Data management method for digital decision support platform

Abstract

The invention relates to the field of intelligent management, and provides a data management method for a digital decision support platform, which makes and adjusts a strategic plan through a strategic management module, ensures that resource allocation is matched with a strategic target, monitors and analyzes financial data through a financial management module to dynamically manage and control financial budget and fund security, and provides a data management method for a digital decision support platform. The marketing activity effect is planned and evaluated through the market management module, the project progress and quality are tracked through the project management module, risk early warning is carried out based on progress reports and quality detection records of a plurality of sub-projects, recruitment, training, performance assessment and salary management are processed through the human resource management module, and the marketing activity effect is evaluated through the human resource management module. And the culture management module analyzes enterprise culture and carries out culture propaganda and training activities. Therefore, through the digital and intelligent management of each module, the decision-making efficiency and management level of an enterprise can be effectively improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent management, and more specifically, to a data management method for a digital decision support platform. Background Art

[0002] In today's era of rapid digital development, enterprises are facing an increasingly complex and changing market environment. Market competition is becoming more intense, industry dynamics are changing rapidly, and customer needs are also constantly evolving. In order to survive and achieve sustainable development in such an environment, enterprises urgently need an efficient decision support system.

[0003] However, traditional digital decision support platforms have many drawbacks. Specifically, strategic management is difficult to dynamically adjust goals according to resources, financial monitoring and analysis are relatively crude, project risk early warning relies on experience and is not timely enough, human resource management efficiency is low, cultural management lacks in-depth and systematic support, etc. These problems not only limit the agility of enterprises, but also seriously affect the accuracy and timeliness of decision-making.

[0004] Therefore, an optimized data management solution for a digital decision support platform is desired. Summary of the Invention

[0005] The present application aims at the disadvantages in the prior art and provides a data management method for a digital decision support platform.

[0006] According to one aspect of the present application, there is provided a data management method for a digital decision support platform, the digital decision support platform including a strategic management module, a financial management module, a market management module, a project management module, a human resource management module, and a cultural management module. Among them, the data management method includes:

[0007] Using the strategic management module to formulate a strategic plan and monitor the strategic process, and analyzing the matching between strategic goals and resource allocation to dynamically adjust the strategic goals;

[0008] Using the financial management module to monitor and analyze financial data to dynamically control financial budgets and fund security;

[0009] Using the market management module to execute the planning and effect evaluation of marketing activities;

[0010] Using the project management module to track the progress and quality of a project and conduct risk early warning. Among them, using the project management module to conduct risk early warning includes conducting risk early warning based on the project-quality semantic core association information between the {project progress report and quality inspection record} of multiple sub-projects;

[0011] The human resource management module is used for managing recruitment, training, performance appraisal, and salary management;

[0012] The culture management module is used to analyze corporate culture and related activities, and conduct cultural publicity and training.

[0013] Since the present application adopts the above technical solutions, it has remarkable technical effects:

[0014] The data management method for a digital decision-making support platform provided by the present application formulates and adjusts strategic plans through a strategic management module to ensure that resource allocation matches strategic goals, monitors and analyzes financial data through a financial management module to dynamically control financial budgets and fund security, plans and evaluates the effects of marketing activities through a market management module, tracks project progress and quality through a project management module, and conducts risk warnings based on progress reports and quality inspection records of multiple sub-projects, processes recruitment, training, performance appraisal, and salary management through a human resource management module, and analyzes corporate culture and conducts cultural publicity and training activities through a culture management module. In this way, through the digital and intelligent management of each module, the decision-making efficiency and management level of the enterprise can be effectively improved. Brief Description of the Drawings

[0015] By describing the embodiments of the present application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of the present application will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present application, and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation to the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 It is a flowchart of the data management method for a digital decision-making support platform according to an embodiment of the present application.

[0017] Figure 2 It is a flowchart of step S4 in the data management method for a digital decision-making support platform according to an embodiment of the present application.

[0018] Figure 3 It is a schematic diagram of data flow of step S4 in the data management method for a digital decision-making support platform according to an embodiment of the present application.

