Multi-dimensional enterprise qualification evaluation method and system

Abstract

The invention discloses a multi-dimensional enterprise qualification evaluation method and system, and the method comprises the steps: obtaining multi-dimensional original data, and obtaining a standardized enterprise multi-dimensional feature data set through employing a data cleaning and normalization preprocessing technology; aiming at a standardized enterprise multi-dimensional feature data set, performing grouping optimization and weight calculation on indexes by utilizing an improved WP-PVC algorithm, and constructing a multi-dimensional evaluation index system; based on the multi-dimensional evaluation index system, a double-standard WP-PVC algorithm is adopted to carry out inter-index correlation analysis and score calculation, and an enterprise qualification comprehensive score result is generated; according to an enterprise qualification comprehensive scoring result, in combination with a multi-logarithm algorithm of random online sorting, calculating a matching degree between the enterprise and various science and technology projects, and outputting a project matching recommendation list and application success rate prediction; and automatically generating an enterprise qualification diagnosis scoring report based on the enterprise qualification comprehensive scoring result and the project matching recommendation list. According to the invention, comprehensive evaluation of enterprise qualification, accurate project matching and scientific application guidance are realized.

Description

Technical Field

[0001] This invention relates to the field of enterprise qualification assessment, and in particular to an enterprise qualification assessment method and system based on a multi-dimensional indicator system, which is used to comprehensively assess an enterprise's financial operations, R&D capabilities, industry position, and other dimensions, and provides project matching recommendations and application success rate predictions. Background Technology

[0002] Enterprise qualification assessment is a crucial basis for enterprises to participate in market competition, apply for projects, and obtain financing. Traditional enterprise assessment techniques mainly include financial statement analysis and expert scoring. Financial statement analysis involves analyzing financial data such as the enterprise's balance sheet and income statement to arrive at an evaluation of the enterprise's financial condition; expert scoring relies on industry experts to subjectively score various aspects of the enterprise based on their personal experience and then summarize the results.

[0003] One existing enterprise qualification assessment technology employs a multi-indicator system and weighted calculation method, incorporating information such as the enterprise's financial data, market performance, and technological capabilities into the assessment scope. By setting different weight coefficients, it calculates the enterprise's comprehensive score. After data collection, this technology processes the data of each indicator through a linear weighted model to form the final assessment result, and classifies the enterprise's qualification level based on thresholds.

[0004] However, this technology has obvious shortcomings: First, the evaluation indicator system is relatively static and it is difficult to make flexible adjustments according to the characteristics of different industries and the development stage of enterprises; second, the interrelationships between indicators are complex, and simple linear weighting is difficult to accurately reflect the comprehensive strength of enterprises; third, the evaluation results have a low degree of matching with the actual project application requirements, and cannot provide enterprises with accurate project application guidance, resulting in a low success rate of project applications. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-dimensional enterprise qualification assessment method and system, which solves the technical problems of static assessment indicator system, simple indicator relationship processing, and low matching degree between assessment results and project applications in the existing technology.

[0006] To achieve the above objectives, this invention provides a multi-dimensional enterprise qualification assessment method, comprising the following steps: We acquire multi-dimensional raw data containing corporate financial operations, R&D capabilities, and industry position, and apply data cleaning and normalization preprocessing techniques to obtain a standardized multi-dimensional feature dataset of enterprises. For the standardized enterprise multidimensional feature dataset, the improved WP-PVC algorithm is used to group and optimize the indicators and calculate the weights, and a multidimensional evaluation indicator system including financial operation, R&D capability level, industry status and brand effect is constructed. Based on the aforementioned multi-dimensional evaluation index system, the dual-standard WP-PVC algorithm is used to perform correlation analysis and scoring calculation among the indicators, generating a comprehensive enterprise qualification score that includes both the overall enterprise qualification score and the sub-dimensional scoring results. Based on the comprehensive score of the enterprise's qualifications, and combined with a multi-logarithm algorithm for random online sorting, the matching degree between the enterprise and various technology projects is calculated, and a project matching recommendation list and application success rate prediction are output. Based on the comprehensive score of the enterprise's qualifications and the project matching recommendation list, an enterprise qualification diagnostic score report is automatically generated, which includes an analysis of the enterprise's strengths, a diagnosis of its weaknesses, and suggestions for project applications.

[0007] Preferably, the step of acquiring multi-dimensional raw data including corporate financial operations, R&D capabilities, and industry position, and applying data cleaning and normalization preprocessing techniques to obtain a standardized multi-dimensional feature dataset of the enterprise, includes: By accessing public data sources including the enterprise credit information disclosure system, tax system, and business registration information system, and combining network data collection technology, we collect basic information and financial data such as enterprise business registration information, historical financial statement data, tax records, and financing history to obtain a basic enterprise information dataset. Collect and obtain data related to enterprise R&D capabilities, including the number and type of enterprise patents, citation rate, R&D investment ratio, proportion of scientific research personnel, and technology transfer status, to obtain enterprise R&D capability dataset; Collect market performance data including enterprise market share, brand awareness, user reviews, industry ranking, and product competitiveness to obtain enterprise industry position dataset; Based on the enterprise basic information dataset, the enterprise R&D capability dataset, and the enterprise industry position dataset, the data is cleaned using techniques including median padding and locality-sensitive hashing. Missing values, outliers, and duplicate values ​​are detected and processed, and the cleaned multidimensional enterprise data is output. Based on the cleaned enterprise multidimensional data, data standardization is performed using methods including maximum and minimum value normalization and Z-score standardization, so that the data in each dimension can be compared, resulting in the standardized enterprise multidimensional feature dataset.

[0008] Preferably, for the standardized enterprise multidimensional feature dataset, the improved WP-PVC algorithm is used to group and optimize the indicators and calculate their weights, constructing a multidimensional evaluation indicator system that includes financial performance, R&D capability level, industry position, and brand effect, including: Based on the standardized enterprise multidimensional feature dataset, and combined with industry expert experience and policy guidance, a candidate set of indicators reflecting the enterprise's financial and operational status, R&D capability level, industry position, and brand effect is selected to establish an initial indicator pool. The enterprise evaluation indicators in the initial pool of indicators are regarded as vertices in the graph, and the relationships between indicators are regarded as edges. The strength of the relationship between indicators is represented by the edge weight. The improved WP-PVC algorithm is applied to group the indicators, dividing them into different groups including financial operations, R&D capabilities, and industry status, thus obtaining indicator groups of different categories. For the different groups of indicators, the improved WP-PVC algorithm is used to set the minimum threshold that each indicator group needs to cover, solve the subset of indicators with the minimum weight, and determine the weight configuration of each indicator. Based on the weight configuration of each indicator, the discrimination and predictive ability of the indicator system are verified through historical sample data. The information gain and variance contribution rate of each indicator are calculated, redundant indicators are eliminated, and the multi-dimensional evaluation indicator system is formed.

[0009] Preferably, the step of using the improved WP-PVC algorithm to perform grouping optimization and weight calculation of indicators further includes: Based on the characteristics of different industries and the development stage of enterprises, the multi-dimensional evaluation index system is adaptively adjusted to generate customized evaluation index systems for different types of enterprises. The adaptive adjustment includes adjusting the weights of indicators in each dimension according to the characteristics of the industry to which the enterprise belongs, adjusting the evaluation focus according to the development stage of the enterprise, and dynamically adjusting the indicator weights according to the current policy environment and support direction.

[0010] Preferably, the step of using the dual-standard WP-PVC algorithm to perform correlation analysis and scoring calculation among indicators based on the multi-dimensional evaluation index system, and generating a comprehensive enterprise qualification score result including a comprehensive enterprise qualification score and multi-dimensional scoring results, includes: The data of various indicators of enterprises in the multi-dimensional evaluation indicator system are standardized to eliminate the influence of dimensions, so that different indicators are comparable and standardized indicator data are obtained. Based on the standardized indicator data, the bi-standard WP-PVC algorithm is applied to analyze the correlation between indicators within the same dimension, construct an indicator correlation network within the dimension, calculate the correlation coefficient between indicators, identify key indicators and auxiliary indicators, and determine the importance ranking of indicators within the dimension by combining the optimized weights of each indicator in the multi-dimensional evaluation indicator system in the evaluation system. Based on the importance ranking of the indicators and the optimization weights, the standardized value of the enterprise on each indicator is multiplied by the corresponding optimization weight, and a non-linear adjustment is made. The results are then aggregated according to the weighted average rule to calculate the enterprise's scores in each dimension of financial operations, R&D capabilities, and industry position, forming a multi-dimensional scoring result. The dual-standard WP-PVC algorithm is applied to dynamically adjust the weights of each dimension based on their contribution to the overall enterprise qualification, taking into account industry characteristics, enterprise type, and application project characteristics. The scores of each dimension are then weighted according to the optimized weights to generate the overall enterprise qualification score, which includes the overall enterprise qualification score and the scores of each dimension.

[0011] Preferably, the dynamic adjustment of the weights of each dimension further includes: Based on the contribution of each dimension to enterprise success in different industries, we analyze the impact of industry characteristics on dimension weights and obtain industry characteristic weight adjustment coefficients. Based on historical successful cases of different types of science and technology project applications, we analyze the performance characteristics of each dimension to obtain the project orientation weight adjustment coefficient. Based on the industry characteristic weight adjustment coefficient and the project orientation weight adjustment coefficient, by setting a minimum coverage threshold for each dimension, the optimal weight allocation scheme is found so that the evaluation results are closest to historical successful cases, and the optimized dimension weight configuration is generated.

[0012] Preferably, the step of calculating the matching degree between the enterprise and various technology projects based on the comprehensive score of the enterprise's qualifications, combined with a multi-logarithm algorithm for random online sorting, and outputting a project matching recommendation list and application success rate prediction includes: Collect information on application requirements, review standards, and funding amounts for various science and technology projects and certifications; establish a structured science and technology project database; and classify and label projects by industry, technology direction, and support stage in multiple dimensions to build a science and technology project database that is updated in real time. Based on the comprehensive score of enterprise qualifications and the real-time updated technology project database, enterprise characteristics and project requirements are transformed into multi-dimensional vectors, the matching degree is calculated, and a matching model between enterprises and projects is constructed. Based on the matching model and successful cases of matching enterprise characteristics with projects in historical application data, a multi-logarithmic algorithm for random online sorting is applied to optimize the project recommendation ranking and generate an optimized project matching sequence. Based on the optimized project matching sequence and historical application case analysis, an application success rate prediction model is constructed to calculate the success probability of enterprises applying for various types of projects, provide enterprises with application decision references, and generate application success rate prediction results. Based on the optimized project matching sequence and the application success rate prediction results, a project recommendation list suitable for enterprises to apply for is generated, sorted by success rate from highest to lowest, and accompanied by matching reasons and application suggestions. The project matching recommendation list and application success rate prediction are then output.

[0013] Preferably, the step of automatically generating an enterprise qualification diagnostic scoring report, which includes enterprise strength analysis, weakness diagnosis, and project application suggestions, based on the comprehensive enterprise qualification scoring result and the project matching recommendation list, includes: Based on the multi-dimensional scoring results in the comprehensive enterprise qualification scoring results, identify the enterprise's advantageous indicators in each dimension, analyze the reasons for the formation and sustainability of the advantages, and generate a section of the enterprise advantage analysis report. Identify the dimensions and indicators in the comprehensive enterprise qualification score that are below a preset threshold, analyze the reasons for the deficiencies, and generate targeted improvement suggestions and improvement paths by combining data from industry benchmark enterprises, thus forming a deficiency diagnosis and improvement suggestion. Based on the comprehensive score of the enterprise's qualifications and the project matching recommendation list, a phased and multi-tiered project application strategy is formulated, including projects that can be applied for in the near future and projects with medium and long-term development goals, and an application plan is generated. By integrating the enterprise strengths analysis report, the deficiencies diagnosis and improvement suggestions, and the application plan, and adaptively selecting the report template and content structure based on the enterprise type, industry characteristics, and evaluation results, the enterprise qualification diagnosis and scoring report is generated.

[0014] This invention also provides a multi-dimensional enterprise qualification assessment system, comprising: The data acquisition module is used to acquire multi-dimensional raw data including corporate financial operations, R&D capabilities, and industry position. It applies data cleaning and normalization preprocessing techniques to obtain a standardized multi-dimensional feature dataset of the enterprise. The indicator system construction module is used to optimize the indicators by grouping and calculating the weights using the improved WP-PVC algorithm on the standardized enterprise multidimensional feature dataset, and to construct a multidimensional evaluation indicator system that includes financial and operational status, R&D capability level, industry position and brand effect. The scoring calculation module is used to perform correlation analysis and scoring calculation between indicators based on the multi-dimensional evaluation index system and the dual-standard WP-PVC algorithm, and generate a comprehensive enterprise qualification score result that includes the comprehensive enterprise qualification score and the sub-dimensional scoring results. The project matching module is used to calculate the matching degree between enterprises and various technology projects based on the comprehensive score of enterprise qualifications and a multi-logarithm algorithm for random online sorting, and output a project matching recommendation list and application success rate prediction. The report generation module is used to automatically generate an enterprise qualification diagnostic scoring report that includes analysis of enterprise strengths, diagnosis of weaknesses, and suggestions for project application, based on the comprehensive enterprise qualification score and the project matching recommendation list.

[0015] The beneficial effects of this invention are as follows: by introducing an improved WP-PVC algorithm to group and optimize indicators and calculate weights, a more scientific and reasonable multi-dimensional evaluation indicator system is constructed; by applying the dual-standard WP-PVC algorithm to perform correlation analysis and scoring calculation among indicators, the overall strength of enterprises is more accurately reflected; by combining a multi-logarithmic algorithm with random online sorting for project matching and success rate prediction, more accurate project application guidance is provided to enterprises; and by intelligently generating enterprise qualification diagnostic scoring reports, comprehensive support is provided for enterprise decision-making, ultimately improving the success rate of enterprise project applications. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the multi-dimensional enterprise qualification assessment method provided in the embodiments of the present invention; Figure 2 This is a flowchart of project matching and application success rate prediction provided in an embodiment of the present invention; Figure 3 This is a structural diagram of the multi-dimensional enterprise qualification assessment system provided in this embodiment of the invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] like Figure 1 As shown, the multi-dimensional enterprise qualification assessment method provided by this invention includes the following steps: S1: Obtain multi-dimensional raw data including corporate financial operations, R&D capabilities, and industry position; apply data cleaning and normalization preprocessing techniques to obtain a standardized multi-dimensional feature dataset of the enterprise. Step S1 is the enterprise multidimensional data collection and preprocessing stage. This step aims to acquire multidimensional raw data encompassing the enterprise's financial operations, R&D capabilities, and industry position, and apply data cleaning and normalization preprocessing techniques to ultimately obtain a standardized enterprise multidimensional feature dataset. In practice, this first requires collecting comprehensive information about the enterprise through various channels, including public data sources, professional databases, and web scraping, to ensure the data's comprehensiveness and accuracy. The collected data covers multiple dimensions such as the enterprise's basic information, financial status, R&D capabilities, and market performance, providing the raw data foundation for subsequent qualification assessments. Subsequently, this raw data undergoes cleaning and standardization, including missing value handling, outlier identification and correction, duplicate data merging, and data normalization, to eliminate the influence of different data sources and different units of measurement, ensuring the comparability of data across all dimensions. Through this step, a structured and standardized enterprise multidimensional feature dataset can be constructed, laying the foundation for subsequent indicator system construction and scoring calculations.

