Intelligent project evaluation method supporting custom standard and multi-dimensional semantic analysis

Abstract

The invention discloses an intelligent project evaluation method supporting a custom standard and multi-dimensional semantic analysis. The method comprises the following steps: constructing an overall standard system according to a project evaluation standard; data arrangement, corpus cleaning, automatic sample generation, sample generation, index evaluation model design and model training are carried out on the basis of collected enterprise historical project materials, finally, an index evaluation model library is generated, and the index evaluation model library is managed in a mode of combining product life cycle management (PLM) and language model (LLM) so as to construct an index evaluation model; performing qualitative evaluation and quantitative evaluation on the enterprise project review material; and comprehensively considering the constructed evaluation standard system, the index evaluation model and the report content quality evaluation result, and intelligently generating a quality evaluation report. The method has the remarkable beneficial effects of improving the evaluation efficiency, reducing the cost, enhancing the evaluation accuracy and objectivity and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of custom standards and semantic analysis evaluation, and in particular to a project intelligent evaluation method supporting custom standards and multi-dimensional semantic analysis. Background Art

[0002] The quality review and management of enterprise projects is of great significance to ensure the long-term success and sustainable development of enterprises. Especially for large state-owned enterprises, this link is a key factor in building an excellent corporate image and ensuring the smooth operation of the national economy. Taking my country's State Grid Corporation as an example, a high-quality supervision and management system is not only an important part of its internal management, but also an important way for it to demonstrate its responsibility and commitment in society.

[0003] In terms of project quality evaluation, state-owned enterprises such as the State Grid Corporation of China have conducted in-depth exploration and innovation, and gradually formed a scientific and systematic evaluation system. This system, on the basis of strictly following relevant national laws, regulations and industry standards, fully combines the characteristics of the enterprise itself and establishes a classification management strategy. Specifically, it is to formulate corresponding evaluation standards and processes in a targeted manner according to different types of projects to ensure that each type of project can be scientifically and reasonably supervised and managed.

[0004] Taking the State Grid's Social Responsibility Rooting Project as an example, the State Grid Corporation of China has formulated a series of guiding documents, including the "Guiding Opinions on the State Grid's Social Responsibility Rooting Project System". These documents not only clarify the importance and goals of the corporate social responsibility rooting project, but also provide detailed evaluation standards and operating guidelines. These standards and guidelines not only reflect the State Grid Corporation of China's high standards for social responsibility projects, but also demonstrate its determination and attitude to actively fulfill its social responsibilities and serve the society. Under the guidance of these documents, the State Grid Corporation of China has continuously optimized the project quality supervision and management process, which has not only improved the internal management efficiency of the enterprise, but also ensured the high-quality implementation of the project through a scientific and rigorous evaluation system, providing high-quality and efficient services to the country and society.

[0005] However, although the traditional enterprise project quality evaluation method has certain historical value and practical basis, it mainly relies on manual review and evaluation reports, and has obvious limitations in efficiency, accuracy and coverage. First, the traditional method is inefficient when processing large-scale data and cannot meet the needs of rapid evaluation. Secondly, due to the reliance on manual evaluation, not only is the labor and time cost high, but the evaluation process is easily affected by personal subjective tendencies, resulting in the inability to guarantee the consistency and accuracy of the evaluation results. In addition, with the increasing number of various types of enterprise projects, traditional evaluation methods have become increasingly difficult to meet the needs of actual evaluation work.

[0006] In summary, existing methods have obvious deficiencies in efficiency, accuracy, and coverage, and are difficult to adapt to the processing needs of large-scale data and the actual needs of enterprises to quickly and accurately evaluate projects. Summary of the invention

[0007] In view of this, the present invention proposes a project intelligent evaluation method that supports customized standards and multi-dimensional semantic analysis to solve the problems of inefficiency, subjectivity and incomplete coverage of project quality review methods in the prior art.