[0019] Figure 4 It is a flowchart of step S43 in the data management method for a digital decision-making support platform according to an embodiment of the present application. Detailed Description of the Embodiments

[0020] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0021] In today's era of rapid digital development, enterprises are facing an increasingly complex and changing market environment. Market competition is becoming increasingly fierce, industry changes are changing with each passing day, and customer needs are constantly upgrading. In order to gain a foothold and achieve sustainable development in such an environment, enterprises urgently need to use efficient decision support systems to enhance their core competitiveness.

[0022] However, traditional digital decision support platforms have many shortcomings. For example, strategic management is difficult to flexibly adjust goals according to the dynamic changes of resources, financial monitoring and analysis methods are too extensive, project risk warnings rely on experience and often lag behind, human resource management efficiency is low, and corporate culture management lacks systematic and in-depth support. These problems not only weaken the company's ability to respond flexibly, but also significantly restrict the accuracy and timeliness of decision-making.

[0023] It should be understood that the digital decision support platform includes a strategic management module, a financial management module, a market management module, a project management module, a human resource management module, and a cultural management module. Based on this, this application proposes a data management method for a digital decision support platform, which provides enterprises with comprehensive management and decision support tools by integrating and analyzing data from various business modules, helping enterprises cope with complex market environments and improving overall operational efficiency and competitiveness. Figure 1 Flowchart of a data management method for a digital decision support platform according to an embodiment of the present application. Figure 1 As shown, the data management method for a digital decision support platform includes: S1, using the strategic management module to formulate strategic plans and monitor strategic progress, and analyzing the matching between strategic goals and resource allocation to dynamically adjust the strategic goals; S2, using the financial management module to monitor and analyze financial data to dynamically control financial budgets and capital security; S3, using the market management module to plan and evaluate the effectiveness of marketing activities; S4, using the project management module to track the progress and quality of the project and conduct risk warnings, wherein the use of the project management module for risk warnings includes conducting risk warnings based on the project-quality semantic core association information between {project progress reports and quality inspection records} of multiple sub-projects; S5, using the human resource management module to manage recruitment, training, performance appraisal and salary management; S6, using the cultural management module to analyze corporate culture and related activities, and conduct cultural publicity and training.

[0024] In step S1, the strategic management module is used to formulate a strategic plan and monitor the strategic process, and analyze the matching degree between the strategic objectives and resource allocation to dynamically adjust the strategic objectives. It should be understood that the strategic management module is the core component of the digital decision-making support platform. Its main function is to assist enterprises in formulating scientific and reasonable strategic plans, effectively monitoring the strategic implementation process, and dynamically adjusting the strategic objectives according to the matching situation between the strategic objectives and resource allocation, so as to ensure that the enterprise strategy is always adapted to the internal and external environment and help the enterprise achieve sustainable development. Specifically, first, the enterprise needs to clarify its long-term and short-term strategic objectives, which are usually determined comprehensively based on various factors such as market analysis, competitive environment assessment, and internal resource status. At the same time, the enterprise also needs to establish a resource allocation model to identify the types and quantities of resources required to achieve each strategic objective. Once the strategic objectives are set, through integrating project management tools and other relevant systems, the strategic management module can track the strategic execution progress in real time, including but not limited to key performance indicators (KPIs), milestone achievement, etc. Then, using data analysis techniques, especially machine learning algorithms, the module continuously evaluates the matching degree between resource allocation and strategic objectives. When a mismatch is found, the module will automatically propose adjustment suggestions, enabling the enterprise management to respond quickly, reallocate resources or adjust the strategic direction to ensure that the enterprise always moves forward along the optimal path. In addition, in order to continuously optimize the strategic execution process, the strategic management module also includes a feedback mechanism that allows opinions and suggestions on the strategic implementation effect to be collected from all levels. Based on this feedback information, the organization can iteratively update its strategic plan, gradually improve its resource allocation plan, and improve the strategic execution efficiency and effect.