[0020] S2: For the standardized enterprise multidimensional feature dataset, the improved WP-PVC algorithm is used to group and optimize the indicators and calculate the weights to construct a multidimensional evaluation indicator system that includes financial and operational status, R&D capability level, industry position and brand effect. Step S2 is the enterprise qualification assessment indicator system construction stage. This step utilizes an improved WP-PVC algorithm to optimize the grouping and weight calculation of indicators based on a standardized multi-dimensional enterprise feature dataset, constructing a multi-dimensional evaluation indicator system encompassing financial operations, R&D capabilities, industry standing, and brand effect. In the specific implementation process, firstly, based on the standardized multi-dimensional enterprise feature dataset and combining industry expert experience and policy guidance, an initial pool containing various potential evaluation indicators is established. Then, these indicators are treated as vertices in a graph structure, and the relationships between indicators are considered as edges. Edge weights represent the strength of the relationships between indicators, constructing an indicator association network. Using the improved WP-PVC algorithm, indicators are divided into different groups, ensuring high correlation within the same group and weak correlation between indicators in different groups, while guaranteeing the assessment coverage of each dimension. Based on this, a minimum coverage threshold is set for each indicator group, the optimal indicator subset and its weight configuration are calculated, and the effectiveness of the indicator system is verified using historical sample data. Redundant indicators are eliminated, forming a concise and efficient multi-dimensional evaluation indicator system. This indicator system can comprehensively reflect the comprehensive strength of enterprises in terms of financial operations, R&D capabilities, and industry standing, providing a scientific basis for enterprise qualification scoring calculations.

[0021] S3: Based on the multi-dimensional evaluation index system, the dual-standard WP-PVC algorithm is used to perform correlation analysis and scoring calculation between indicators, and generate a comprehensive enterprise qualification score result that includes the comprehensive enterprise qualification score and the sub-dimensional scoring results; Step S3 is the comprehensive enterprise qualification score calculation stage. This step, based on a multi-dimensional evaluation index system, uses the dual-standard WP-PVC algorithm to analyze the correlation between indicators and calculate scores, generating a comprehensive enterprise qualification score that includes both the overall enterprise qualification score and the sub-dimensional scores. In the specific implementation process, firstly, the enterprise's data for each indicator is standardized to eliminate the influence of dimensions and ensure comparability between different indicators. Then, the dual-standard WP-PVC algorithm is applied to analyze the correlation between indicators within the same dimension, constructing an indicator correlation network within the dimension, identifying key and auxiliary indicators, and determining the importance ranking of indicators within the dimension. Based on this, according to the indicator importance ranking and optimized weights, the enterprise's scores in various dimensions such as financial operations, R&D capabilities, and industry position are calculated, forming sub-dimensional score results. Finally, based on the contribution of different dimensions to the overall enterprise qualification, considering industry characteristics, enterprise type, and application project characteristics, the weights of each dimension are dynamically adjusted, and the scores of each dimension are weighted according to the optimized weights to generate the comprehensive enterprise qualification score. This scoring method based on the dual-standard WP-PVC algorithm can fully consider the interaction between indicators and industry characteristics, making the scoring results more objective and accurate, and providing a reliable basis for subsequent project matching and qualification diagnosis.

[0022] S4: Based on the comprehensive score of the enterprise's qualifications, and combined with the multi-logarithm algorithm for random online sorting, calculate the matching degree between the enterprise and various technology projects, and output a project matching recommendation list and application success rate prediction; Step S4 is the project matching and application success rate prediction stage. This step calculates the matching degree between enterprises and various science and technology projects based on the comprehensive enterprise qualification score and a multi-logarithmic algorithm for random online sorting, outputting a project matching recommendation list and a predicted application success rate. In the specific implementation process, firstly, information such as application conditions, review standards, and funding amounts for various science and technology projects and qualification certifications is collected to establish a structured science and technology project database. Projects are then categorized and labeled in multiple dimensions to build a real-time updated science and technology project library. Next, based on the comprehensive enterprise qualification score and the science and technology project library, enterprise characteristics and project requirements are transformed into multi-dimensional vectors, and the matching degree is calculated to construct a matching model between enterprises and projects. Then, a multi-logarithmic algorithm for random online sorting is applied to optimize the project recommendation ranking based on successful cases of enterprise characteristics matching projects in historical application data, generating an optimized project matching sequence. Based on this, an application success rate prediction model is constructed to calculate the probability of enterprises successfully applying for various projects, providing enterprises with application decision-making references. Finally, based on the optimized project matching sequence and the application success rate prediction results, a recommended list of projects suitable for enterprises is generated, sorted by success rate, with matching reasons and application suggestions. This project matching method, based on a random online sorting multi-logarithm algorithm, can effectively improve the accuracy and diversity of recommendations, providing enterprises with more valuable guidance on project applications.

[0023] S5: Based on the comprehensive score of the enterprise's qualifications and the project matching recommendation list, automatically generate an enterprise qualification diagnostic score report that includes analysis of the enterprise's strengths, diagnosis of its weaknesses, and suggestions for project application.

[0024] Step S5 is the intelligent diagnostic report generation stage. Based on the comprehensive enterprise qualification score and the project matching recommendation list, this step automatically generates an enterprise qualification diagnostic score report containing enterprise strength analysis, weakness diagnosis, and project application suggestions. In practice, firstly, based on the multi-dimensional score results in the comprehensive enterprise qualification score, the advantages of the enterprise in each dimension are identified, the reasons for these advantages and their sustainability are analyzed, and a strength analysis report paragraph is generated. Then, dimensions and indicators where the enterprise's score is below a preset threshold are identified, the reasons for these shortcomings are analyzed, and targeted improvement suggestions and improvement paths are generated by combining data from industry benchmark enterprises, forming a weakness diagnosis and improvement suggestion. Next, based on the comprehensive enterprise qualification score and the project matching recommendation list, a phased, multi-tiered project application strategy is formulated, including projects that can be applied for in the near term and projects with medium- and long-term development goals, generating an application plan. Finally, the strength analysis report paragraph, weakness diagnosis and improvement suggestions, and application plan are integrated, and the report template and content structure are adaptively selected according to the enterprise type, industry characteristics, and evaluation results to generate a complete enterprise qualification diagnostic score report. This intelligent report generation method not only provides objective evaluation results but also offers targeted improvement suggestions and application strategies, providing comprehensive support and guidance for enterprise decision-making.

[0025] In this embodiment, the acquisition of multi-dimensional raw data including corporate financial operations, R&D capabilities, and industry position, followed by the application of data cleaning and normalization preprocessing techniques, yields a standardized multi-dimensional feature dataset of the enterprise, including: S1.1: By accessing public data sources including the Enterprise Credit Information Disclosure System, the Tax System, and the Business Registration Information System, and combining network data collection technology, we collect basic information and financial data such as enterprise business registration information, historical financial statement data, tax records, and financing history to obtain a basic enterprise information dataset. S1.2: Collect and obtain data related to enterprise R&D capabilities, including the number and type of enterprise patents, citation rate, R&D investment ratio, proportion of scientific research personnel, and technology transfer status, to obtain enterprise R&D capability dataset; S1.3: Collect market performance data including enterprise market share, brand awareness, user reviews, industry ranking, and product competitiveness to obtain enterprise industry position dataset; S1.4: Based on the enterprise basic information dataset, the enterprise R&D capability dataset, and the enterprise industry position dataset, apply techniques including median padding and local sensitive hashing to clean the data, detect and process missing values, outliers, and duplicate values, and output the cleaned multidimensional enterprise data. S1.5: Based on the cleaned enterprise multidimensional data, data standardization is performed using methods including maximum and minimum value normalization and Z-score standardization, so that the data in each dimension can be compared, and the standardized enterprise multidimensional feature dataset is obtained.

[0026] Step S1.1 involves collecting basic enterprise information and financial data. This step utilizes publicly available data sources, including the Enterprise Credit Information Publicity System, the Tax System, and the Business Registration Information System, combined with web data collection technology, to gather basic information and financial data such as business registration information, historical financial statements, tax records, and financing history, thus obtaining a basic enterprise information dataset. During implementation, application programming interfaces (APIs) or web scraping technologies are used to connect to various publicly available government data platforms, such as the National Enterprise Credit Information Publicity System, local tax systems, and business administration systems, to obtain basic information such as the enterprise's registered capital, establishment date, legal representative, equity structure, and change records, as well as financial statement data such as balance sheets, profit and loss statements, and cash flow statements. Simultaneously, supplementary information such as the enterprise's development history, management team, financing history, and major events is collected from the enterprise's official website, industry portals, and financial media through web data collection technology. This data collectively constitutes the basic enterprise information dataset, comprehensively reflecting the enterprise's basic status and financial health, providing fundamental data support for qualification assessment.

[0027] Step S1.2 is the data collection stage for R&D capabilities and technical indicators. This step collects data related to the company's R&D capabilities, including the number and type of patents, citation rate, R&D investment ratio, proportion of R&D personnel, and technology transfer status, to obtain a corporate R&D capability dataset. In practice, this involves connecting to professional data sources such as the State Intellectual Property Office's patent database, science and technology project management system, and human resource database to collect information on the company's patent applications and authorizations, including the number, authorization time, and validity period of invention patents, utility model patents, and design patents. It also involves analyzing the technological field distribution of patents, the number of citations, and patent families to assess the quality and influence of patents. Furthermore, data such as R&D investment amount and ratio, number of R&D personnel and their proportion of total employees, and number and output value of technology transfer projects are extracted from the company's annual reports, science and technology statistical annual reports, and recruitment information to comprehensively evaluate the company's R&D strength and innovation capabilities. These data collectively constitute the corporate R&D capability dataset, providing an objective basis for evaluating the company's technological innovation capabilities.

[0028] Step S1.3 involves collecting data on industry position and market performance. This step gathers market performance data including market share, brand awareness, user reviews, industry ranking, and product competitiveness to obtain a dataset on the company's industry position. During implementation, market share data and industry ranking information are collected from channels such as industry data released by industry associations, research reports from market research institutions, and commercial databases. Simultaneously, web crawling and natural language processing technologies are used to collect user reviews of the company's products and services from social media, e-commerce platforms, and industry forums, and brand reputation index is calculated using sentiment analysis algorithms. Furthermore, data on the company's media exposure, advertising spending, brand value assessment, and competitiveness indicators such as product pricing, product update frequency, and market response speed are also collected. This data aggregation forms the company's industry position dataset, comprehensively reflecting the company's market position and brand influence, providing a data foundation for assessing the company's market competitiveness.

[0029] Step S1.4 is the data cleaning and outlier handling stage. Based on the enterprise basic information dataset, enterprise R&D capability dataset, and enterprise industry position dataset, this step applies techniques including median imputation and locality-sensitive hashing (LSH) to clean the data, detect and process missing values, outliers, and duplicates, and output cleaned multidimensional enterprise data. In practice, the data integrity is first checked to identify missing values, and appropriate processing methods are selected based on the data characteristics. For structural missing values, median imputation, average imputation, or imputation based on values ​​from similar enterprises are used; for unstructured missing values, statistical model prediction or labeling as special values ​​may be used. For outliers, locality-sensitive hashing (LSH) technology is used to quickly identify outliers in the data. This technology maps high-dimensional data to a low-dimensional space, making the hash values ​​of similar data similar, thus efficiently discovering data points that do not conform to normal distribution patterns. Once outliers are identified, the system decides whether to correct (e.g., truncate to a reasonable range), delete, or retain but label based on the reliability of the data source and the degree of anomaly. For duplicate data, the system compares the data source, timestamp, and integrity, selecting to retain the highest quality version or merge information from multiple sources. Through these data cleaning steps, the system can significantly improve data quality, reduce noise impact in subsequent analysis, and output reliable cleaned multidimensional enterprise data.

[0030] Step S1.5 is the data normalization and standardization process. Based on the cleaned multidimensional enterprise data, this step uses methods including maximum-minimum value normalization and Z-score standardization to standardize the data, making the data comparable across dimensions and obtaining a standardized multidimensional enterprise feature dataset. During implementation, the distribution characteristics and numerical range of each indicator are first analyzed, and appropriate standardization methods are selected for different types of indicators. For indicators with clear upper and lower bounds (such as rating indicators), the maximum-minimum value normalization method is used to linearly map the data to the [0,1] interval. The calculation formula is as follows: Xnorm = (X - Xmin) / (Xmax - Xmin), where X is the original value, and Xmin and Xmax are the minimum and maximum values ​​of the indicator, respectively. For indicators that follow a normal distribution or whose data distribution characteristics need to be considered (such as financial ratios, growth rates, etc.), the system uses the Z-score standardization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The calculation formula is as follows: Xstd = (X-μ) / σ, where μ is the mean and σ is the standard deviation.

[0031] Furthermore, the directionality of the indicators is considered. Different processing methods are used for positive indicators ("the bigger the better") and negative indicators ("the smaller the better") to ensure that all indicators, after standardization, follow a unified "the bigger the better" direction. For indicators with non-linear relationships, non-linear transformation methods such as logarithmic transformation may be applied to make the data distribution more reasonable. Through these standardization processes, indicator data with different dimensions and ranges are transformed to a comparable scale, forming a standardized multi-dimensional feature dataset for enterprises, providing a foundation for subsequent scoring calculations and indicator system construction.

[0032] The improved Weighted Partial Vertex Cover (WP-PVC) algorithm is an optimized and upgraded version of the traditional WP-PVC algorithm, primarily improving time complexity, extending to weighted settings, and employing a dual-standard algorithm design. Regarding time complexity optimization, the algorithm significantly improves efficiency by solving a linear programming problem with a complexity of n to the power of ω (where ω is the exponent of matrix multiplication, currently at an optimal value of approximately 2.373). It also eliminates the enumeration step, reducing computational overhead, and removes the extra ε factor in the approximation, improving approximation accuracy. In terms of extending to weighted settings, the algorithm can handle weighted graphs where both vertices and edges have weights, enabling its application to more complex practical problems, such as optimizing indicator weights in enterprise qualification assessment. Regarding the dual-standard algorithm design, when the number of groups is large, the algorithm can control the cost of the solution while approximately satisfying the profit target, achieving a balance between accuracy and efficiency by setting an appropriate approximation ratio. In this invention, the improved WP-PVC algorithm is applied to the indicator system construction stage. By treating enterprise evaluation indicators as vertices in a graph and the relationships between indicators as edges, the scientific grouping and weight optimization of indicators are achieved, thus constructing a more concise and efficient evaluation indicator system.