[0008] The specific technical solutions of the present invention are as follows:

[0009] A project intelligent evaluation method supporting customized standards and multi-dimensional semantic analysis, comprising:

[0010] Constructing an overall standard system based on the project evaluation standards, including customizing the project evaluation standards, decomposing the evaluation standards according to project types or regions, and semantically parsing the evaluation standards to construct an evaluation index system, which covers the economic benefit indicators, social benefit indicators, and technical indicators of the project;

[0011] Based on the collected historical project materials of the enterprise, data collation, corpus cleaning and automatic sample generation, sample generation, indicator evaluation model design, and model training are performed, and finally an indicator evaluation model library is generated. The indicator evaluation model library is managed in a combination of product lifecycle management PLM and language model LLM to construct an indicator evaluation model;

[0012] Understand the content of enterprise project review materials, intelligently analyze the material content, match the evaluation standards with the material content through cognitive computing, conduct qualitative and quantitative evaluations to calculate the degree of completion and quality evaluation, and thus complete the report content quality evaluation;

[0013] Comprehensively consider the constructed evaluation standard system, indicator evaluation model and the results of report content quality assessment, make a quality assessment of the overall work by setting algorithms and rules, and intelligently generate a quality assessment report.

[0014] Specifically, in the step of semantic analysis of evaluation criteria in constructing the overall standard system, the evaluation criteria in the form of natural language are analyzed into quantifiable and operational evaluation indicators through multi-dimensional semantic analysis technology.

[0015] Specifically, in the sample generation step of constructing the indicator evaluation model, in addition to generating samples from historical project materials, supplementary samples are generated through data simulation and expert experience according to the requirements of the evaluation indicator system to improve the generalization ability of the model.

[0016] Specifically, in the step of designing an indicator evaluation model in constructing an indicator evaluation model, a machine learning algorithm is selected as the basic model framework based on the constructed evaluation indicator system and the generated samples, and the model structure is designed according to the characteristics of the project and the distribution of the data.

[0017] Specifically, in the step of understanding the content of enterprise project review materials in the report content quality assessment, natural language processing technology is used to convert the review materials containing detailed project introductions, progress reports, and results presentations into a form that can be understood by computers.

[0018] Specifically, when constructing the evaluation index system, for projects represented by the State Grid Social Responsibility Rooted Project, content matching is carried out according to the review materials of the rooted project and the rooted project evaluation criteria after standard decomposition to determine the evaluation basis and scope, and qualitative evaluation is carried out with reference to natural language semantic understanding, machine learning models and large language models. Quantitative evaluation is carried out based on the analysis results of the four-level indicator problems, including quantification of the four-level indicators and scoring of the indicators at each level. The final score is divided into four levels of ABCD to make a project classification judgment and determine whether the prediction plan is passed.

[0019] Specifically, in the cognitive computing step of the report content quality assessment, cognitive computing technology is used to simulate human thinking, match the information in the review materials with the evaluation criteria of the indicator evaluation model, and determine the degree of conformity between the two.

[0020] Specifically, in the qualitative and quantitative evaluation links of the report content quality assessment, for projects represented by the State Grid Social Responsibility Rooting Project, the qualitative and quantitative evaluation methods in the above-mentioned evaluation index system are adopted.

[0021] Specifically, when constructing the indicator evaluation model, the quantitative evaluation results are used to fit the expert voting training process, the indicators are assigned initial scores based on the qualitative evaluation results, historical data are trained, the indicator learning weights are used to fit the expert voting numbers, and the indicator learning weights are used to adjust the indicator scores.

[0022] Specifically, according to the comparison between the model evaluation results and the expert evaluation results and the comparison curve between the model score and the expert vote, if the model evaluation results are close to the expert evaluation results and the fitting degree of the comparison curve between the model score and the expert vote is high, it indicates that the evaluation method has high accuracy and reliability.

[0023] The beneficial effects of the present invention are:

[0024] 1. Improve efficiency and reduce costs: This method significantly improves evaluation efficiency, shortens evaluation cycle, and reduces labor costs.