[0025] In step S2, the financial management module is used to monitor and analyze financial data to dynamically control the financial budget and fund security. Specifically, the financial management module automatically collects and integrates financial data from multiple sources such as bank accounts, sales systems, procurement systems, etc., ensuring that managers can obtain the latest financial information and make timely and effective decisions based on this information. At the same time, it supports flexible budgeting tools, allowing enterprises to formulate detailed budget plans according to factors such as historical data and market forecasts, and then quickly detect any deviations by comparing actual expenditures with the budget in real time, thus ensuring the strict implementation of the budget. In addition, the financial management module also provides a cash flow forecasting function to help enterprises better plan future fund requirements and prevent possible liquidity risks. In-depth analysis of the cost structure is also one of its major features. By dissecting costs in detail, it identifies cost-saving opportunities and puts forward optimization suggestions to help enterprises improve economic efficiency. Moreover, it can automatically generate various financial statements including income statements, balance sheets, and cash flow statements, and supports customized report design to meet the needs of different reports. Generally speaking, the financial management module can comprehensively and real-time monitor and deeply analyze the financial data of enterprises, helping enterprises to promptly discover financial risks, optimize resource allocation, and improve the efficiency of financial management and the scientific nature of decision-making.

[0026] In step S3, the market management module is used to execute the planning and effect evaluation of marketing activities. It should be understood that the market management module helps enterprises accurately grasp market dynamics, formulate scientific and effective marketing strategies, and quantitatively evaluate the effects of marketing activities in a data-driven manner to optimize resource allocation, enhance market competitiveness, and improve the return on marketing investment. Specifically, the market management module integrates and analyzes data from multiple channels such as social media, website access records, and sales data to deeply understand market demand, consumer preferences, and competitor dynamics. Based on the analysis results of this data, the module can identify different market segments and formulate targeted marketing strategies for each segment, thereby more accurately positioning the target market and improving the effectiveness of marketing activities. The market management module also provides a complete set of tools for creating, managing, and tracking marketing activities, including advertising placement, promotional activities, and content marketing, allowing users to set the goals, budgets, schedules, and expected outcomes of the activities. In addition, by integrating multiple online and offline marketing channels, it ensures the consistency of information and the maximization of coverage, supports cross-channel coordination, and realizes unified marketing message transmission. The intuitive visual dashboard provided by the module can also display key indicators and analysis results, helping managers quickly understand the market situation and the performance of marketing activities. By real-time monitoring of key performance indicators (such as click-through rate, conversion rate, ROI, etc.), the performance of marketing activities can be evaluated in a timely manner and marketing strategies can be adjusted according to the feedback to optimize future marketing plans, thereby maximizing the return on investment.

[0027] In step S4, the project management module is used to track the progress and quality of the project and conduct risk early warning. Among them, using the project management module to conduct risk early warning includes conducting risk early warning based on the project-quality semantic core association information between the {project progress reports and quality inspection records} of multiple sub-projects. It should be understood that the project progress is directly related to whether the project can be completed on time. If the progress gets out of control, it may lead to delivery delays, missing the best market opportunity or facing default risks; the project quality determines whether the project results meet the expected standards. Low-quality results may not meet customer needs and damage the enterprise's reputation; risk early warning is to discover potential problems in advance to avoid the deterioration of problems affecting the project progress and quality. Therefore, tracking these three key elements is a necessary condition for ensuring the successful delivery of the project. However, traditional project risk early warning relies on personal experience and manual analysis, lacking automated and intelligent support, resulting in inaccurate early warning. Moreover, this method can often only identify known risks and is difficult to discover potential complex problems.

[0028] Based on this, the project management module is further used to conduct the risk early warning based on the project-quality semantic core association information between the {project progress reports and quality inspection records} of multiple sub-projects. Specifically, the technical concept of this application is to obtain the {project progress reports and quality inspection records} of multiple sub-projects, use natural language processing technology based on deep learning to conduct semantic understanding of each {project progress report and quality inspection record}, and then conduct core semantic feature interaction on the {semantic features of the project progress report text and the semantic features of the quality inspection record} of each sub-project, so as to intelligently judge whether there are risks in the project according to the multi-dimensional analysis representation between the semantic interaction representations of the project progress report-quality inspection records of multiple sub-projects after interaction. Compared with manual analysis, by conducting core semantic feature interaction on the project progress reports and quality inspection records of each sub-project, this application can discover the deep-level associations between different documents, thereby enabling more accurate risk prediction and early warning.