[0033] The dual-standard WP-PVC algorithm is a special variant of the WP-PVC algorithm, characterized by simultaneously considering two optimization objectives: minimizing the total weight of the selected vertex set (cost minimization) and ensuring that the sum of the edges covered in each edge group reaches a specified threshold (profit maximization). This algorithm decomposes the original problem into two subproblems using Lagrange relaxation, one focusing on cost minimization and the other on profit maximization. A balance is then achieved between the two objectives by adjusting the Lagrange multipliers. In this invention, the dual-standard WP-PVC algorithm is applied to the enterprise qualification scoring calculation process to analyze the correlation between indicators, dynamically adjust dimensional weights, and adapt to the evaluation needs of different industries and enterprise types. This algorithm features flexibility, adaptability, multi-objective optimization, and interpretability. It can flexibly balance evaluation cost and accuracy while ensuring basic coverage, automatically adjust evaluation standards for different industries and enterprise types, consider both cost minimization and information coverage maximization objectives, and ensure that the evaluation results have a clear mathematical basis and are interpretable.

[0034] The random online ranking multi-logarithmic algorithm is an algorithm for solving complex ranking and matching problems. This algorithm combines the characteristics of random sampling, online learning, and multi-logarithmic models. In this algorithm, the system first constructs multiple sub-models through multiple random samplings. Each sub-model processes the relationship between features and the target variable based on logarithmic transformation. Then, the prediction results of these sub-models are combined to form the final ranking decision. Simultaneously, it possesses online learning capabilities, continuously adjusting model parameters based on user feedback and new data to optimize the ranking results. In this invention, the random online ranking multi-logarithmic algorithm is applied to the project matching and success rate prediction stage to optimize project recommendation ranking and improve the accuracy and diversity of recommendations. By learning the feature patterns of historical successful cases, this algorithm can more accurately predict the success rate of enterprises applying for different projects. Furthermore, by balancing exploration and utilization strategies, it considers recommending projects with less historical application data but potentially higher success rates, increasing the diversity of recommendations and providing enterprises with more valuable project application guidance.

[0035] Locality-Sensitive Hashing (LSH) is an algorithm for approximate nearest neighbor search in high-dimensional spaces. Its core idea is to design a special hash function so that data points that are close in the original space still have a high probability of being mapped to the same bucket after hashing, while data points that are far apart have a lower probability of being mapped to the same bucket. In data cleaning, LSH technology is used to quickly identify outliers and duplicates in data. For outlier detection, LSH maps high-dimensional data to a low-dimensional hash space. Similar data points are mapped to the same or adjacent hash buckets, while outliers that are far from most data points are mapped to sparse hash buckets, thus being quickly identified. For duplicate detection, the system compares the hash values ​​of data points, efficiently discovering data records that are highly similar but not completely identical. This is particularly important for processing enterprise data from different sources, as information from the same enterprise may have subtle differences in different systems. Compared to traditional pairwise comparison methods, LSH technology has a significant efficiency advantage when processing large-scale high-dimensional data, reducing the time complexity from O(n²) to near linear time O(n), which greatly improves the efficiency of data cleaning.

[0036] Median imputation is a technique for handling missing values ​​in data. It replaces missing values ​​in a specific feature or variable with the median of that feature or variable. Compared to mean imputation, median imputation is less sensitive to outliers and better preserves the data's distribution characteristics, especially when the data is skewed. In enterprise data processing, data such as financial indicators and market performance often exhibit skewed distributions (e.g., a few companies perform exceptionally well or poorly), and median imputation can provide more reasonable estimates. In practice, the system selects an appropriate imputation range based on the data's grouping characteristics. For example, it can calculate the median by grouping by industry, size, or region, and then use the median of the corresponding group to imput missing values, ensuring the reasonableness and representativeness of the imputed values. Furthermore, these imputed values ​​are labeled so that the reliability of the data can be considered in subsequent analyses. As a simple and effective method for handling missing values, median imputation minimizes disturbance to the original data distribution while ensuring data integrity, providing a reliable data foundation for subsequent scoring calculations.

[0037] Maximum-minimum normalization (MPN) is a commonly used data standardization method that maps original data to the [0,1] interval through linear transformation, making data with different dimensions and ranges comparable. The calculation formula is: Xnorm = (X - Xmin) / (Xmax - Xmin), where X is the original value, and Xmin and Xmax are the minimum and maximum values ​​of the indicator, respectively. This method preserves the distribution shape of the original data while unifying the range of all indicators, facilitating subsequent calculations and comparisons. In enterprise qualification assessment, MPN is particularly suitable for indicators with clear upper and lower bounds, such as scoring indicators and percentage indicators. However, it is sensitive to outliers; extreme values ​​may compress the distribution range of most normal values, so outlier handling is usually required before application. In this invention, a suitable normalization method is selected based on the characteristics of different indicators. For indicators with clear boundaries and relatively stable distributions, MPN is used to ensure that the performance of all enterprises on this indicator is mapped to a uniform scale, laying the foundation for fair comparison of multi-dimensional scores.

[0038] Z-score standardization (also known as standardized score or standard deviation standardization) is a method to transform data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The calculation formula is: Xstd = (X - μ) / σ, where μ is the mean and σ is the standard deviation. This method considers the distribution characteristics of the data and is particularly suitable for indicators that approximately follow a normal distribution, such as many financial ratios and growth rates. The advantage of Z-score standardization is that it considers the dispersion of the data, providing a reasonable relative position for indicators with different volatility, while effectively handling the impact of outliers. In enterprise qualification assessment, Z-score standardization can reflect a company's performance relative to the industry average on specific indicators; positive values ​​indicate above-average performance, negative values ​​indicate below-average performance, and larger absolute values ​​indicate greater deviation from the average. In this invention, Z-score standardization is used for indicators that require consideration of industry averages and distribution characteristics, ensuring that the company's score reflects not only its absolute performance but also its relative position within the industry, providing a basis for fair comparison of companies across different industries.

[0039] Information gain is an indicator that measures the contribution of a feature to a classification task. Originating from the concept of entropy in information theory, it represents the reduction in uncertainty of the system before and after the introduction of a feature. The formula is: IG(Y, X) = H(Y) - H(Y|X), where H(Y) is the entropy of the target variable Y, and H(Y|X) is the conditional entropy of Y given the feature X. A higher information gain indicates a greater contribution of the feature to classification or prediction. In the construction of an enterprise qualification assessment indicator system, information gain is used to evaluate the contribution of each indicator to the final assessment result, helping the system identify and retain key indicators with high information content and strong discriminative power. In practice, the information gain of each candidate indicator is calculated, and the indicators are sorted from highest to lowest information gain, prioritizing indicators with high information gain for inclusion in the assessment system. Furthermore, mutual information between indicators is considered to avoid selecting redundant indicator combinations, ensuring that the final indicator system is both concise and comprehensive. Through information gain analysis, the system can construct an assessment system containing the most discriminative indicators, improving the accuracy and efficiency of qualification assessment.

[0040] The variance contribution rate is an indicator that measures the extent to which a feature explains the variance of a target variable, reflecting the proportion of variability in the target variable that the feature can explain. In multi-feature models, the variance contribution rate can be calculated by dividing the partial regression sum of squares of the feature by the total regression sum of squares. The higher the variance contribution rate, the greater the contribution of that feature to explaining the volatility of the target variable, and the stronger its discriminative ability. In the validation of the enterprise qualification assessment indicator system, the variance contribution rate is used to analyze the contribution of each indicator to the overall assessment variance, helping the system determine the relative importance of each indicator. In practice, regression models between enterprise qualification scores and each indicator are constructed based on historical sample data, the variance contribution rate of each indicator is calculated, and then indicators with higher variance contribution rates are selected as key assessment indicators. Through variance contribution rate analysis, indicators that best reflect the differences between enterprises can be identified, optimizing the discriminative ability of the indicator system and ensuring that the assessment results can effectively distinguish enterprises with different qualification levels, providing a foundation for accurate project matching and qualification diagnosis.

[0041] In this embodiment, for the standardized enterprise multidimensional feature dataset, the improved WP-PVC algorithm is used to group and optimize the indicators and calculate their weights, constructing a multidimensional evaluation indicator system that includes financial performance, R&D capability level, industry position, and brand effect, including: S2.1: Based on the standardized enterprise multidimensional feature dataset, and combined with industry expert experience and policy guidance, select a candidate set of indicators that reflect the enterprise's financial and operational status, R&D capability level, industry position and brand effect, and establish an initial indicator pool. S2.2: Treat the enterprise evaluation indicators in the initial pool of indicators as vertices in the graph, the relationship between indicators as edges, and the strength of the relationship between indicators as the edge weight. Apply the improved WP-PVC algorithm to group the indicators, and divide the indicators into different groups including financial operations, R&D capabilities, and industry status to obtain indicator groups of different groups. S2.3: For the different groups of indicators, use the improved WP-PVC algorithm to set the minimum threshold that each indicator group needs to cover, solve the subset of indicators with the minimum weight, and determine the weight configuration of each indicator. S2.4: Based on the weight configuration of each indicator, verify the discrimination and predictive ability of the indicator system through historical sample data, calculate the information gain and variance contribution rate of each indicator, eliminate redundant indicators, and form the multi-dimensional evaluation indicator system.

[0042] Step S2.1 is the initial indicator pool construction stage. This step, based on a standardized multi-dimensional enterprise feature dataset and combined with industry expert experience and policy guidance, selects a candidate set of indicators reflecting a company's financial and operational status, R&D capabilities, industry position, and brand effect, thus establishing the initial indicator pool. In the specific implementation process, the key aspects for evaluating a company's qualifications are first determined from three main dimensions: financial and operational status, R&D capabilities, and industry position and brand effect. For the financial and operational status dimension, financial indicators such as debt-to-equity ratio, net profit margin, revenue growth rate, cash flow ratio, current ratio, and asset turnover are selected to reflect the company's profitability, solvency, operational efficiency, and growth potential. For the R&D capabilities dimension, indicators such as R&D investment ratio, number of patents (divided into invention patents, utility model patents, and design patents), patent quality (e.g., citation rate, authorization rate), technology transfer rate, proportion of R&D personnel, and technological leadership are selected to comprehensively evaluate the company's innovation capabilities and technological strength. For the industry position and brand effect dimension, indicators such as market share, brand awareness, customer satisfaction, media exposure, industry ranking, and product competitiveness are selected to reflect the company's position and influence in the market. In addition, indicators aligned with national strategies, such as performance indicators related to green and low-carbon development and digital transformation, will be included based on current policy guidance. Through expert consultation and literature review, and by integrating opinions from all parties, an initial pool of dozens of candidate indicators will be established to lay the foundation for subsequent indicator optimization.

[0043] Step S2.2 is the indicator grouping and partitioning stage. This step treats the enterprise evaluation indicators in the initial indicator pool as vertices in a graph, and the relationships between indicators as edges. Edge weights represent the strength of the relationship between indicators. An improved WP-PVC algorithm is applied to group the indicators into different groups, including financial operations, R&D capabilities, and industry position, resulting in different indicator groups. In the implementation process, firstly, correlation coefficients or mutual information values ​​between indicators are calculated based on historical data to construct an indicator correlation matrix. Higher correlation coefficients or mutual information values ​​indicate a strong correlation between two indicators, and these values ​​are used as the weights of the edges in the graph. For example, there may be a high correlation between "R&D investment ratio" and "number of patents," so the edges connecting them will be assigned higher weights; while the correlation between "debt-to-equity ratio" and "brand awareness" may be low, and the edges connecting them will have lower weights. These indicators and their relationships are constructed into a weighted undirected graph, and then the improved WP-PVC algorithm is applied for community detection and indicator grouping. This algorithm finds the optimal indicator grouping scheme by minimizing the connection cost between indicators within a group while maximizing the coverage benefit of indicators within the group. Specifically, the algorithm first assigns an initial group to each indicator based on semantic similarity and statistical correlation. Then, through iterative optimization, it adjusts the group assignments of the indicators until a globally optimal or near-optimal grouping scheme is found. Ultimately, all indicators are divided into different groups such as financial operations, R&D capabilities, and industry position. Indicators within each group have high internal correlation and can collectively reflect the company's performance in that dimension.

[0044] Step S2.3 is the indicator weight optimization calculation stage. This step uses the improved WP-PVC algorithm to set the minimum threshold that each indicator group needs to cover for different indicator groups, solves for the minimum weight subset of indicators, and determines the weight configuration of each indicator. In specific implementation, a coverage threshold is set for each indicator group, representing the minimum interpretability or information coverage that the selected indicators within that group need to achieve. This threshold can be set based on expert experience or determined through statistical methods, such as ensuring that the selected indicators can explain more than 80% of the variance of that dimension. Each indicator in the indicator group is considered a vertex in a graph, the importance or complexity of the indicator is considered the vertex weight (cost), the relationship between indicators is considered an edge, and the contribution of the indicator to that dimension is considered the edge weight (return). The goal of the improved WP-PVC algorithm is to find a vertex subset with the minimum total weight (lowest cost) such that the total weight (return) of the edges covered by this subset reaches at least the preset threshold. The algorithm solves this optimization problem using methods such as linear programming relaxation and rounding, double standard approximation, or greedy strategies to find the optimal indicator subset. For example, in the R&D capability dimension, key indicators such as "R&D investment ratio," "number of invention patents," and "technology transfer rate" might be selected, while indicators with redundant information or low contribution are discarded. Based on the selected subset of indicators, the relative weight of each indicator is further calculated. The weight value reflects the importance of the indicator in the evaluation of this dimension, providing a basis for subsequent scoring calculations.

[0045] Step S2.4 is the validity verification step of the indicator system. Based on the weight configuration of each indicator, this step verifies the discrimination and predictive ability of the indicator system using historical sample data, calculates the information gain and variance contribution rate of each indicator, eliminates redundant indicators, and forms a multi-dimensional evaluation indicator system. During implementation, the constructed indicator system is verified using historical sample data with known evaluation results (such as recognized high-tech enterprises, enterprises that have successfully applied for science and technology projects, etc.). First, the information gain of each indicator is calculated to assess its contribution to the final classification result. Information gain is based on the concept of entropy, calculating the reduction in uncertainty before and after the introduction of features, with the formula IG(Y,X) = H(Y) - H(Y|X), where H(Y) is the entropy of the target variable Y, and H(Y|X) is the conditional entropy of Y under known feature X. A higher information gain indicates a greater contribution of the indicator to the enterprise qualification assessment. Second, the variance contribution rate of each indicator is calculated to assess the indicator's explanatory power for the overall score variance. The variance contribution rate can be calculated using methods such as principal component analysis (PCA) or partial least squares regression (PLS), reflecting the extent to which an indicator explains the variation in the overall score. Indicators with high variance contribution rates have stronger discriminative power and can better differentiate between companies with different qualification levels. Based on the analysis results of information gain and variance contribution rate, redundant or low-contribution indicators are identified and eliminated, such as those with information gain below a preset threshold or those that are highly correlated with other indicators and have low variance contribution rates. Through repeated verification and optimization, a concise and efficient multi-dimensional evaluation indicator system is finally formed. This system contains the most representative and discriminative key indicators, which can accurately assess the comprehensive qualification level of enterprises.