[0025] 2. Enhance the accuracy and objectivity of evaluation: Improve the accuracy and objectivity of evaluation and effectively reduce the impact of human errors and subjective biases.

[0026] 3. Provide scientific and systematic evaluation tools: Provide enterprises with a scientific and systematic evaluation tool to help promote project implementation and ensure project quality.

[0027] 4. Improve the company’s image and execution: The application of this method can also improve the company’s social image and responsibility execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0029] Figure 1 It is a schematic diagram of the overall process of the project intelligent evaluation method supporting customized standards and multi-dimensional semantic analysis of the present invention;

[0030] Figure 2 This is a partial screenshot of the enterprise project evaluation index system constructed by the present invention based on the State Grid's social responsibility rooted project evaluation standard;

[0031] Figure 3 It is a flow chart of the qualitative and quantitative evaluation method of enterprise projects based on the large language model of the present invention;

[0032] Figure 4 A flowchart of the expert voting training process for fitting the quantitative evaluation results of the present invention;

[0033] Figure 5 A schematic diagram showing the comparison between the evaluation results of the model proposed in the present invention and the evaluation results of experts;

[0034] Figure 6 Schematic diagram of the comparison curve between the model score of the present invention and the expert vote. DETAILED DESCRIPTION

[0035] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0036] The present invention proposes a project intelligent evaluation method that supports customized standards and multi-dimensional semantic analysis. Figure 1 As shown, the method includes:

[0037] Step 1: Build an overall standard system based on the enterprise project evaluation standards: Decompose the enterprise project evaluation standards in detail, divide them according to different regions or specific types, and parse them into an evaluation index system. This step specifically includes customizing project evaluation standards, decomposing by type / region, semantically parsing evaluation standards, and building an evaluation index system.

[0038] Customized project evaluation criteria: Enterprise projects are diverse and unique, and different enterprises formulate different evaluation criteria based on their own strategic goals, industry characteristics and other factors. This method allows enterprises to customize project evaluation criteria, which lays the foundation for the flexibility of the evaluation system. For example, an innovative technology company may pay more attention to the innovation and technological foresight of the project, while a traditional manufacturing company may pay more attention to the cost control and production efficiency improvement of the project.

[0039] Decomposition by type / region: In order to analyze the evaluation criteria more carefully, they are decomposed by project type (such as R&D projects, production projects, marketing projects, etc.) or region (such as different production bases, sales regions, etc.). This decomposition method helps to accurately evaluate the characteristics of projects of different types or regions. Taking regional decomposition as an example, if an enterprise faces different market environments, policies and regulations in different regions, decomposing the evaluation criteria by region can better reflect the actual situation of local projects.

[0040] Semantic analysis of evaluation criteria: Evaluation criteria often exist in the form of natural language, and semantic analysis is required to transform them into quantifiable and operational evaluation indicators. Through multi-dimensional semantic analysis technology, we can deeply understand the connotation of evaluation criteria. For example, for the criterion of "project innovation", semantic analysis can decompose it into sub-indicators of multiple dimensions such as technological innovation, business model innovation, and management innovation, thereby constructing a more comprehensive evaluation indicator system.

[0041] Construct an evaluation index system: Based on the above steps, the customized, decomposed and semantically parsed evaluation criteria are transformed into a specific evaluation index system. This index system is the basis for subsequent evaluation work and covers all aspects of the project, such as the project's economic benefit indicators (cost, income, etc.), social benefit indicators (environmental impact, employment promotion, etc.), technical indicators (technical difficulty, technical innovation, etc.), etc.

[0042] Step 2, indicator evaluation model construction: based on the collected enterprise historical project materials, the corpus is cleaned and samples are automatically generated, and the indicator evaluation model is designed in combination with the established standard indicator system, and then the model is trained and the indicator evaluation model library is formed. That is, the indicator evaluation model is comprehensively constructed based on the evaluation indicator system and the processed enterprise historical project materials, including data collation, corpus cleaning and automatic generation, sample generation, indicator evaluation model design, model training, and indicator evaluation model library generation.