[0029] Figure 2 It is a flowchart of step S4 in the data management method for a digital decision support platform according to an embodiment of the present application. Figure 3 It is a schematic diagram of data flow in step S4 in the data management method for a digital decision support platform according to an embodiment of the present application. As Figure 2 and Figure 3As shown, step S4 includes: S41, obtaining the {project progress reports and quality inspection records} of the multiple sub-projects; S42, respectively performing text semantic understanding on each of the {project progress reports and quality inspection records} in the {project progress reports and quality inspection records} of the multiple sub-projects to obtain the {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects; S43, performing project-quality core semantic feature interaction representation on each group of corresponding project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors in the {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects to obtain multiple sub-project progress report-quality inspection record semantic interaction representation vectors; S44, performing multi-dimensional semantic comprehensive analysis on the multiple sub-project progress report-quality inspection record semantic interaction representation vectors to obtain a project comprehensive progress report-quality inspection record semantic interaction feature vector; S45, based on the project comprehensive progress report-quality inspection record semantic interaction feature vector, obtaining a risk prediction result, where the risk prediction result is used to indicate whether there is a risk in the project.

[0030] In step S41, the {project progress reports and quality inspection records} of the multiple sub-projects are obtained. It should be understood that the project progress report of each sub-project will include information on the completion status of each task in the sub-project, information on the completion of important milestones in the sub-project, information on the comparison between the actual completion progress and the planned progress, information on resource usage, etc.; while the quality inspection record of each sub-project includes information on the inspection method, information on the qualification status of the inspection object, specific indicator data information, and defect description information, etc. Specifically, the data of the {project progress reports and quality inspection records} of a single sub-project can enable the model to understand the progress and quality of a single sub-project under the project, and through comprehensive analysis of the {project progress reports and quality inspection records} of multiple sub-projects, the risk status of the project can be evaluated as a whole, which helps to more accurately identify project risks and issue early warnings, so that the project management party can take effective preventive measures to ensure the smooth progress of the project.

[0031] In step S42, text semantic understanding is performed on each of the {project progress reports and quality inspection records} of the multiple sub-projects to obtain {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects. Specifically, in the embodiment of the present application, step S42 includes: passing each of the {project progress reports and quality inspection records} of the multiple sub-projects through a text semantic understanding device based on the DeBERTa model to obtain {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects. Correspondingly, considering that each of the {project progress reports and quality inspection records} contains information about the project progress, such as the comparison between the actual progress and the planned progress of the project, the semantic information of the completed tasks, and the semantic information of the quality inspection records obtained by inspecting and testing the project product or service quality, therefore, in order to further understand and analyze the semantic information expressed in each data and the text meaning between the contexts, in the technical solution of the present application, each of the {project progress reports and quality inspection records} of the multiple sub-projects is passed through a text semantic understanding device based on the DeBERTa model to obtain {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects. It should be understood that the DeBERTa model has powerful performance in the field of natural language processing. By improving the relative position encoding and decoupling the attention mechanism, it can capture the semantic information and context relationship in the text more accurately. Project progress reports and quality inspection records usually contain rich professional terms, complex semantic expressions, and context dependencies. Using the DeBERTa model can better understand the meaning of these texts and extract more accurate and comprehensive semantic features compared to some other simple text processing methods to accurately parse and understand their semantics.

[0032] In step S43, for each group of corresponding project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors in the {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects, project-quality core semantic feature interaction representation is performed to obtain multiple sub-project progress report-quality inspection record semantic interaction representation vectors. Specifically, Figure 4 FIG. is a flowchart of step S43 in the data management method for a digital decision support platform according to an embodiment of the present application. As Figure 4As shown, step S43 includes: S431, extracting the project progress report text semantic coding vector and quality inspection record semantic coding vector of the first sub-project from the {project progress report text semantic coding vector and quality inspection record semantic coding vector} of the multiple sub-projects; S432, respectively calculating the feature granularity feature and eigenvalue granularity feature between the project progress report text semantic coding vector and quality inspection record semantic coding vector of the first sub-project to obtain a project progress report-quality semantic feature granularity response interaction coding vector and a project progress report-quality semantic feature eigenvalue granularity response interaction coding vector; S433, fusing the project progress report-quality semantic feature granularity response interaction coding vector and the project progress report-quality semantic feature eigenvalue granularity response interaction coding vector to obtain a progress report-quality inspection record semantic interaction representation vector of the first sub-project.