[0046] Principal Component Analysis (PCA) is a commonly used dimensionality reduction technique and feature extraction method. It projects original high-dimensional data into a new low-dimensional space through linear transformation, maximizing the variance of the transformed data along the projection direction. In enterprise qualification assessment, PCA is used to analyze the variance contribution rate of each assessment indicator and identify the most discriminative key indicators. The principle of PCA is to calculate the eigenvalues ​​and eigenvectors of the data covariance matrix, sort the eigenvectors by eigenvalue, and select the eigenvectors corresponding to the k largest eigenvalues ​​as the projection direction, thus constructing a new feature space. The proportion of each eigenvalue to the total sum of eigenvalues ​​is the variance contribution rate of the corresponding principal component, reflecting the degree to which that principal component explains the variance of the original data. In validating the indicator system, the contribution of each indicator to the assessment results is determined by calculating the loading coefficient of each indicator in the principal components and the variance contribution rate of the corresponding principal components. Indicators with high variance contribution rates have stronger discriminative power and can better distinguish enterprises with different qualification levels. PCA analysis can scientifically optimize the indicator system, retaining those indicators with high information content and strong discriminative power, while eliminating those that are redundant or have low contribution, thereby improving the efficiency and accuracy of the evaluation.

[0047] Partial Least Squares Regression (PLS) is a statistical method that combines principal component analysis (PCA) and multiple linear regression, particularly suitable for handling cases of multicollinearity among independent variables. Unlike PCA, PLS considers information from both the independent variable matrix X and the dependent variable matrix Y during dimensionality reduction, seeking the latent structure of X that best explains the variation in Y. In enterprise qualification assessment, PLS is used to analyze the contribution of each assessment indicator to the final assessment result, especially when there is a high correlation between indicators. The basic principle of PLS ​​is to project X and Y into the latent variable space through an iterative algorithm, so that the latent variables can both explain the variation in X to the greatest extent and have the strongest correlation with Y. By calculating the Variable Importance in Projection (VIP) value of each indicator in the PLS model, the contribution of each indicator to the assessment result can be quantified. Indicators with a VIP value greater than 1 are generally considered important indicators and should be retained in the assessment system; while indicators with a VIP value less than 0.5 may contribute little and can be considered for removal. PLS analysis allows us to identify key indicators that truly have a significant impact on the evaluation results, based on the interrelationships between indicators, and to construct a more concise and effective evaluation indicator system.

[0048] Community detection is an important problem in graph theory and network analysis, aiming to identify tightly connected groups of nodes (communities or modules) in a network, where the connection density between nodes within a community is high, while the connection density between communities is low. In enterprise qualification assessment indicator grouping, community detection techniques are used to classify indicators into different groups, such as financial operations, R&D capabilities, and industry status. Commonly used community detection algorithms include the Louvain algorithm based on modularity optimization, the Infomap algorithm based on information flow, and the Label Propagation Algorithm (LPA). In this invention, the improved WP-PVC algorithm incorporates the idea of ​​community detection, treating indicators as nodes in a graph and the relationships between indicators as edges. By optimizing the objective function (minimizing the connection cost between indicators within a group while maximizing the coverage benefit of indicators within a group), scientific grouping of indicators is achieved. The key to community detection lies in defining an appropriate objective function and selecting an effective optimization algorithm. Within the WP-PVC framework, by setting minimum coverage thresholds for different groups, it is ensured that each indicator group fully reflects the company's capabilities in that dimension. Simultaneously, by minimizing the total weight, the indicator groups are kept concise and efficient. Community discovery technology enables the construction of well-structured and internally consistent indicator groups, providing a scientific basis for dimension scoring calculations.

[0049] Lagrangian relaxation is a mathematical technique for solving constrained optimization problems. It transforms the problem into an unconstrained or less constrained problem by incorporating the constraints into the objective function. In the dual-standard WP-PVC algorithm, Lagrangian relaxation is used to handle two competing objectives: cost minimization (minimizing the total weight of the selected indices) and profit maximization (ensuring sufficient information coverage). The basic idea of ​​Lagrangian relaxation is to introduce the Lagrange multiplier λ, transforming the original constrained optimization problem—minimizing f(x) with the constraint g(x) ≤ 0—into a Lagrangian function. The optimization problem of the index weights is described in the bi-standard WP-PVC algorithm as: minimizing... Constraints Here, x(v) represents whether to select indicator v, y(e) represents whether to cover association e, w(v) is the weight (cost) of the indicator, p(e) is the benefit of the association, and Ti is the minimum coverage threshold for group i. By introducing the Lagrange multiplier λi, the problem is transformed into minimizing By adjusting the value of λi, a balance is found between the two objectives: ensuring that the selected indicator set has a small weight (low cost) while ensuring sufficient information coverage (high return). Lagrange relaxation enables the bi-standard WP-PVC algorithm to find near-optimal solutions in complex multi-objective optimization problems, providing a scientific method for indicator selection and weight calculation in enterprise qualification assessment.

[0050] Furthermore, the method of using the improved WP-PVC algorithm to group and optimize indicators and calculate weights also includes: S2.5: Based on the characteristics of different industries and the development stage of enterprises, the multi-dimensional evaluation index system is adaptively adjusted to generate customized evaluation index systems for different types of enterprises; The adaptive adjustment includes adjusting the weights of indicators in each dimension according to the characteristics of the industry to which the enterprise belongs, adjusting the evaluation focus according to the development stage of the enterprise, and dynamically adjusting the indicator weights according to the current policy environment and support direction.

[0051] Step S2.5 is the adaptive indicator system generation stage. This step adaptively adjusts the multi-dimensional evaluation indicator system based on the characteristics of different industries and the development stage of enterprises, generating customized evaluation indicator systems for different types of enterprises. In specific implementation, firstly, the weights of each dimension indicator are adjusted according to the characteristics of the industry to which the enterprise belongs. The key competitive points differ across industries. For example, for high-tech industries (such as chips and new materials), the weight of the R&D capability dimension will be increased because R&D innovation is the core competitiveness of enterprises in this industry; while for consumer goods industries (such as food and clothing), the brand effect dimension may receive higher weight because brand influence is crucial to the success of such enterprises. Secondly, the evaluation focus is adjusted according to the development stage of the enterprise. For start-up enterprises, more attention may be paid to their innovation capabilities, team structure, and growth potential, while the requirements for financial stability and market share are reduced; for growth-stage enterprises, more emphasis is placed on their expansion capabilities, revenue growth rate, and market penetration rate; for mature enterprises, more emphasis may be placed on their stable profitability, market position, and sustainable development potential. Furthermore, the indicator weights are dynamically adjusted according to the current policy environment and support directions to make the evaluation results more aligned with policy guidance. For example, when the country strongly supports green and low-carbon development, the weight of environmental protection-related indicators will be increased; when policies encourage digital transformation, the weight of digitalization indicators will be increased accordingly. This multi-dimensional adaptive adjustment ensures that the evaluation system can accurately capture the core competitiveness and development potential of different types of enterprises, improving the relevance and practicality of the evaluation results.

[0052] In this embodiment, based on the multi-dimensional evaluation index system, the dual-standard WP-PVC algorithm is used to perform correlation analysis and scoring calculation among the indicators, generating a comprehensive enterprise qualification score result that includes a comprehensive enterprise qualification score and multi-dimensional scoring results, including: S3.1: Standardize the data of various indicators of enterprises in the multi-dimensional evaluation index system to eliminate the influence of dimensions, make different indicators comparable, and obtain standardized indicator data. S3.2: Based on the standardized indicator data, the dual-standard WP-PVC algorithm is applied to analyze the correlation between indicators within the same dimension, construct an indicator correlation network within the dimension, calculate the correlation coefficient between indicators, identify key indicators and auxiliary indicators, and determine the importance ranking of indicators within the dimension by combining the optimized weights of each indicator in the multi-dimensional evaluation indicator system in the evaluation system. S3.3: Based on the importance ranking of the indicators and the optimization weights, the standardized value of the enterprise on each indicator is multiplied by the corresponding optimization weight, non-linear adjustment is performed, and the results are summarized according to the weighted average rule to calculate the enterprise's scores in each dimension of financial operation, R&D capability, and industry position, forming a multi-dimensional scoring result. S3.4: Apply the dual-standard WP-PVC algorithm to dynamically adjust the weights of each dimension based on their contribution to the overall enterprise qualification, taking into account industry characteristics, enterprise type, and application project characteristics. Then, calculate the weighted average of the scores for each dimension according to the optimized weights to generate the overall enterprise qualification score, which includes the overall enterprise qualification score and the scores for each dimension.

[0053] Step S3.1 is the standardization process for indicator data. This step standardizes the data of various enterprise indicators in the multi-dimensional evaluation indicator system, eliminating the influence of dimensions and making different indicators comparable, thus obtaining standardized indicator data. During implementation, the data type and distribution characteristics of each indicator are first identified, and an appropriate standardization method is selected. For continuous numerical indicators (such as financial ratios, R&D investment ratios, etc.), two main standardization methods are used: maximum-minimum normalization and Z-score standardization. Maximum-minimum normalization linearly maps the data to the [0,1] interval, suitable for indicators with clear upper and lower bounds, with the formula Xnorm = (X - Xmin) / (Xmax - Xmin). Z-score standardization converts the data into a standard normal distribution with a mean of 0 and a standard deviation of 1, suitable for indicators that approximately follow a normal distribution, with the formula Xstd = (X - μ) / σ. For graded indicators (such as credit rating, qualification rating, etc.), a clear numerical mapping relationship is established, for example, mapping AAA, AA, and A grades to values ​​of 1.0, 0.8, and 0.6, respectively. For Boolean indicators (such as whether a certain qualification is possessed), they are typically converted to binary values ​​of 0 / 1. Furthermore, the directionality of the indicators is considered. Different standardization processes are used for positive indicators (such as net profit margin) that are "the larger the better" and negative indicators (such as debt-to-equity ratio) that are "the smaller the better," ensuring that all indicators, after standardization, follow a unified "the larger the better" direction. For highly skewed indicators, logarithmic or other non-linear transformations may be performed before standardization to make the data distribution more reasonable. Through these standardization processes, indicator data of different dimensions and ranges are transformed to a comparable scale, laying the foundation for subsequent correlation analysis and scoring calculations.

[0054] Step S3.2 is the intra-dimensional indicator correlation analysis step. Based on standardized indicator data, this step applies the bi-standard WP-PVC algorithm to analyze the correlation between indicators within the same dimension, constructing an intra-dimensional indicator correlation network, calculating correlation coefficients between indicators, identifying key and auxiliary indicators, and determining the importance ranking of indicators within the dimension by combining the optimized weights of each indicator in the multi-dimensional evaluation indicator system. In specific implementation, firstly, an indicator correlation network is constructed for each dimension (e.g., financial operations, R&D capabilities, industry position), calculating the correlation coefficients or mutual information values ​​between indicators within the dimension to form a correlation matrix. Through this correlation data, the correlation structure between indicators can be visualized, identifying which indicators have strong inter-correlation. Then, the bi-standard WP-PVC algorithm is applied to identify key indicators. In the bi-standard WP-PVC problem, indicators are considered as vertices in a graph, the correlation between indicators is considered as edges, the complexity or acquisition cost of an indicator is considered as vertex weight, and the correlation strength between indicators is considered as edge weight. The algorithm aims to find a subset of vertices (indicator subsets) that minimizes the total weight while satisfying a certain coverage threshold. By solving this optimization problem, we can identify the core indicators (those with high information content and strong independence) and auxiliary indicators (those highly correlated with the core indicators) within a dimension. The dual-standard algorithm is characterized by its ability to simultaneously consider both cost minimization and coverage maximization, achieving a balance between the two. It also determines the influence paths and weight transfer relationships between indicators, constructing a structured indicator evaluation network. Finally, by combining the optimized weights of the indicators in the evaluation system (calculated in step S2.3), the final importance ranking of the indicators within the dimension is determined, providing a basis for subsequent scoring calculations.

[0055] Step S3.3 is the dimensional scoring calculation step. This step, based on the importance ranking of indicators and optimized weights, multiplies the standardized value of the enterprise on each indicator by the corresponding optimized weight, performs non-linear adjustments, and summarizes the results according to a weighted average rule to calculate the enterprise's scores in each dimension: financial operations, R&D capabilities, and industry position, forming a multi-dimensional scoring result. In the implementation process, firstly, the standardized values ​​of the enterprise on each indicator are obtained; these values ​​reflect the enterprise's original performance in each aspect. Then, these standardized values ​​are multiplied by the indicator importance weights determined in step S3.2 to obtain the weighted indicator scores. Considering that some indicators may have threshold effects or non-linear effects, non-linear adjustments are made to the scores of some indicators. For example, the R&D investment ratio may have an optimal range; too low indicates insufficient innovation, while too high may indicate unreasonable resource allocation. The debt-to-equity ratio has similar characteristics; too low may mean low capital utilization efficiency, while too high may bring financial risks. By setting appropriate non-linear transformation functions (such as piecewise linear functions, Gaussian functions, or S-shaped functions), the scores of these indicators are adjusted to more accurately reflect the enterprise's true performance on that indicator. After adjustment, the scores of all indicators within each dimension are aggregated according to a weighted average rule to calculate the company's total score across dimensions such as financial operations, R&D capabilities, and industry standing. Furthermore, the company's dimensional scores are compared with industry benchmarks or historical best practices to provide an assessment of its relative position within the industry. Ultimately, the scores for each dimension are generated, providing a clear visual representation of the company's strengths and weaknesses in various aspects and forming the basis for calculating the overall score.

[0056] Step S3.4 is the inter-dimensional weight optimization step. This step applies the dual-standard WP-PVC algorithm to dynamically adjust the weights of each dimension based on their contribution to the overall enterprise qualification, taking into account industry characteristics, enterprise type, and application project characteristics. The scores for each dimension are then weighted according to the optimized weights to generate a comprehensive enterprise qualification score that includes both the overall enterprise qualification score and the scores for each dimension. In practice, firstly, based on historical data analysis, the contribution of different dimensions (such as financial operations, R&D capabilities, and industry position) to the overall success of the enterprise is established, creating a basic model of dimension importance. Then, considering the characteristics of the industry in which the enterprise operates, the dimension weights are adjusted. Different industries have different key competitive factors. For example, in the high-tech industry, R&D capability may be the most critical success factor; in the FMCG industry, brand effect may be more important; in manufacturing, production efficiency and cost control may be more decisive. The relative importance of each dimension is adjusted according to industry characteristics. Secondly, considering the type and development stage of the enterprise, the weights are further adjusted. For example, for startups, the weights of innovation capability and growth potential may be increased; for mature enterprises, the weights of market position and stability may be higher. Furthermore, the characteristics of the projects applied for by enterprises are analyzed, and weights are adjusted to suit the specific application requirements of each project. A dual-standard WP-PVC algorithm is applied for weight optimization. The algorithm aims to find a dimensional weight configuration scheme that most closely approximates the evaluation results with historical successful cases, while ensuring the minimum coverage threshold for each dimension (ensuring comprehensive evaluation). Finally, the enterprise's scores on each dimension are weighted according to the optimized weights to generate a comprehensive qualification score for the enterprise. Combining the detailed scores for each dimension, a complete score result containing both the comprehensive qualification score and the dimensional scores is output, providing comprehensive data support for subsequent project matching and qualification diagnosis.