[0043] Data collation: Collect historical project materials of the enterprise, which contain rich project information, such as project plans, execution status, achievements, etc. Sort these data, remove irrelevant information, unify data formats, etc., for subsequent processing.

[0044] Corpus cleaning and automatic generation: Since historical project materials may have data noise, non-standard expressions and other problems, corpus cleaning is required. At the same time, samples are automatically generated based on the cleaned corpus, and these samples will be used for model training. For example, the project's various indicator data and corresponding evaluation results are organized into sample data to provide data support for the relationship between model learning indicators and evaluation results.

[0045] Sample generation: In addition to generating samples from historical project materials, supplementary samples can also be generated through data simulation, expert experience, etc. according to the requirements of the evaluation index system to improve the generalization ability of the model.

[0046] Design indicator evaluation model: Design a suitable indicator evaluation model based on the constructed evaluation indicator system and generated samples. You can choose machine learning algorithms (such as neural networks, decision trees, etc.) or traditional statistical models (such as regression models, etc.) as the basic model framework, and design the model structure according to the characteristics of the project and the distribution of the data.

[0047] Training model: Use the generated samples to train the designed model and adjust the model parameters so that the model can accurately output reasonable evaluation results based on the input project indicator data.

[0048] Generate an indicator evaluation model library (PLM+LLM): The indicator evaluation model obtained after training can be stored in the model library, which can be managed by combining product lifecycle management (PLM) and language model (LLM). PLM can manage the version and scope of application of the model, and LLM can provide support for the semantic understanding and optimization of the model, making it convenient to call the appropriate model in different project scenarios.

[0049] Step 3, report content quality assessment: intelligently analyze the content of the enterprise project review materials, match it with the evaluation criteria of the designed indicator evaluation model, and finally perform completion calculation and quality assessment in combination with the indicator model library. That is, the report content quality assessment is completed on the basis of step 2, including understanding the content of the enterprise project review materials, intelligent analysis of the material content, matching the evaluation criteria with the material content through cognitive computing, and calculating the completion and quality assessment (qualitative and quantitative evaluation).

[0050] Understanding the content of enterprise project review materials: Understanding the content of enterprise project review materials is the first step in the evaluation. These review materials may include detailed introductions of projects, progress reports, and achievement presentations. Through natural language processing technology, these text contents are converted into a form that computers can understand.

[0051] Intelligent analysis of material content: Further intelligent analysis of the content of the review materials is performed to extract information related to the evaluation indicators. For example, the actual cost and progress of the project can be extracted from the project progress report, and the technical achievements and economic benefits of the project can be extracted from the results display.

[0052] Cognitive computing: Using cognitive computing technology, the information in the review materials is matched with the evaluation criteria of the indicator evaluation model. Cognitive computing can simulate the way humans think, analyze and judge complex information, and determine the degree of conformity between the project situation in the review materials and the evaluation criteria.

[0053] Matching evaluation criteria with material content: Through cognitive computing, accurately match the content in the review materials with the evaluation criteria, find the corresponding evaluation indicators, and determine their weights and evaluation levels in the evaluation criteria.

[0054] Qualitative evaluation and quantitative evaluation: On the basis of matching, qualitative evaluation (such as whether the innovation of the project is high, medium, low, etc.) and quantitative evaluation (such as whether the cost saving rate of the project is 10%, etc.) are carried out. Qualitative evaluation can provide an overall qualitative description of the project, while quantitative evaluation can give specific numerical results. The combination of the two can evaluate the project more comprehensively.

[0055] Completion calculation and quality assessment: Based on the results of qualitative and quantitative assessments, calculate the completion of the project, that is, the ratio of the actual completion of the project to the expected goal. At the same time, evaluate the overall quality of the project, comprehensively consider various aspects of the project, such as economic benefits, social benefits, technical level, etc., and obtain the quality assessment results of the project.

[0056] Step 4: Comprehensively consider the evaluation standard system, indicator evaluation model and report content quality assessment constructed in steps 1-3 to make a quality assessment of the overall work, and then intelligently generate a quality assessment report.