[0033] It should be understood that although the project progress report and quality inspection record are texts that describe the project from different perspectives, there is actually a close potential connection between them. For example, a delay in project progress may affect the results of quality inspection, or a situation where the quality does not meet the standard may lead to an adjustment of the project progress. In order to be able to uncover these connections hidden behind the text and thus more comprehensively understand the actual situation of the project. Based on this, the present application introduces a project-quality core semantic feature interaction representation mechanism to perform semantic interaction on each corresponding project progress report text semantic coding vector and quality inspection record semantic coding vector in the {project progress report text semantic coding vector and quality inspection record semantic coding vector} of the multiple sub-projects to obtain multiple sub-project progress report-quality inspection record semantic interaction representation vectors. In particular, in this mechanism, the self-correlation decoupling technology is used to accurately extract the core anchor point information. At the same time, through two-level interaction modeling, the internal connection between the project progress and quality semantic coding vectors is deeply explored from the two dimensions of feature granularity and eigenvalue granularity, which helps the model to more comprehensively understand the project situation.

[0034] Specifically, first, extract the semantic encoding vectors of the project progress report text and the semantic encoding vectors of the quality inspection records of the first sub-project from the {semantic encoding vectors of the project progress report text and the semantic encoding vectors of the quality inspection records} of the multiple sub-projects. It should be understood that in a complex project, it is usually composed of multiple sub-projects, and each sub-project has its unique characteristics and progress. Extracting the {semantic encoding vectors of the project progress report text and the semantic encoding vectors of the quality inspection records} of the first sub-project can conduct a more in-depth and detailed analysis of this sub-project. That is, by focusing on a single sub-project, the model can more clearly understand the specific situation of its progress and quality, as well as the potential relationship between the two, and then can more accurately grasp the potential risk factors in this sub-project, thereby laying a foundation for the overall evaluation of subsequent projects.

[0035] Specifically, in the embodiment of the present application, the step S432 includes: constructing a semantic autocorrelation association matrix of the semantic encoding vectors of the project progress report text and the semantic encoding vectors of the quality inspection records of the first sub-project to obtain a semantic autocorrelation association matrix of the project progress report text and a semantic autocorrelation association matrix of the quality inspection records. This process can be expressed by the formula:

[0036]

[0037] where v1 is the semantic encoding vector of the project progress report text, v2 is the semantic encoding vector of the quality inspection record, T is the transpose operation, and φ(·) is a feature mapping function, such as a linear mapping or a non-linear kernel function. M1 is the semantic autocorrelation association matrix of the project progress report text, and M2 is the semantic autocorrelation association matrix of the quality inspection records;

[0038] Perform self-correlation core information decoupling and anchoring on the semantic autocorrelation association matrix of the project progress report text and the semantic autocorrelation association matrix of the quality inspection records respectively to obtain a semantic core information anchoring encoding vector of the project progress report text and a semantic core information anchoring encoding vector of the quality inspection records. This process can be expressed by the formula:

[0039]

[0040] where M1 is the semantic autocorrelation association matrix of the project progress report text, M2 is the semantic autocorrelation association matrix of the quality inspection records, f Anchor (·) is the operation of performing core information decoupling and anchoring, decouple(·) is the feature decoupling operation, x 11 , x 12 , x 1i and x 1n are respectively the row vectors of M1, W 1i and b1i are x respectively 1i the corresponding project progress report weight matrix and project progress report bias vector, is matrix multiplication, is the project progress report scoring vector, e 1i is the i-th project progress report text semantic significance factor in the set of project progress report text semantic significance factors, Sigmoid is the normalization function, a 1i is the i-th normalized project progress report text semantic significance factor in the set of normalized project progress report text semantic significance factors, n is the number of row vectors in the M1, c1 is the project progress report text semantic core information anchoring coding vector, x 21 , x 22 , x 2i and x 2m are the respective row vectors of the M2, W 2i and b 2i are x respectively 2i the corresponding quality inspection record weight matrix and quality inspection record bias vector, is the quality inspection record scoring vector, e 2i is the i-th quality inspection record semantic significance factor in the set of quality inspection record semantic significance factors, a 2i is the i-th normalized quality inspection record semantic significance factor in the set of normalized quality inspection record semantic significance factors, m is the number of row vectors in the M2, and the quantities of m and n are equal, c2 is the quality inspection record semantic core information anchoring coding vector;