[0057] Furthermore, the dynamic adjustment of the weights of each dimension also includes: S3.4.1: Based on the contribution of each dimension to the success of enterprises in different industries, analyze the impact of industry characteristics on dimension weights and obtain the industry characteristic weight adjustment coefficient; S3.4.2: Based on historical successful cases of different types of science and technology project applications, analyze the performance characteristics of each dimension to obtain the project orientation weight adjustment coefficient; S3.4.3: Based on the industry characteristic weight adjustment coefficient and the project orientation weight adjustment coefficient, by setting the minimum coverage threshold for each dimension, find the optimal weight allocation scheme so that the evaluation results are closest to historical successful cases, and generate the optimized dimension weight configuration.

[0058] Step S3.4.1 is the calculation of industry characteristic weight adjustment coefficients. This step analyzes the impact of industry characteristics on dimension weights based on the contribution of each dimension to enterprise success in different industries, and obtains industry characteristic weight adjustment coefficients. In the specific implementation process, a large amount of historical industry data is first collected and analyzed, including multi-dimensional performance indicators and ultimate success indicators (such as whether projects were successfully applied for, whether high-tech enterprise certification was obtained, etc.) for enterprises in different industries. Through statistical analysis and machine learning methods, a correlation model between dimension performance and enterprise success in each industry is established. For example, it may be found that in the integrated circuit industry, the correlation between R&D capability dimension and enterprise success reaches 0.8, while in traditional manufacturing, this correlation may only be 0.5. Based on these correlation strength data, a benchmark model of dimension importance is established for different industries. Then, the industry to which the enterprise belongs is matched with these benchmark models, and the corresponding dimension importance coefficients are extracted. To handle cases with blurred industry boundaries or cross-industry enterprises, fuzzy clustering or multi-label classification methods are used, allowing enterprises to belong to multiple industry categories simultaneously, and the impact of industry characteristics is calculated according to similarity weights. Finally, a set of industry-specific weight adjustment coefficients is generated. These coefficients reflect the degree of influence of the company's industry on the importance of each dimension. For example, the industry adjustment coefficient for the R&D capability dimension might be 1.2 (indicating that R&D capability is more important than usual in this industry), while the adjustment coefficient for the financial and operational dimension might be 0.9 (indicating slightly lower than usual importance). These adjustment coefficients will play a crucial role in subsequent dimension weight optimization, ensuring that the evaluation results fully consider the impact of industry characteristics.

[0059] Step S3.4.2 is the calculation of project-oriented weight adjustment coefficients. This step analyzes the performance characteristics of each dimension based on historical successful cases of different types of science and technology project applications to obtain project-oriented weight adjustment coefficients. In the implementation process, a database containing historical science and technology project application data is first constructed, recording the multi-dimensional performance indicators of enterprises and the project application results (success or failure). These projects are classified according to characteristics such as type, field, and funding intensity, such as basic research projects, technology development projects, technology transfer projects, and industrialization promotion projects. For each type of project, the performance differences of successful and unsuccessful applicants in various dimensions (financial operations, R&D capabilities, industry position, etc.) are analyzed to identify the key dimensions most decisive for project application success. For example, it may be found that for basic research projects, R&D capability is a key factor for success, with a correlation coefficient of 0.85; while for industrialization promotion projects, market position and financial stability may be more important, with correlation coefficients of 0.75 and 0.7, respectively. By comparing the characteristic patterns of different types of projects, a mapping relationship between project type and dimension importance is generated. When a company prepares to apply for a specific type of project, it extracts the corresponding dimension importance coefficients based on the project type, which serve as adjustment coefficients for project orientation weights. For example, if the company's goal is to apply for technology research and development projects, the project orientation adjustment coefficient for the R&D capability dimension might be 1.3, for the financial and operational dimension it might be 0.8, and for the industry position dimension it might be 0.7. These adjustment coefficients reflect the emphasis different types of projects place on the company's capabilities in each dimension, providing a basis for subsequent optimization of dimension weights and project orientation adjustments.

[0060] Step S3.4.3 optimizes the dimension weight configuration. Based on industry-specific weight adjustment coefficients and project-oriented weight adjustment coefficients, this step finds the optimal weight allocation scheme by setting a minimum coverage threshold for each dimension. This ensures the evaluation results most closely resemble historical success stories, generating the optimized dimension weight configuration. In practice, firstly, the preliminary dimension weight adjustment values ​​are calculated using the industry-specific adjustment coefficients obtained in step S3.4.1 and the project-oriented adjustment coefficients obtained in step S3.4.2. This calculation can employ a product method (e.g., wadj = wbase × coefindustry × coefproject) or a weighted average method (e.g., wadj = wbase × (α × coefindustry + β × coefproject), where α and β are parameters reflecting the relative importance of industry and project factors). Then, a minimum coverage threshold is set for each dimension to ensure the evaluation results adequately reflect the company's capabilities across all aspects, preventing certain important dimensions from being completely ignored. These thresholds can be set based on expert experience or learned from historical data using statistical methods, such as ensuring each dimension accounts for at least 15% of the total score. Next, an optimization model is constructed. The objective function is to minimize the difference between the predicted score and the scores of historical successful cases, with the constraints that the sum of the weights of each dimension is 1 and that the weight of each dimension is not lower than its minimum coverage threshold. This optimization problem can be solved using linear programming, quadratic programming, or heuristic algorithms (such as genetic algorithms and simulated annealing). The weight configuration scheme that is closest to historical successful cases will be selected, while ensuring the comprehensiveness and balance of the evaluation. Finally, a set of optimized dimension weight configurations is generated. These weights reflect the combined influence of the characteristics of the company's industry and the requirements of the target project on the importance of each dimension, providing a scientific weight basis for the comprehensive score calculation of the company's qualifications, ensuring that the evaluation results are both consistent with industry characteristics and meet the specific requirements of the project application.

[0061] Fuzzy clustering is a clustering analysis method that allows data points to belong to multiple clusters simultaneously. Unlike traditional hard clustering, fuzzy clustering assigns a membership vector to each data point, representing the degree to which it belongs to each cluster. In enterprise qualification assessment, fuzzy clustering is used to handle situations where industry boundaries are ambiguous or where enterprises span multiple industries. It allows enterprises to belong to multiple industry categories simultaneously and calculates the impact of industry characteristics based on similarity weights. In adjusting industry characteristic weights, firstly, based on the enterprise's multidimensional feature data, the fuzzy clustering algorithm is used to classify the enterprise into different industry categories, obtaining the enterprise's membership degree to each industry. Then, combining the dimensional importance coefficients of each industry and the enterprise's industry membership degree, a weighted average industry characteristic weight adjustment coefficient is calculated. For example, if an enterprise's membership degree to the electronics and information industry is 0.7 and its membership degree to the new materials industry is 0.3, then the industry adjustment coefficient for the R&D capability dimension might be 0.7 × 1.3 + 0.3 × 1.1 = 1.24. Through fuzzy clustering, situations with ambiguous industry boundaries can be handled more accurately, making the assessment results more consistent with the actual situation of the enterprise.

[0062] Multi-label classification is a machine learning task in which each instance can belong to multiple categories or labels simultaneously. Unlike traditional multi-class classification, which requires each instance to belong to only one category, multi-label classification allows an instance to have multiple labels. In enterprise qualification assessment, multi-label classification is used to handle situations where an enterprise may belong to multiple industries or business areas simultaneously. It assigns multiple industry labels to each enterprise and calculates the comprehensive impact of industry characteristics based on these labels. Commonly used multi-label classification algorithms include binary association, classifier chains, label power sets, and algorithm adaptation. In industry characteristic weight adjustment, a multi-label classification model is first constructed. Based on the enterprise's feature data (such as main business description, product type, technical field, etc.), it predicts which industry categories the enterprise belongs to and calculates the confidence or probability of each category. Then, combining the dimensional importance coefficients of each industry and the confidence of the industry labels, a weighted average industry characteristic weight adjustment coefficient is calculated. Compared to traditional single-industry classification methods, multi-label classification can more accurately capture the diversified business characteristics of enterprises, providing a more reasonable basis for weight adjustment for assessments across different dimensions, and improving the accuracy and relevance of qualification assessments.

[0063] like Figure 2 As shown, the process of calculating the matching degree between enterprises and various technology projects based on the comprehensive score of enterprise qualifications, combined with a multi-logarithm algorithm for random online sorting, and outputting a project matching recommendation list and application success rate prediction includes: S4.1: Collect information on the application conditions, review standards, and funding amounts for various science and technology projects and certifications; establish a structured science and technology project database; and classify and label projects in multiple dimensions, including industry, technology direction, and support stage, to build a science and technology project database that is updated in real time. S4.2: Based on the comprehensive score of the enterprise qualifications and the real-time updated technology project database, the enterprise characteristics and project requirements are transformed into multi-dimensional vectors, the matching degree is calculated, and a matching model between enterprises and projects is constructed. S4.3: Based on the matching model and successful cases of matching enterprise characteristics with projects in historical application data, apply a random online sorting multi-logarithm algorithm to optimize the project recommendation ranking and generate an optimized project matching sequence; S4.4: Based on the optimized project matching sequence and historical application case analysis, construct an application success rate prediction model, calculate the success probability of enterprises applying for various types of projects, provide enterprises with application decision reference, and generate application success rate prediction results; S4.5: Based on the optimized project matching sequence and the application success rate prediction results, generate a project recommendation list suitable for the enterprise to apply for, sort it according to the success rate, and attach matching reasons and application suggestions, and output the project matching recommendation list and application success rate prediction.

[0064] Step S4.1 is the construction of the science and technology project database. This step collects information on various science and technology projects and certifications, including application requirements, review standards, and funding amounts. A structured database of science and technology projects is established, and projects are categorized and labeled in multiple dimensions, including industry sector, technical direction, and support stage, to create a real-time updated database. In the specific implementation process, science and technology project information is first collected through various channels, including official websites of national, provincial, and municipal science and technology management departments, science and technology project management documents, annual science and technology plan guidelines, and public announcements of projects from previous years. The collected project information covers core elements such as project name, application time, support direction, funding amount, application requirements, and review standards. For unstructured project description text, natural language processing (NLP) technology is used to extract information, converting the unstructured text into structured data. NLP technology mainly includes named entity recognition (NAME) and relation extraction. The former is used to identify entities such as project type, applicant, and funding amount in the text, while the latter is used to extract relationships between entities, such as "application requirements" and "review standards." Next, the collected projects were categorized and labeled in multiple dimensions, including industry sector (such as information technology, advanced manufacturing, and biomedicine), technology direction (such as artificial intelligence, new materials, and green energy), support stage (such as basic research, applied research and development, technology transfer, and industrialization), and funding intensity (such as major, key, and general projects). These labels can be automatically generated through rule matching and keyword extraction, or manually corrected by experts to ensure the accuracy of the classification. To ensure the timeliness and accuracy of the project database, a regular update mechanism was established. Web crawlers regularly retrieve the latest project information from various sources, and an information change detection mechanism was also set up to promptly identify and update changes in project application conditions and review standards. Through these measures, a comprehensive, clearly structured, and real-time updated science and technology project database was constructed, providing data support for subsequent project matching and analysis.

[0065] Step S4.2 is the enterprise-project matching model construction stage. Based on the comprehensive enterprise qualification scoring results and a real-time updated technology project database, this step transforms enterprise characteristics and project requirements into multi-dimensional vectors, calculates the matching degree, and constructs an enterprise-project matching model. In the implementation process, a suitable representation method for enterprise-project matching is first designed to transform the enterprise qualification scoring results and project application conditions into comparable multi-dimensional feature vectors. For the enterprise feature vector, scores are extracted for various dimensions such as financial operations, R&D capabilities, and industry status, as well as basic characteristics such as enterprise size, years of establishment, and industry. For the project feature vector, core elements such as project application conditions, review criteria, and funding directions are extracted and transformed into structured element-value pairs. To handle ambiguous conditions in text descriptions, semantic similarity calculation technology is introduced. This technology calculates the semantic similarity between texts using word vector models (such as Word2Vec and GloVe) or pre-trained language models (such as BERT and RoBERTa), converting ambiguous text descriptions into numerical similarity indicators. Next, a matching degree calculation method is designed to calculate the similarity between the enterprise feature vector and the project requirement vector. The basic matching degree calculation uses weighted cosine similarity or Euclidean distance, assigning different weights to key and non-key conditions. A threshold condition screening mechanism is also implemented, prioritizing the checking of projects based on hard conditions (such as company size restrictions, years of establishment requirements, etc.), and directly excluding company-project pairs that do not meet the threshold conditions. For complex matching relationships, a multi-level matching strategy is adopted: first, the matching degree is calculated separately on multiple dimensions such as financial indicators, R&D capabilities, and market performance; then, the matching degrees of these dimensions are combined to obtain an overall matching score. To improve matching accuracy, a learning mechanism based on historical application data is introduced, continuously adjusting the parameters and weights of the matching algorithm by analyzing historically successful and unsuccessful company-project pairs. Through these methods, an accurate and interpretable company-project matching model is constructed, laying the foundation for subsequent project recommendations and success rate prediction.

[0066] Step S4.3 is the project recommendation optimization stage. This step optimizes the project recommendation ranking based on successful cases of matching enterprise characteristics with projects in the matching model and historical application data, applying a random online sorting multi-logarithmic algorithm to generate an optimized project matching sequence. In practice, firstly, enterprise characteristic data of historically successful applications for various projects are collected and analyzed to establish a feature-success rate correlation model. The random online sorting multi-logarithmic algorithm (ROSMLM) is an algorithm that combines random sampling, online learning, and multi-logarithmic models, particularly suitable for handling ranking problems with time dynamics and multiple factors. The core idea of ​​this algorithm is to construct multiple sub-models through multiple random samplings. Each sub-model processes the relationship between features and success rates based on logarithmic transformation (logarithmic transformation can effectively handle long-tailed data, making the model more stable). Then, the prediction results of these sub-models are combined to form the final ranking decision. In practice, firstly, multiple data subsets are randomly extracted from historical data, each subset containing project application data of different periods and types; then, a log-linear model is constructed for each subset, in the form of... Where p is the probability of successful application, and X1 to X... n The variables are enterprise characteristics, and β is the model parameter. Next, the predictions from all sub-models are integrated, and a comprehensive prediction result is obtained through weighted averaging or a voting mechanism. To balance "exploration" and "utilization," the ranking not only considers projects with high predicted success rates but also, with a certain probability, projects with less historical data but potentially higher success rates. This strategy is similar to the e-greedy epsilon-greedy strategy or the Upper Confidence Bound (UCB) algorithm. It also possesses online learning capabilities, continuously adjusting model parameters based on the latest application results to ensure the recommendation ranking is up-to-date. Through these technologies, an optimized project matching sequence that considers both matching accuracy and diversity is generated, providing enterprises with a more comprehensive reference for project selection.