[0057] Comprehensively consider the results of the three parts: standard system construction, indicator evaluation model construction, and report content quality assessment. The standard system construction provides a framework and basis for the evaluation, the indicator evaluation model construction provides specific methods and tools for the evaluation, and the report content quality assessment is a direct evaluation of the project review materials. Integrate the results of these three parts and evaluate the overall work quality through certain algorithms and rules. For example, you can set weights according to the importance of each part, and then calculate the overall work quality score by weight. Finally, generate a quality assessment report based on the results of the overall work quality assessment. This report will elaborate on the project's evaluation results, existing problems, suggestions for improvement, etc., and provide an important reference for the company's project decision-making, management and improvement.

[0058] Figure 2 The enterprise project evaluation index system built based on the State Grid's social responsibility rooted project evaluation standards is demonstrated. The construction process includes:

[0059] First, the review materials of the rooted project are matched with the evaluation criteria of the rooted project after the standard decomposition. By comparing the review materials with the evaluation criteria, the basic basis and scope of the evaluation can be determined. In the State Grid Social Responsibility Rooted Project, the review materials may include the implementation of the project, the impact on society, etc., while the evaluation criteria clearly define the requirements and expectations in these aspects. Accurate content matching can ensure the pertinence of subsequent evaluations.

[0060] Next, refer to the three sections of natural language semantic understanding, machine learning model and large language model for qualitative evaluation. Natural language semantic understanding helps to deeply interpret the semantic information in the review materials and explore the implicit project characteristics and achievements; the machine learning model can make a preliminary qualitative judgment on the project based on the existing data and patterns; the large language model, with its powerful language processing capabilities, conducts a comprehensive qualitative analysis of the overall situation of the project. The synergy of these three sections can comprehensively and deeply evaluate the project qualitatively, for example, judging whether the project has qualitative characteristics such as innovation and sustainability in fulfilling its social responsibilities.

[0061] Then, the project is quantitatively evaluated based on the analysis results of the four-level indicator questions. The four-level indicators provide a detailed quantitative dimension for the evaluation. The quantitative evaluation is first quantified according to the four-level indicators, which means converting each four-level indicator into a quantifiable value, such as quantifying the specific indicators such as the amount of resource input and the number of beneficiaries of the project. Then, the indicator scores are carried out level by level, and the four-level indicators are gradually summarized to the third-level, second-level and finally first-level indicator scores. This step-by-step summary method can systematically integrate the indicator information at all levels and comprehensively reflect the overall situation of the project.

[0062] According to the final score, the project is divided into four levels, ABCD, to determine whether the project is passed. If passed, the predicted output is 1, and if not, the output is zero. This grading method is simple and clear, and can provide companies with intuitive project evaluation results, making it easier for companies to make decisions. For example, A-level means that the project performs well in all aspects and fully meets or exceeds the expected standards, while D-level means that the project has many problems and fails to meet the basic requirements.

[0063] When the prediction result fails, the cross-loss is calculated and the four-level indicators are re-quantified, and the above scoring steps are repeated. This mechanism ensures the accuracy and reliability of the evaluation results. Through re-quantification and scoring, the project can be analyzed and adjusted more carefully to avoid misjudgment caused by the limitations of one-time evaluation.

[0064] Figure 3 Shown Figure 1 The third stage in the method flow. First, the evaluation materials of the rooted project are matched with the evaluation criteria of the rooted project after the standard decomposition. This link is like the cornerstone of building a building, which ensures the accuracy and pertinence of the evaluation. The evaluation materials contain various actual conditions of the project, while the evaluation criteria are the scale to measure the project. The matching of the two can accurately locate the performance of the project in each evaluation dimension.

