a cost evaluation model according to claim 1, wherein, Perform data augmentation and generation processing on the data in each domain in the cost evaluation of the e-government informatization project to obtain heterogeneous data in each domain, including: Collect and sort out various evaluation data affecting the cost of e-government informatization projects, and classify the evaluation data by category to obtain the data of each domain; Perform data augmentation processing on the data of each domain, and perform data generation processing on the data in each domain in an adversarial manner according to the context information and the dependency relationship between the data of each domain to obtain heterogeneous data in each domain.

7. A cost evaluation model generation device, characterized in that, Including: A heterogeneous data acquisition module for performing data augmentation and generation processing on the data in each domain in the cost evaluation of the e-government informatization project to obtain heterogeneous data in each domain; An evaluation data generation module for performing data matching and data alignment on the heterogeneous data in each domain, and using the weighted average algorithm to perform data fusion on the aligned heterogeneous data to generate the cost evaluation data set of the e-government informatization project; A feature data generalization module for decomposing the data in the cost evaluation data set into shared feature data and specific feature data; Use a feature generalization network to perform data generalization on the shared feature data and specific feature data; An evaluation model construction module, configured to dynamically fuse the shared feature data and specific feature data after data generalization to obtain a feature representation data set; according to the feature representation data set, construct and train a cost evaluation model for the e-government informatization project by using a pre-training method to obtain a trained cost evaluation model; Wherein, decomposing the data in the cost evaluation data set into shared feature data and specific feature data includes: After segmenting the data in the cost evaluation data set by using a tokenizer and the BPE algorithm, performing feature extraction through a Word2Vec model to obtain vector features of each word; Using a self-attention mechanism to perform correlation calculation on the vector features to generate a feature matrix; performing weighted summation on the vocabulary in the cost evaluation data set according to the feature matrix to obtain a self-attention vector; decomposing the self-attention vector into shared features and specific features; Deleting the shared features corresponding to the variance of the shared features being less than a preset variance threshold, and deleting the specific features corresponding to the variance of the specific features being less than a preset variance threshold to obtain shared feature data and specific feature data.

8. A cost evaluation method, characterized in that, Including: Inputting the evaluation data of the target e-government informatization project into the trained cost evaluation model, performing domain division and regional division on the evaluation data, determining the cost evaluation criteria for the evaluation data, and outputting the cost evaluation result of the target e-government informatization project; Wherein, the trained cost evaluation model is obtained by using the cost evaluation model generation method described in any one of claims 1-6.

9. A cost evaluation device, characterized in that Including: An evaluation result acquisition module, configured to input the evaluation data of the target e-government informatization project into the trained cost evaluation model, perform domain division and regional division on the evaluation data, determine the cost evaluation criteria for the evaluation data, and output the cost evaluation result of the target e-government informatization project; Wherein, the trained cost evaluation model is obtained by using the cost evaluation model generation method described in any one of claims 1-6.

Images