[0041] Perform project progress report-quality feature granularity response interaction coding on the project progress report text semantic core information anchoring coding vector and the quality inspection record semantic core information anchoring coding vector to obtain the project progress report-quality semantic feature granularity response interaction coding vector, and this process can be expressed by the formula:

[0042]

[0043] wherein, c1 is the project progress report text semantic core information anchoring coding vector, c2 is the quality inspection record semantic core information anchoring coding vector, is pointwise addition by position, W VT and b VT are the response weight matrix and response bias vector respectively, tanh is the hyperbolic tangent function, E granular is the project progress report-quality semantic feature granularity response interaction coding vector;

[0044] An item progress report - quality eigenvalue granularity response interactive encoding is performed on the item progress report text semantic core information anchored encoding vector and the quality inspection record semantic core information anchored encoding vector to obtain the item progress report - quality semantic eigenvalue granularity response interactive encoding vector. This process can be expressed by the formula:

[0045]

[0046] where c1 is the item progress report text semantic core information anchored encoding vector, c2 is the quality inspection record semantic core information anchored encoding vector, and E value is the item progress report - quality semantic eigenvalue granularity response interactive encoding vector.

[0047] It should be understood that by constructing the semantic autocorrelation association matrix of the item progress report text semantic encoding vector and the quality inspection record semantic encoding vector of the first sub - project, the originally hidden correlations between the components in the original feature vector can be clearly shown. Taking the item progress report text semantic encoding vector as an example, it may cover semantic information such as the specific completion status of the tasks of this sub - project, the actual input and usage status of resources, etc. These information are originally intertwined in the vector. Through the constructed item progress report text semantic autocorrelation association matrix, the potential relationship between the task completion status and resource usage can be clearly presented. For example, insufficient resources may lead to a lag in task progress; or the relationship between the reasons for schedule delays and the task execution order, etc. For the quality inspection record semantic encoding vector, this vector contains semantic contents such as the standards based on which the inspection is carried out, the final inspection results, and the detailed descriptions of the discovered defects. After constructing the quality inspection record semantic autocorrelation association matrix, the correlation between the inspection standards and the inspection results can be clearly seen. For example, whether stricter inspection standards correspond to more accurate inspection results, and the internal relationship between the defect description and the inspection method, etc. That is, the generated semantic autocorrelation association matrix not only retains the global structure of the original feature vector but also captures its internal semantic patterns.

[0048] Accordingly, considering that the semantic autocorrelation correlation matrix of the project progress report text and the semantic autocorrelation correlation matrix of the quality inspection record contain a large amount of semantic information about the progress and quality of sub-projects, not all of this information is equally important for the overall evaluation and understanding of subsequent projects. In actual data analysis, it is necessary to screen out the key and core parts from the complex information in order to make decisions more efficiently. Based on this, it is necessary to perform autocorrelation core information decoupling and anchoring processing on the semantic autocorrelation correlation matrix of the project progress report text and the semantic autocorrelation correlation matrix of the quality inspection record respectively. Through autocorrelation core information decoupling and anchoring processing, by means of feature distillation and other means, it is possible to effectively highlight the parts in the original project progress report text semantics and quality inspection record semantics that are highly correlated with the core semantics. For example, in the generated semantic core information anchoring coding vector of the project progress report text, the key factors directly affecting the project progress will be highlighted, such as the completion status of tasks on the critical path, the reasonable allocation of resources, etc.; in the semantic core information anchoring coding vector of the quality inspection record, the information related to core quality semantics such as quality standards and defect severity will be highlighted. In this way, in subsequent analysis and decision-making, the key information of sub-projects can be grasped more quickly and accurately. It should be noted that the "anchor" concept implicit in the semantic core information anchoring coding vector of the project progress report text and the semantic core information anchoring coding vector of the quality inspection record is intended to capture global information. By constraining the information entropy, these "anchors" can ensure semantic consistency in the compressed space, enabling the model to have a more stable understanding and expression of core semantics when processing different project data. For example, in the project progress reports of different projects, the core anchor representation can maintain a consistent understanding of the key factors of project progress, avoiding semantic deviations caused by data changes, thereby improving the generalization ability of the model and enabling it to be more accurately applied to different project scenarios.