[0065] Subsequently, a qualitative evaluation is conducted with reference to three sections: natural language semantic understanding, machine learning model, and large language model. Natural language semantic understanding plays an important role in interpreting the semantic connotation of the review materials. It can deeply explore the potential information in the materials, such as special circumstances during the implementation of the project or results that are not clearly stated. Machine learning models, whether classification, regression, or matching models, can make preliminary qualitative classifications of projects based on existing data patterns. The large language model is fine-tuned through prompt instructions to conduct qualitative analysis of the project from a more macro and comprehensive perspective, taking into account multiple factors of the project, such as the project's innovation, sustainability, and other characteristics that are difficult to simply quantify.

[0066] Next, the project is quantitatively evaluated based on the analysis results of the four-level indicator questions. The four-level indicators provide a detailed quantitative framework for project evaluation. Quantitative evaluation starts with the quantification of the four-level indicators. This process converts the four-level indicators into measurable values ​​by setting weight matrices and other methods. Then, the indicators are scored step by step from the fourth level to the third level, the second level, and then the first level. This step-by-step summary method can comprehensively and systematically integrate the information at all levels of the project, and finally summarize the first-level indicator score.

[0067] According to the final scoring results, the projects are divided into four levels, ABCD, for project grading. This grading method clearly reflects the overall level of the project. For example, A-level represents that the project performs well in all aspects, while D-level means that the project has many deficiencies. At the same time, the prediction scheme is determined based on the grading. If it passes, the prediction result is one, and if it fails, it is zero. When the prediction result is not passed, the cross-loss is continued to be calculated and the four-level indicators are re-quantified, and then the previous scoring steps are repeated. This mechanism ensures the reliability of the evaluation results, avoids misjudgments caused by errors that may exist in one-time evaluations, and makes the evaluation results more truly reflect the actual situation of the project, providing a solid basis for enterprise project decision-making and improvement.

[0068] In the process of project intelligent evaluation, Figure 4 The quantitative evaluation results shown are a key step in the expert voting training process. First, the indicators are assigned initial scores based on the qualitative evaluation results, which is the basis for building a quantitative evaluation system. The initial scores can reflect the preliminary evaluation of the project on each indicator.

[0069] Next, historical data is trained, which utilizes data resources from past projects. By fitting expert votes with indicator learning weights, the experience and judgment of experts can be incorporated into the evaluation system. Expert votes represent the opinions of people with professional knowledge and rich experience in related fields on project indicators. Combining them with indicator learning weights can make the evaluation results more authoritative and reasonable.

[0070] Then, the indicator scores are adjusted using indicator learning weights. This operation can optimize the indicator scores based on expert experience and historical data, so that the final scores more accurately reflect the actual situation of the project.

[0071] from Figure 5 and Figure 6 It can be seen that the model evaluation results proposed in the present invention are compared with the expert evaluation results, as well as the model score and expert vote comparison curves. These comparisons show the relationship between the model of the present invention and the expert evaluation. If the model evaluation results are close to the expert evaluation results, and the model score and expert vote comparison curve have a high degree of fit, this indicates that the evaluation method of the present invention has high accuracy and reliability. It can replace the traditional evaluation method that relies only on the subjective judgment of experts to a certain extent, and at the same time combine the experience of experts to provide a more scientific, objective and comprehensive solution for project evaluation.

[0072] The beneficial effects of the present invention are:

[0073] 1. Improve efficiency and reduce costs: This method significantly improves evaluation efficiency, shortens evaluation cycle, and reduces labor costs.

[0074] 2. Enhance the accuracy and objectivity of evaluation: Improve the accuracy and objectivity of evaluation and effectively reduce the impact of human errors and subjective biases.

[0075] 3. Provide scientific and systematic evaluation tools: Provide enterprises with a scientific and systematic evaluation tool to help promote project implementation and ensure project quality.

[0076] 4. Improve the company’s image and execution: The application of this method can also improve the company’s social image and responsibility execution.