[0067] Step S4.4 is the application success rate prediction stage. Based on optimized project matching sequences and historical application case analysis, this step constructs an application success rate prediction model to calculate the probability of success for enterprises applying for various projects, providing a reference for application decisions and generating application success rate prediction results. In implementation, a comprehensive success rate prediction model considering multiple factors is first constructed. Unlike traditional binary classification models, this model considers not only the enterprise's own qualifications (such as financial indicators, R&D capabilities, market performance, etc.), but also project characteristics (such as the intensity of competition, funding intensity, technical threshold, etc.), time factors (such as policy cycles, economic environment), and the matching degree between the enterprise and the project, among other multi-dimensional factors. A Probabilistic Graphical Model (PGM) is used to construct this complex causal network. PGM uses directed or undirected graphs to represent the conditional dependencies between random variables, effectively modeling uncertainties under the interaction of multiple factors. Within the PGM framework, Bayesian networks or conditional random fields are used to construct the prediction model, and model parameters are learned through historical data. To improve prediction accuracy, an ensemble learning approach is employed, combining the prediction results of multiple base models (such as logistic regression, decision trees, and support vector machines) and deriving the final prediction through weighted voting or stacking. Unlike simply outputting a fixed success rate value, this approach provides a success rate range estimate. By calculating the confidence interval of the prediction, it indicates the degree of uncertainty, such as "The success rate of this project application is between 65% and 75%, with a confidence level of 90%." Sensitivity analysis is also performed, observing changes in the output results by altering the values ​​of input variables to identify the key factors that have the greatest impact on the success rate, providing targeted improvement suggestions for enterprises. Finally, a project success rate prediction result is generated, including the predicted success rate, confidence interval, key influencing factors, and improvement suggestions, providing a scientific basis for enterprises' project application decisions.

[0068] Step S4.5 is the project recommendation list generation stage. Based on the optimized project matching sequence and application success rate prediction results, this step generates a project recommendation list suitable for the company's applications, sorted by success rate from highest to lowest, and includes matching reasons and application suggestions. The output is the project matching recommendation list and the predicted application success rate. In practice, the recommended projects are first ranked by comprehensively considering multiple dimensions. Besides predicting the success rate, factors such as project funding amount, application difficulty, application time window, and alignment with the company's development strategy are also considered, constructing a multi-objective ranking model. This multi-objective ranking model employs the Pareto optimality principle, seeking a balance between multiple potentially conflicting objectives to ensure that the recommended projects have good overall performance in all aspects. To enhance the diversity and coverage of recommendations, a grouping recommendation strategy is adopted, categorizing recommended projects according to different dimensions, including success rate (e.g., high success rate, medium success rate, challenging projects), project type (e.g., basic research, technology development, technology transfer, industrialization), application difficulty (e.g., simple application, moderate difficulty, complex application), and application time (e.g., soon-to-be-applied, mid-term planning, long-term goals). Detailed matching reasons and application suggestions are also generated for each recommended project. The matching reasons section analyzes in which aspects the company meets the project requirements and in which aspects it has advantages, making the recommendation explainable; the application suggestions section provides specific application preparation advice, including key points for material preparation, application time planning, and precautions, making the recommendation actionable. Furthermore, based on the principle of collaborative filtering, the successful application patterns of other companies similar to the target company are analyzed, providing reference information such as "similar companies have also applied...", broadening the recommendation perspective. Finally, a structured project matching recommendation list and application success rate prediction are output, providing comprehensive and accurate support for companies' project application decisions.

[0069] Semantic similarity calculation is a technique for measuring the semantic similarity between two texts. In enterprise-project matching models, it is used to handle ambiguous conditions in text descriptions, converting them into quantifiable similarity metrics. Methods for semantic similarity calculation mainly include word vector-based methods and pre-trained language model-based methods. Word vector models (such as Word2Vec, GloVe, and FastText) map words to a low-dimensional dense vector space and measure word similarity by calculating the cosine similarity between vectors. Text similarity can be calculated using a weighted average of word vectors or other aggregation methods. Pre-trained language models (such as BERT bidirectional encoder representation, RoBERTa robust optimization, and XLNet extended language network) capture contextual semantic information of text through self-supervised learning, generating context-relevant text representations and capturing semantic information more accurately. In this paper, a BERT-based semantic similarity calculation method is adopted. First, a domain-adaptive fine-tuned BERT model is used to generate semantic representation vectors for enterprise descriptions and project requirements. Then, the similarity between these vectors (such as cosine similarity, dot product, or Euclidean distance) is calculated. For example, when a project requirement is described as "having independently developed core technologies," while a company is described as "possessing multiple invention patents and proprietary technologies," their high semantic relevance can still be identified through semantic similarity calculation, despite the slight discrepancy, thus supporting the matching degree calculation. Furthermore, a domain-knowledge-enhanced semantic matching method is introduced, incorporating expertise from the fields of technology projects and company evaluation to improve the accuracy and relevance of semantic understanding.

[0070] Probabilistic Graphical Models (PGMs) are a class of statistical models that represent the conditional dependencies between random variables using a graph structure. In application success rate prediction, they are used to construct complex causal networks to simulate uncertainties caused by the interaction of multiple factors. PGMs mainly include two categories: Bayesian Networks and Markov Random Fields (MRFs). A Bayesian Network is a directed acyclic graph model where nodes represent random variables and edges represent conditional dependencies. Each node has a conditional probability table representing the probability distribution of that node when its parent node takes a specific value. A Markov Random Field is an undirected graphical model that defines the interaction between variables through a potential function. In application success rate prediction, a Bayesian network was constructed, using variables such as company characteristics (e.g., financial status, R&D capabilities), project characteristics (e.g., intensity of competition, technological barriers), and environmental factors (e.g., policy guidance, economic environment) as network nodes. The conditional probability relationships between nodes were learned through historical data. This graph-based approach can intuitively represent the causal relationships between variables and handle uncertainty and missing values ​​in the data. For example, it can be inferred that companies with "high R&D investment but average patent quality" have a higher probability of success in projects with "intense competition but strong policy support." Furthermore, PGM is highly interpretable, visually demonstrating the impact of various factors on success rates through a graph structure, providing companies with more transparent predictive results.

[0071] Sensitivity analysis is a method to assess the influence of different factors by changing the values ​​of input variables in a model and observing changes in the output. In the application success rate prediction stage, sensitivity analysis is used to identify the key factors that have the greatest impact on the success rate, providing targeted improvement suggestions for enterprises. The main methods of sensitivity analysis include single-factor sensitivity analysis and global sensitivity analysis. Single-factor sensitivity analysis changes only one input variable at a time, keeping other variables constant, and observes the changes in the output; global sensitivity analysis considers the changes of multiple input variables and their interactions, such as variance decomposition or regression-based methods. In this study, a two-stage sensitivity analysis strategy based on the Morris screening method and the Sobol index was adopted: first, the computationally efficient Morris screening method was used to initially identify important factors; then, the more accurate but complex Sobol method was applied to these factors for in-depth analysis. Through sensitivity analysis, the influence of different factors on the application success rate can be quantified. For example, it was found that for every 1 percentage point increase in the "R&D investment ratio," the success rate increased by an average of 3.5 percentage points; while for every 5 percentage point decrease in the "debt-to-equity ratio," the success rate increased by an average of 2 percentage points. These quantitative results help companies clarify improvement directions and priorities, concentrate resources on improving the most impactful key factors, and maximize the success rate and effectiveness. At the same time, sensitivity analysis can identify robustness and vulnerabilities in the model, improving the reliability and interpretability of prediction results.

[0072] Multi-objective ranking models are ranking methods that seek a balance among multiple potentially conflicting objectives. In the project recommendation list generation stage, they are used to comprehensively rank recommended projects by considering multiple factors such as predicted success rate, project funding amount, and application difficulty. The core challenge of multi-objective ranking lies in how to handle the trade-offs between objectives. The main methods include weighted summation, Pareto optimality, and the analytic hierarchy process (AHP). Weighted summation assigns weights to each objective and calculates a weighted total score for ranking, which is simple but struggles to handle nonlinear trade-offs. Pareto optimality, based on the concept of "non-dominance," seeks solutions where no objective can be improved without harming other objectives, preserving diverse trade-offs. The AHP constructs a hierarchical decision structure and determines weights through pairwise comparisons, making it suitable for handling subjective judgment factors. This paper adopts a Pareto optimality-based multi-objective ranking framework, combined with user preference learning technology, to dynamically adjust the importance of different objectives. In practice, the process first calculates the score for each project across dimensions such as predicted success rate, funding amount, and application difficulty. Then, it identifies Pareto non-dominated solutions (projects where improvement in one dimension is impossible without sacrificing others), forming the optimal candidate recommendation set. For projects in this set, the specific needs and preferences of the enterprise (e.g., prioritizing success rate or funding amount) are further considered. By learning from historical interaction data or directly consulting users, the weights of different objectives are adjusted to generate the final ranking result. This multi-objective ranking method can find the most suitable recommendation scheme for the enterprise's actual situation in a complex decision space, balancing short-term success rate and long-term development potential, providing enterprises with comprehensive and personalized project recommendations.

[0073] Collaborative filtering is a classic recommendation method that makes recommendations based on the assumption that "similar users have similar preferences" or "similar projects are liked by similar users." In the project recommendation list generation stage, collaborative filtering is used to analyze the project application patterns of other companies similar to the target company, providing reference information such as "similar companies have also applied for..." and broadening the recommendation perspective. Collaborative filtering is mainly divided into user-based collaborative filtering and project-based collaborative filtering. User-based collaborative filtering first identifies user groups similar to the target user and then recommends projects liked by these similar users but not yet explored by the target user. Project-based collaborative filtering, on the other hand, finds other projects similar to those already explored by the user and recommends them. In this paper, companies are considered "users," and projects are considered "projects," constructing a company-project interaction matrix to record the history and results of company project applications. Matrix factorization-based collaborative filtering methods, such as Singular Value Decomposition (SVD) or latent semantic models, are used to map companies and projects to a common latent feature space, calculating the similarity between companies and projects. Through collaborative filtering, some projects that might be overlooked based on content matching but are actually very suitable for companies can be discovered. For example, it might find that "small and medium-sized software companies similar to you have successfully applied for the 'Digital Economy Innovation and Development Special Fund,' and we suggest you consider applying." This kind of recommendation based on similar corporate behavior complements content-matching recommendations, increasing the breadth and diversity of recommendations and helping companies discover more potential opportunities.

[0074] In this embodiment, the automatic generation of an enterprise qualification diagnostic scoring report, which includes enterprise strength analysis, weakness diagnosis, and project application suggestions, based on the comprehensive enterprise qualification score and the project matching recommendation list, includes: S5.1: Based on the multi-dimensional scoring results in the comprehensive enterprise qualification scoring results, identify the enterprise's advantageous indicators in each dimension, analyze the reasons for the formation and sustainability of the advantages, and generate the enterprise advantage analysis report paragraph; S5.2: Identify the dimensions and indicators in the comprehensive enterprise qualification score that are lower than the preset threshold, analyze the reasons for the deficiencies, and generate targeted improvement suggestions and improvement paths by combining data from industry benchmark enterprises, thus forming a deficiency diagnosis and improvement suggestion; S5.3: Based on the comprehensive score of the enterprise's qualifications and the project matching recommendation list, formulate a phased and multi-tiered project application strategy, including projects that can be applied for in the near future and projects with medium and long-term development goals, and generate an application plan; S5.4: Integrate the enterprise strengths analysis report paragraphs, the deficiencies diagnosis and improvement suggestions, and the application plan; adaptively select the report template and content structure based on the enterprise type, industry characteristics, and evaluation results; and generate the enterprise qualification diagnosis and scoring report.

[0075] Step S5.1 is the enterprise advantage analysis stage. Based on the multi-dimensional scoring results of the enterprise qualification comprehensive evaluation, this step identifies the enterprise's advantageous indicators in each dimension, analyzes the reasons for and sustainability of these advantages, and generates an enterprise advantage analysis report. In the specific implementation process, the enterprise's advantageous indicators are first identified, i.e., indicators whose scores are higher than the industry average or competitors. The Z-score analysis method is used to calculate the deviation of the enterprise's performance on each indicator from the industry average. Z-score = (Enterprise indicator value - Industry average) / Industry standard deviation. Indicators with a Z-score greater than 1 are considered significant advantages for the enterprise. Simultaneously, the percentile ranking method is used to calculate the enterprise's relative position in the industry, such as "R&D investment ratio is in the top 15% of the industry." After identifying the advantageous indicators, the reasons for their formation are analyzed, and attribution analysis techniques are used to uncover the driving factors behind the advantages. Attribution analysis is a method that determines the main influencing factors by decomposing and tracking changes in indicators. In this process, combined with historical enterprise data and industry development trends, the analysis examines whether the advantages stem from long-term strategic investment, policy support, team strength, technological accumulation, or market opportunities. Next, the sustainability of the advantages is assessed, that is, the stability of these advantages and their future sustainable development capability. Time series analysis methods, such as the Autoregressive Integrated Moving Average (ARIMA) model or trend decomposition analysis, are used to study the historical trends of the company's advantage indicators and assess their stability. Simultaneously, combined with industry development trends and competitive landscape analysis, the sustainability and growth potential of these advantages in the future are predicted. Based on the above analysis, a structured corporate advantage analysis report is generated, including an overview of the advantage indicators, quantification of the degree of advantage, analysis of the reasons for their formation, sustainability assessment, and competitive advantage recommendations, providing a clear advantage positioning for corporate decision-making.

[0076] Step S5.2 is the deficiency diagnosis and improvement suggestion stage. This step identifies dimensions and indicators in the enterprise's comprehensive qualification score that are below a preset threshold, analyzes the reasons for the deficiency, and generates targeted improvement suggestions and improvement paths by combining data from industry benchmark enterprises, thus forming the deficiency diagnosis and improvement suggestions. In the implementation process, the enterprise's deficiency indicators are first identified, i.e., dimensions and indicators whose scores are below the preset threshold. The preset threshold can be set based on the industry average (e.g., below the industry average by one standard deviation) or based on project application requirements (e.g., below the key threshold value for a certain type of project application). For the identified deficiency indicators, a gap analysis method is used to quantify the gap between the enterprise and the industry average or benchmark enterprises, and analyze the severity of the gap and its impact on the overall score. Gap analysis is a method to determine the direction of improvement by comparing the difference between the current state and the target state. The priority of different deficiency indicators is evaluated by calculating the "gap ratio" (the ratio of the enterprise's indicator value to the target value) and the "impact weight" (the contribution of the indicator to the overall score). Next, benchmarking analysis is conducted using data from industry benchmark enterprises, a method to find improvement paths by learning from industry best practices. We select high-performing companies within the industry or those excelling in specific indicators as benchmarks, analyze their successful strategies and practices, and extract lessons learned. Based on this, we generate targeted improvement suggestions, including short-term, actionable optimization measures and medium- to long-term strategic adjustments. Short-term measures primarily target indicators requiring minimal investment and yielding quick results, such as improving internal management systems and optimizing financial structures. Medium- to long-term directions focus on capabilities requiring substantial investment and long-term accumulation, such as enhancing R&D capabilities and brand building. We also provide implementation paths and phased goals for each improvement suggestion to help companies develop feasible improvement plans. Ultimately, we produce a structured deficiency diagnosis and improvement recommendation, including an overview of the deficient indicators, gap analysis, benchmarking results, improvement suggestions, and implementation paths, providing scientific guidance for companies to enhance their capabilities.