[0077] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A project intelligent evaluation method supporting customized standards and multi-dimensional semantic analysis, characterized in that: include: Construct an overall standard system based on the project evaluation standards, including customizing the project evaluation standards, decomposing the evaluation standards according to project types or regions, and semantically parsing the evaluation standards to construct an evaluation index system, which covers the economic benefit indicators, social benefit indicators, and technical indicators of the project; Based on the collected historical project materials of the enterprise, data collation, corpus cleaning and automatic sample generation, sample generation, indicator evaluation model design, and model training are performed, and finally an indicator evaluation model library is generated. The indicator evaluation model library is managed in a combination of product lifecycle management PLM and language model LLM to construct an indicator evaluation model; Understand the content of enterprise project review materials, intelligently analyze the material content, match the evaluation standards with the material content through cognitive computing, conduct qualitative and quantitative evaluations to calculate the degree of completion and quality evaluation, and thus complete the report content quality evaluation; Comprehensively consider the constructed evaluation standard system, indicator evaluation model and the results of report content quality assessment, make a quality assessment of the overall work by setting algorithms and rules, and intelligently generate a quality assessment report.

2. The project intelligent evaluation method according to claim 1, characterized in that: In the semantic analysis step of evaluation criteria in constructing the overall standard system, the evaluation criteria in the form of natural language are analyzed into quantifiable and operational evaluation indicators through multi-dimensional semantic analysis technology.

3. The project intelligent evaluation method according to claim 1, characterized in that: In the sample generation step of constructing the indicator evaluation model, in addition to generating samples from historical project materials, supplementary samples are generated through data simulation and expert experience according to the requirements of the evaluation indicator system to improve the generalization ability of the model.

4. The project intelligent evaluation method according to claim 1, characterized in that: In the step of designing the indicator evaluation model in constructing the indicator evaluation model, a machine learning algorithm is selected as the basic model framework based on the constructed evaluation indicator system and the generated samples, and the model structure is designed according to the characteristics of the project and the distribution of the data.

5. The project intelligent evaluation method according to claim 1, characterized in that: In the step of understanding the content of enterprise project review materials in the report content quality assessment, natural language processing technology is used to convert the review materials containing detailed project introductions, progress reports, and results presentations into a form that can be understood by computers.

6. The project intelligent evaluation method according to claim 1, characterized in that: When constructing the evaluation index system, for projects represented by the State Grid Social Responsibility Rooted Project, content matching is carried out according to the review materials of the rooted project and the rooted project evaluation criteria after standard decomposition to determine the evaluation basis and scope, and qualitative evaluation is carried out with reference to natural language semantic understanding, machine learning models and large language models. Quantitative evaluation is carried out based on the analysis results of the four-level indicator problems, including four-level indicator quantification and level-by-level indicator scoring. The final score is divided into four levels of ABCD to make project grading and determine whether the prediction plan is passed; if the prediction plan does not pass, the cross-loss is calculated and the four-level indicator quantification and scoring steps are repeated.

7. The project intelligent evaluation method according to claim 6, characterized in that: In the cognitive computing step of the report content quality assessment, cognitive computing technology is used to simulate human thinking, match the information in the review materials with the evaluation criteria of the indicator evaluation model, and determine the degree of conformity between the two.

8. The project intelligent evaluation method according to claim 6, characterized in that: In the qualitative and quantitative evaluation stages of the report content quality assessment, for projects represented by the State Grid Corporation of China's Social Responsibility Rooting Project, the qualitative and quantitative evaluation methods in the above-mentioned evaluation index system are adopted.

9. The project intelligent evaluation method according to claim 1, characterized in that: When constructing the indicator evaluation model, the quantitative evaluation results are used to fit the expert voting training process, the indicators are assigned initial scores based on the qualitative evaluation results, historical data are trained, the indicator learning weights are used to fit the expert voting numbers, and the indicator learning weights are used to adjust the indicator scores.

10. The project intelligent evaluation method according to claim 9, characterized in that: According to the comparison between the model evaluation results and the expert evaluation results and the comparison curve between the model score and the expert vote, if the model evaluation results are close to the expert evaluation results and the fitting degree of the comparison curve between the model score and the expert vote is high, it indicates that this evaluation method has high accuracy and reliability.

Images