[0077] Step S5.3 is the project application planning stage. Based on the company's comprehensive qualification score and the project matching recommendation list, this step develops a phased, multi-tiered project application strategy, including projects that can be applied for in the near term and projects representing medium- to long-term development goals, generating an application plan. In implementation, recommended projects are first categorized according to time and difficulty levels. In terms of time, projects are divided into near-term projects (applicable within 3-6 months), medium-term projects (planning within 6-18 months), and long-term projects (development goals of more than 18 months). In terms of difficulty, projects are divided into basic projects (high success rate, low preparation threshold), advanced projects (requiring improvement in certain areas), and challenging projects (requiring comprehensive capability improvement). Project dependency graph technology is used to construct the logical relationships between projects. This method uses a directed graph to represent the dependencies between projects, helping companies clarify their application path and development trajectory. In the dependency graph, nodes represent specific projects, and edges represent dependencies or optimal application order. For example, successful application for some basic projects may be a prerequisite or favorable factor for applying for higher-level projects. Based on dependency graph analysis, a tiered application strategy is developed. This is a step-by-step approach, progressing from easy to difficult projects. Successfully applying for simpler projects accumulates experience and qualifications, laying the foundation for more challenging projects. Furthermore, a resource-constrained planning approach is adopted, taking into account the company's development plans and resource conditions, to optimize the timeline and resource allocation for project applications while considering the company's human and financial limitations. For each planned project, detailed application timelines, preparation cycles, key material lists, and application guidelines are provided, forming an actionable application guide. Finally, a structured project application plan is generated, including short-term application priorities, medium- and long-term development paths, tiered application strategies, and specific operational guidelines, providing comprehensive planning and guidance for the company's technology project applications.

[0078] Step S5.4 is the diagnostic report generation stage. This step integrates the enterprise strengths analysis report, the deficiency diagnosis and improvement suggestions, and the application plan. Based on the enterprise type, industry characteristics, and assessment results, it adaptively selects the report template and content structure to generate an enterprise qualification diagnostic scoring report. During implementation, a suitable report template is first selected based on the enterprise type, industry, and assessment results. Different types of enterprises (e.g., startups, growth-stage enterprises, mature enterprises), different industries (e.g., high-tech enterprises, manufacturing enterprises, service enterprises), and enterprises with different assessment results (e.g., those with clear advantages, those with balanced development, those with a single breakthrough) are suited to different report structures and focuses. Template adaptation technology is used to dynamically adjust the report's content structure, focus, and expression based on enterprise characteristics and assessment results. Template adaptation is a technology that automatically selects and adjusts the output template based on the characteristics of the target object, matching the most suitable report template for different types of enterprises through a preset rule base or machine learning methods. Next, the enterprise strengths analysis, deficiency diagnosis and improvement suggestions, and project application plan generated in the previous steps are integrated to form a complete report content. Natural language generation technology is used to transform the data analysis results into fluent and professional text descriptions. Natural Language Generation (NLG) is a technology that converts structured data into natural language text, comprising three main steps: content planning (determining what information to convey), sentence planning (organizing the order and structure of information), and implementation (generating specific sentences that conform to grammar and style). To enhance the readability and intuitiveness of the report, it automatically generates accompanying data visualization charts, such as radar charts displaying the company's scores across various dimensions, bar charts comparing the company's performance with the industry average, and heatmaps showing the strengths and weaknesses of each indicator. It also adds personalized executive summaries and action recommendations based on the company's specific circumstances, highlighting the most critical findings and the most urgent action points. Ultimately, it generates a complete, comprehensive, and targeted enterprise qualification diagnostic scoring report, including an overview of the company's basic information, multi-dimensional scoring result analysis, analysis of the company's strengths and weaknesses, improvement suggestions, and project application planning, presented in a combination of charts and text, providing comprehensive support for enterprise decision-making.

[0079] Z-score analysis is a statistical method used to assess the degree of deviation of data points from the overall average. In the analysis of a company's strengths, it is used to identify which indicators a company performs significantly better than the industry average. The formula for calculating the Z-score is Z = (X - μ) / σ, where X is the observed value (a certain indicator value of the company), μ is the overall average (the industry average), and σ is the overall standard deviation (the industry standard deviation). The Z-score measures the standard deviation of a data point from the mean; a positive value indicates above-average performance, a negative value indicates below-average performance, and a larger absolute value indicates a greater degree of deviation. In practical applications, data on various indicators of companies within the industry are collected, the industry average and standard deviation are calculated, and then the Z-scores for each indicator of the target company are calculated. Generally, indicators with a Z-score greater than 1 (i.e., one standard deviation above the industry average) are considered significant strengths of the company, and indicators with a Z-score greater than 2 are considered outstanding strengths. For example, if a company's R&D investment ratio Z-score is 1.8 and its patent grant number Z-score is 2.3, these two can be identified as strength indicators of the company, with the patent grant number being a particularly outstanding strength. The advantage of Z-score analysis lies in its consideration of the data distribution characteristics. It compares not only absolute values ​​but also the degree of variability of indicators across the industry, making the strength of advantages between different indicators comparable. This method helps companies objectively understand their relative position in the industry and identify their true competitive advantages.

[0080] Attribution analysis is a method for identifying key influencing factors by decomposing and tracking changes in indicators. In the analysis of a company's strengths, it is used to uncover the driving factors behind those strengths. The core idea of ​​attribution analysis is to decompose a composite outcome (such as a company's strength indicator) into multiple contributing factors and quantify the contribution of each factor. The main attribution analysis methods include hierarchical attribution, marginal contribution attribution, and model-based attribution. Hierarchical attribution constructs a hierarchical structure of indicators, decomposing the contribution of each level from top to bottom; marginal contribution attribution assesses the independent contribution of a factor by calculating its marginal effect in the presence or absence of other factors; and model-based attribution analyzes the influence coefficients of each factor by establishing statistical models (such as linear regression and decision trees). This paper employs a hybrid attribution analysis strategy: for financial indicators with clear calculation formulas (such as net profit margin), hierarchical attribution is used to decompose them into the contributions of factors such as revenue growth and cost control; for complex indicators such as R&D capabilities and market performance, machine learning-based attribution models are used, such as Shapley Additive Explanations (SHAP) analysis, which is based on the Shapley value concept in game theory to fairly allocate the contribution of each feature to the prediction result. Through attribution analysis, the intrinsic mechanisms of corporate advantages can be revealed. For example, it was found that "the advantage in the number of patents mainly stems from continuous high R&D investment (contribution rate 65%) and effective industry-academia-research cooperation (contribution rate 25%)." This in-depth analysis helps companies better understand the reasons for the formation of their own advantages and provides direction for consolidating and expanding their advantages.

[0081] Time series analysis is a statistical method that studies a sequence of data points arranged chronologically to discover patterns and trends. In the analysis of a company's competitive advantages, it is used to assess the stability of its competitive advantages and its future sustainable development capabilities. The main methods of time series analysis include decomposition, smoothing, the Autoregressive Integrated Moving Average (ARIMA) model, and state-space models. Decomposition methods break down a time series into trend, seasonal, and random components, analyzing the characteristics of each component separately. Smoothing methods (such as moving averages and exponential smoothing) highlight long-term trends by eliminating short-term fluctuations. ARIMA combines autoregression, differencing, and moving averages, making it suitable for modeling non-stationary time series. State-space models can handle complex time series with multivariate and multi-time-scale characteristics. This paper employs a multi-level time series analysis approach: First, Seasonal-Trend decomposition using LOESS (Locally Weighted Regression-based Seasonal Trend Decomposition) is used to decompose the historical data of each company's competitive advantage into trend, seasonal, and residual components, identifying long-term trends and cyclical fluctuations. Then, stability tests, such as the Augmented Dickey-Fuller test (ADF test), are applied to the trend components to determine the stationarity of the time series. For stable competitive advantages, the ARIMA model is used to predict future trends. For indicators with significant growth or volatility, more complex predictive models are constructed by incorporating external factors (such as industry development and policy changes). Through time series analysis, the stability of a company's competitive advantage can be assessed, such as "R&D investment ratio has maintained stable growth over the past 5 years with low volatility, and is expected to maintain a leading position for the next 2 years." This dynamic advantage assessment is more forward-looking than static cross-sectional analysis, providing more valuable references for the company's strategic planning.

[0082] Gap analysis is a method for determining improvement directions by comparing the current state with a target state. In the deficiency diagnosis and improvement suggestion stage, it is used to quantify the gap between a company and the industry average or benchmark companies. The basic steps of gap analysis include: determining the comparison benchmark (such as the industry average, benchmark companies, or project application thresholds); measuring the gap between current performance and the benchmark; analyzing the causes and impacts of the gap; and developing strategies to narrow the gap. In this paper, gap analysis adopts a multi-dimensional evaluation framework, calculating not only the absolute gap (the difference between the company's indicator value and the benchmark value), but also the relative gap (the percentage of the difference relative to the benchmark value) and the standardized gap (the difference divided by the industry standard deviation). A "gap priority index" is also introduced, comprehensively considering the size of the gap, the importance of the indicator, and the difficulty of improvement. The calculation formula is: Priority Index = Gap Size × Indicator Weight / Difficulty of Improvement. This index helps companies identify the most pressing deficiencies. For example, it might be found that "the company's R&D personnel ratio is 15 percentage points lower than the industry average (relative gap 30%), while the technology transfer rate is 5 percentage points lower (relative gap 20%). Considering the indicator weight and the difficulty of improvement, it is recommended to prioritize improving the technology transfer rate." Based on the gap analysis results, specific improvement suggestions are generated, including goal setting (such as "planning to increase the proportion of R&D personnel to the industry average within one year"), specific measures (such as "optimizing the R&D personnel recruitment process and strengthening cooperation with universities"), and phased checkpoints, forming a formative improvement path.

[0083] Benchmarking is a method for finding improvement paths by learning from industry best practices. In the deficiency diagnosis and improvement suggestion stage, it is used to analyze the gap between a company and industry benchmarks and extract lessons that can be learned. The main types of benchmarking analysis include internal benchmarking (comparing with best practices within the organization), competitive benchmarking (comparing with direct competitors), functional benchmarking (comparing with leading companies in a specific functional area), and general benchmarking (comparing with outstanding companies in different industries). This paper adopts a multi-level benchmarking framework: first, industry-level benchmarking is conducted, selecting the company with the best overall performance in the same industry as the overall benchmark; then, indicator-level benchmarking is conducted, selecting the company with the best performance in each indicator for the company's shortcomings as the specific benchmark; finally, strategy-level benchmarking is conducted, analyzing the successful strategies and practices of benchmark companies in relevant fields. Through data mining and pattern recognition techniques, potential success factors are extracted from the publicly available information of benchmark companies, such as R&D management models, talent development mechanisms, and intellectual property strategies. For example, it may be found that "the patent quality advantage of industry-leading company A mainly comes from its unique 'R&D crowdsourcing + internal review' innovation model, and it is recommended that the company learn from and establish a similar open innovation platform." Benchmarking analysis not only helps companies clarify their improvement directions, but also provides specific and feasible implementation references, transforming the abstract "improvement indicators" into concrete "learning best practices", making improvement suggestions more actionable and effective.

[0084] Project dependency graphs are a method of representing the dependencies between projects using directed graphs. In the project application planning stage, they are used to construct the logical relationships between projects, helping companies clarify their application paths and development trajectories. In a project dependency graph, nodes represent specific technology projects, and edges represent dependencies or optimal application order between projects. Dependencies are mainly categorized into three types: strong dependencies (preceding projects are necessary conditions for subsequent projects, such as the success of certain basic projects being a hard requirement for applying for advanced projects), weak dependencies (preceding projects contribute to the success of subsequent projects but are not necessary conditions), and mutual exclusion (two projects cannot be applied for simultaneously or their success probabilities affect each other). Methods for constructing project dependency graphs include expert knowledge encoding, historical data analysis, and rule-based reasoning. This paper employs a graph learning method based on historical data, automatically discovering dependencies between projects by analyzing a large number of companies' project application sequences and success patterns. Graph mining algorithms, such as frequent subgraph mining and path analysis, are used to identify common patterns in successful application paths. Based on the project dependency graph, graph algorithms (such as topological sorting and critical path analysis) are applied to generate the optimal application sequence, considering project dependency constraints, time windows, and resource limitations to form a reasonable phased application plan. The visualization of project dependency diagrams can also intuitively help companies understand the relationships between different projects and the best application path, providing a clear reference for strategic decision-making.

[0085] The tiered application strategy is a progressive approach to project applications, starting with easier projects and gradually increasing in difficulty. In the project application planning stage, it helps companies accumulate experience and qualifications by successfully applying for simpler projects, laying the foundation for applying for more challenging ones. The core concept of the tiered application strategy is to view project applications as a gradual improvement process, rather than an isolated, one-off event. Through a carefully designed application sequence, it maximizes the overall success rate and amount of funding obtained by the company. The main steps of the tiered application strategy include: project grading (categorizing projects into basic, intermediate, and challenging levels based on difficulty, funding scale, and impact); capability matching (assessing the match between the company's current capabilities and the project levels); path planning (designing a project application path that gradually improves from the current capability level to the target level); and feedback adjustment (continuously adjusting subsequent application plans based on application results). This tiered application strategy employs a "small steps, quick progress" approach, suggesting that companies first apply for small projects with high success rates and short preparation periods to rapidly accumulate application experience and project management capabilities. Next, they should apply for medium-difficulty projects with substantial funding to enhance their research strength and project execution capabilities. Finally, they should attempt to apply for high-difficulty, high-funding major projects to achieve a leapfrog development of their technological capabilities. A "tiered leap condition" is also designed, meaning that when a company performs exceptionally well in a certain level of projects (e.g., successfully applying for multiple projects with significant execution results), it can consider directly challenging higher-level projects to accelerate its development. This tiered application strategy considers both the gradual improvement of a company's capabilities and the possibility of rapid development, providing companies with a scientifically sound and reasonable project application path.

[0086] Resource-constrained programming (RCP) is a planning method for optimizing activity scheduling under limited resource conditions. In the project application planning stage, it is used to optimize the timing and resource allocation of project applications while considering the constraints of a company's human and financial resources. The core problem of RCP is how to find the optimal or near-optimal activity scheduling scheme under various constraints (such as resource limitations, time windows, and priority dependencies). Commonly used solution methods include exact algorithms (such as branch and bound, dynamic programming), heuristic algorithms (such as genetic algorithms, simulated annealing), and constraint programming. In this paper, project application planning is modeled as a resource-constrained project scheduling problem, where project application preparation and execution are the activities to be scheduled, the company's human and financial resources are limited resources, the dependencies between projects are priority constraints, and the application time window is a time constraint. This problem is solved using a combination of priority rule-based heuristic algorithms and local search. Priority rules include Shortest Processing Time First (SPT), Earliest Deadline First (EDD), and Highest Resource Requirement First (HRD), which are used to generate an initial feasible solution; then, local search methods (such as moving bottleneck activities and adjusting resource allocation) are used to improve the initial solution. It also takes into account the uncertainty of project applications, introducing robust optimization techniques to reserve appropriate buffers in resource allocation to cope with possible plan changes. Through resource-constrained planning, it can generate an application timeline that maximizes project value while conforming to the company's resource realities, avoiding the problem of the company spreading resources too thinly due to applying for multiple projects simultaneously, which could lead to a decline in application quality.

[0087] Template adaptation technology is a technique that automatically selects and adjusts output templates based on the characteristics of the target object. In the diagnostic report generation process, it is used to dynamically adjust the content structure, focus, and expression of the report based on the company's type, industry, and assessment results. The basic idea of ​​template adaptation is to divide the report generation process into two stages: template selection and content filling. In the template selection stage, the most suitable basic template is selected from a pre-set template library based on the company's characteristics (such as size, industry, and development stage) and assessment results (such as strengths and weaknesses). In the content filling stage, the proportion, level of detail, and expression of each part of the template are adjusted based on specific data and analysis results to generate a personalized final report. In this paper, template adaptation adopts a combination of rule-based and machine learning methods: on the one hand, a series of report templates and adjustment rules are pre-set for different types of companies, such as "for R&D-oriented companies, the analysis of technological innovation capabilities should be more detailed"; on the other hand, the template selection and adjustment strategy is continuously optimized by learning from historically generated reports and user feedback. In practice, decision trees or case-based reasoning methods are used for template selection, finding the templates used in the most similar historical cases based on the company's multidimensional characteristics. Then, rule-based content generation adjusts the content weight and level of detail in each part of the template; for example, more detailed descriptions are used for the company's strengths, while general content is given brief explanations. Furthermore, the use of technical terminology is adjusted according to the company's industry characteristics to ensure the report language is both professional and easy to understand. This adaptive report generation method ensures that each diagnostic report accurately matches the company's actual situation and needs, improving the report's relevance and practicality.

[0088] Natural Language Generation (NLG) is a technology that converts structured data into natural language text. In the diagnostic report generation stage, it is used to transform data analysis results into fluent and professional text descriptions. The basic NLG process includes three main stages: content planning, sentence planning, and implementation. Content planning determines the information content and structure to be expressed, such as determining which analysis results and recommendations to include in the report; sentence planning organizes the order of information expression and sentence structure, such as how to combine multiple related data points into coherent paragraphs; the implementation stage is responsible for generating specific sentences that meet grammatical and stylistic requirements. In this paper, NLG employs a hybrid approach of template-based and neural network-based methods. For highly structured content with relatively fixed expressions (such as data descriptions and standard evaluation conclusions), a template-based method is used, pre-setting the sentence structure and only replacing specific numerical values ​​and evaluation words; for content requiring more natural expression (such as in-depth analysis and personalized recommendations), a neural network-based generation method is used, such as a fine-tuning model based on the GPT series of generative pre-trained transformers, which can generate more flexible and diverse expressions. To ensure the quality of the generated text, a multi-level quality control mechanism is also introduced, including grammar checking, factual consistency verification, and style consistency evaluation.

[0089] like Figure 3 As shown, the present invention also provides a multi-dimensional enterprise qualification assessment system, including: The data acquisition module 601 is used to acquire multi-dimensional raw data including corporate financial operations, R&D capabilities, and industry position, and apply data cleaning and normalization preprocessing techniques to obtain a standardized multi-dimensional feature dataset of the enterprise. The indicator system construction module 602 is used to optimize the indicators by grouping and calculating the weights using the improved WP-PVC algorithm for the standardized enterprise multidimensional feature dataset, and to construct a multidimensional evaluation indicator system that includes financial and operational status, R&D capability level, industry position and brand effect. The scoring calculation module 603 is used to perform correlation analysis and scoring calculation between indicators based on the multi-dimensional evaluation index system and the dual-standard WP-PVC algorithm, and generate a comprehensive enterprise qualification score result that includes the comprehensive enterprise qualification score and the sub-dimensional score results. The project matching module 604 is used to calculate the matching degree between the enterprise and various science and technology projects based on the comprehensive score of the enterprise's qualifications and combined with a random online sorting multi-logarithm algorithm, and output a project matching recommendation list and application success rate prediction. The report generation module 605 is used to automatically generate an enterprise qualification diagnostic scoring report containing enterprise strength analysis, weakness diagnosis, and project application suggestions based on the comprehensive enterprise qualification scoring results and the project matching recommendation list.

[0090] The technical solution of this invention constructs an enterprise evaluation index system through an improved WP-PVC algorithm, uses a dual-standard WP-PVC algorithm to calculate the comprehensive score of enterprise qualifications, and combines a multi-logarithmic algorithm with random online sorting for project matching and success rate prediction. This provides enterprises with comprehensive and accurate qualification evaluation services and project application guidance, effectively solving the technical problems of static evaluation index systems, simple index relationship processing, and low matching degree between evaluation results and project applications in existing technologies.

[0091] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-dimensional enterprise qualification assessment method, characterized in that, Includes the following steps: We acquire multi-dimensional raw data containing corporate financial operations, R&D capabilities, and industry position, and apply data cleaning and normalization preprocessing techniques to obtain a standardized multi-dimensional feature dataset of enterprises. For the standardized enterprise multidimensional feature dataset, the improved WP-PVC algorithm is used to group and optimize the indicators and calculate the weights, and a multidimensional evaluation indicator system including financial operation, R&D capability level, industry status and brand effect is constructed. Based on the multi-dimensional evaluation index system, the dual-standard WP-PVC algorithm is used to perform correlation analysis and scoring calculation between indicators, generating a comprehensive enterprise qualification score that includes the comprehensive enterprise qualification score and the sub-dimensional scoring results. Based on the comprehensive score of the enterprise's qualifications, and combined with a multi-logarithm algorithm for random online sorting, the matching degree between the enterprise and various technology projects is calculated, and a project matching recommendation list and application success rate prediction are output. Based on the comprehensive score of the enterprise's qualifications and the project matching recommendation list, an enterprise qualification diagnostic score report is automatically generated, which includes an analysis of the enterprise's strengths, a diagnosis of its weaknesses, and suggestions for project applications.

2. The method according to claim 1, characterized in that, The process involves acquiring multi-dimensional raw data encompassing corporate financial operations, R&D capabilities, and industry standing. Preprocessing techniques such as data cleaning and normalization are then applied to obtain a standardized multi-dimensional feature dataset of the enterprise, including: By accessing public data sources including the enterprise credit information disclosure system, tax system, and business registration information system, and combining network data collection technology, we collect basic information and financial data such as enterprise business registration information, historical financial statement data, tax records, and financing history to obtain a basic enterprise information dataset. Collect and obtain data related to enterprise R&D capabilities, including the number and type of enterprise patents, citation rate, R&D investment ratio, proportion of scientific research personnel, and technology transfer status, to obtain enterprise R&D capability dataset; Collect market performance data including enterprise market share, brand awareness, user reviews, industry ranking, and product competitiveness to obtain enterprise industry position dataset; Based on the enterprise basic information dataset, the enterprise R&D capability dataset, and the enterprise industry position dataset, the data is cleaned using techniques including median padding and locality-sensitive hashing. Missing values, outliers, and duplicate values ​​are detected and processed, and the cleaned multidimensional enterprise data is output. Based on the cleaned enterprise multidimensional data, data standardization is performed using methods including maximum and minimum value normalization and Z-score standardization, so that the data in each dimension can be compared, resulting in the standardized enterprise multidimensional feature dataset.

3. The method according to claim 1, characterized in that, For the standardized enterprise multidimensional feature dataset, the improved WP-PVC algorithm is used to group and optimize the indicators and calculate their weights, constructing a multidimensional evaluation indicator system that includes financial performance, R&D capability level, industry position, and brand effect, including: Based on the standardized enterprise multidimensional feature dataset, and combined with industry expert experience and policy guidance, a candidate set of indicators reflecting the enterprise's financial and operational status, R&D capability level, industry position, and brand effect is selected to establish an initial indicator pool. The enterprise evaluation indicators in the initial pool of indicators are regarded as vertices in the graph, and the relationships between indicators are regarded as edges. The strength of the relationship between indicators is represented by the edge weight. The improved WP-PVC algorithm is applied to group the indicators, dividing them into different groups including financial operations, R&D capabilities, and industry status, thus obtaining indicator groups of different categories. For the different groups of indicators, the improved WP-PVC algorithm is used to set the minimum threshold that each indicator group needs to cover, solve the subset of indicators with the minimum weight, and determine the weight configuration of each indicator. Based on the weight configuration of each indicator, the discrimination and predictive ability of the indicator system are verified through historical sample data. The information gain and variance contribution rate of each indicator are calculated, redundant indicators are eliminated, and the multi-dimensional evaluation indicator system is formed.

4. The method according to claim 3, characterized in that, The method of using the improved WP-PVC algorithm to group and optimize indicators and calculate weights also includes: Based on the characteristics of different industries and the development stage of enterprises, the multi-dimensional evaluation index system is adaptively adjusted to generate customized evaluation index systems for different types of enterprises. The adaptive adjustment includes adjusting the weights of indicators in each dimension according to the characteristics of the industry to which the enterprise belongs, adjusting the evaluation focus according to the development stage of the enterprise, and dynamically adjusting the indicator weights according to the current policy environment and support direction.

5. The method according to claim 1, characterized in that, Based on the multi-dimensional evaluation index system, the dual-standard WP-PVC algorithm is used to perform correlation analysis and scoring calculation among the indicators, generating a comprehensive enterprise qualification score result that includes both a comprehensive enterprise qualification score and multi-dimensional scoring results, including: The data of various indicators of enterprises in the multi-dimensional evaluation indicator system are standardized to eliminate the influence of dimensions, so that different indicators are comparable and standardized indicator data are obtained. Based on the standardized indicator data, the bi-standard WP-PVC algorithm is applied to analyze the correlation between indicators within the same dimension, construct an indicator correlation network within the dimension, calculate the correlation coefficient between indicators, identify key indicators and auxiliary indicators, and determine the importance ranking of indicators within the dimension by combining the optimized weights of each indicator in the multi-dimensional evaluation indicator system in the evaluation system. Based on the importance ranking of the indicators and the optimization weights, the standardized value of the enterprise on each indicator is multiplied by the corresponding optimization weight, and a non-linear adjustment is made. The results are then aggregated according to the weighted average rule to calculate the enterprise's scores in each dimension of financial operations, R&D capabilities, and industry position, forming a multi-dimensional scoring result. The dual-standard WP-PVC algorithm is applied to dynamically adjust the weights of each dimension based on their contribution to the overall enterprise qualification, taking into account industry characteristics, enterprise type, and application project characteristics. The scores of each dimension are then weighted according to the optimized weights to generate the overall enterprise qualification score, which includes the overall enterprise qualification score and the scores of each dimension.

6. The method according to claim 5, characterized in that, The dynamic adjustment of the weights of each dimension also includes: Based on the contribution of each dimension to enterprise success in different industries, we analyze the impact of industry characteristics on dimension weights and obtain industry characteristic weight adjustment coefficients. Based on historical successful cases of different types of science and technology project applications, we analyze the performance characteristics of each dimension to obtain the project orientation weight adjustment coefficient. Based on the industry characteristic weight adjustment coefficient and the project orientation weight adjustment coefficient, by setting a minimum coverage threshold for each dimension, the optimal weight allocation scheme is found so that the evaluation results are closest to historical successful cases, and the optimized dimension weight configuration is generated.

7. The method according to claim 1, characterized in that, The process involves calculating the matching degree between enterprises and various technology projects based on the comprehensive score of enterprise qualifications, combined with a multi-logarithm algorithm for random online sorting, and outputting a project matching recommendation list and application success rate prediction, including: Collect information on application requirements, review standards, and funding amounts for various science and technology projects and certifications; establish a structured science and technology project database; and classify and label projects by industry, technology direction, and support stage in multiple dimensions to build a science and technology project database that is updated in real time. Based on the comprehensive score of enterprise qualifications and the real-time updated technology project database, enterprise characteristics and project requirements are transformed into multi-dimensional vectors, the matching degree is calculated, and a matching model between enterprises and projects is constructed. Based on the matching model and successful cases of matching enterprise characteristics with projects in historical application data, a multi-logarithmic algorithm for random online sorting is applied to optimize the project recommendation ranking and generate an optimized project matching sequence. Based on the optimized project matching sequence and historical application case analysis, an application success rate prediction model is constructed to calculate the success probability of enterprises applying for various types of projects, provide enterprises with application decision references, and generate application success rate prediction results. Based on the optimized project matching sequence and the application success rate prediction results, a project recommendation list suitable for enterprises to apply for is generated, sorted by success rate from highest to lowest, and accompanied by matching reasons and application suggestions. The project matching recommendation list and application success rate prediction are then output.

8. The method according to claim 1, characterized in that, Based on the comprehensive enterprise qualification score and the project matching recommendation list, an enterprise qualification diagnostic score report is automatically generated, which includes enterprise strength analysis, weakness diagnosis, and project application suggestions. Based on the multi-dimensional scoring results in the comprehensive enterprise qualification scoring results, identify the enterprise's advantageous indicators in each dimension, analyze the reasons for the formation and sustainability of the advantages, and generate a section of the enterprise advantage analysis report. Identify the dimensions and indicators in the comprehensive enterprise qualification score that are below a preset threshold, analyze the reasons for the deficiencies, and generate targeted improvement suggestions and improvement paths by combining data from industry benchmark enterprises, thus forming a deficiency diagnosis and improvement suggestion. Based on the comprehensive score of the enterprise's qualifications and the project matching recommendation list, a phased and multi-tiered project application strategy is formulated, including projects that can be applied for in the near future and projects with medium and long-term development goals, and an application plan is generated. By integrating the enterprise strengths analysis report, the deficiencies diagnosis and improvement suggestions, and the application plan, and adaptively selecting the report template and content structure based on the enterprise type, industry characteristics, and evaluation results, the enterprise qualification diagnosis and scoring report is generated.

9. A multi-dimensional enterprise qualification assessment system, characterized in that, include: The data acquisition module is used to acquire multi-dimensional raw data including corporate financial operations, R&D capabilities, and industry position. It applies data cleaning and normalization preprocessing techniques to obtain a standardized multi-dimensional feature dataset of the enterprise. The indicator system construction module is used to optimize the indicators by grouping and calculating the weights using the improved WP-PVC algorithm on the standardized enterprise multidimensional feature dataset, and to construct a multidimensional evaluation indicator system that includes financial and operational status, R&D capability level, industry position and brand effect. The scoring calculation module is used to perform correlation analysis and scoring calculation between indicators based on the multi-dimensional evaluation index system and the dual-standard WP-PVC algorithm, and generate a comprehensive enterprise qualification score result that includes the comprehensive enterprise qualification score and the sub-dimensional scoring results. The project matching module is used to calculate the matching degree between enterprises and various technology projects based on the comprehensive score of enterprise qualifications and a multi-logarithm algorithm for random online sorting, and output a project matching recommendation list and application success rate prediction. The report generation module is used to automatically generate an enterprise qualification diagnostic scoring report that includes analysis of enterprise strengths, diagnosis of weaknesses, and suggestions for project application, based on the comprehensive enterprise qualification score and the project matching recommendation list.

Images