[0049] It should be understood that performing project progress report-quality feature granularity response interaction encoding on the semantic core information anchored encoding vector of the project progress report text and the semantic core information anchored encoding vector of the quality inspection record can accurately capture the semantic coupling between project progress and quality at the micro level. Specifically, in the semantic core information anchored encoding vector of the project progress report text, semantic units such as task progress nodes and resource allocation progress are included; the semantic core information anchored encoding vector of the quality inspection record covers semantic units such as the achievement of quality standards, defect types, and severity. By exploring the response relationships between the two in these semantic units, the internal connections at the micro level can be mined. For example, if the progress of a key task is delayed, it may be related to the material defects found in the quality inspection of this task. Through feature granularity response interaction encoding, this micro-level semantic coupling can be captured, and then the hidden causal relationship between project progress and quality can be revealed, providing a more detailed perspective for in-depth understanding of the actual situation of the project. That is, through interaction encoding at the feature granularity level, the accuracy of the sub-project analysis can be comprehensively improved. It deeply integrates and analyzes the core information of project progress and quality at the micro level, enabling the model to gain insights into the project status from a more subtle perspective. When analyzing potential project risks, it no longer solely relies on macro indicators such as schedule delays or quality problems, but can delve into the specific semantic unit level to analyze the mutual influence between schedule and quality-related factors, thereby more accurately assessing the risk situation of the sub-project.

[0050] Correspondingly, considering that although the semantic core information anchored encoding vector of the project progress report text and the semantic core information anchored encoding vector of the quality inspection record have extracted key information from the semantic level after previous processing, staying only at the semantic abstraction layer is not sufficient to comprehensively understand the complex relationship between project progress and quality. Through project progress report-quality feature value granularity response interaction encoding, returning to more underlying numerical operations can delve into the numerical level of the data and mine the interdependent relationships that are not easily detected at the semantic level. Specifically, through element-wise operations, the interdependent relationship between project progress and quality feature values can be accurately characterized. For example, in the progress report, there are certain associations between feature values such as the quantity and time of resource investment and feature values such as the frequency and severity of defect occurrences in the quality inspection record. Through feature value granularity response interaction encoding, these associations can be clearly revealed, which is beneficial to enriching the representation dimensions of the interaction modeling between the original data and further enhancing the model's representation ability of the relationship between project progress and quality.

[0051] Finally, fuse the project progress report-quality semantic feature granularity response interaction encoding vector and the project progress report-quality semantic feature value granularity response interaction encoding vector to obtain the progress report-quality inspection record semantic interaction representation vector of the first sub-project. The above process can be expressed by the formula:

[0052] V f = concat[E granular ; E value

[0053] where E granular is the project progress report - quality semantic feature granularity response interaction coding vector, E value is the project progress report - quality semantic feature value granularity response interaction coding vector, concat[·;·] is the concatenation operation, and V f is the progress report - quality detection record semantic interaction representation vector of the said progress report.

[0054] It should be understood that the project progress report - quality semantic feature granularity response interaction coding vector focuses on the relationship mining between semantic units and reveals the coupling situation of project progress and quality at the micro - semantic level; while the project progress report - quality semantic feature value granularity response interaction coding vector focuses on the numerical relationship at the data bottom layer. By fusing these two vectors, information at different levels from semantics to numerics can be integrated to construct a progress report - quality detection record semantic interaction representation vector that comprehensively reflects the relationship between project progress and quality. In this way, a comprehensive expression of the sub - project situation can be realized, avoiding information omission.

[0055] Preferably, in another example of this application, fusing the project progress report - quality semantic feature granularity response interaction coding vector and the project progress report - quality semantic feature value granularity response interaction coding vector to obtain the progress report - quality detection record semantic interaction representation vector of the first sub - project includes: performing project progress report - quality growth distribution perturbation correction on the project progress report - quality semantic feature granularity response interaction coding vector and the project progress report - quality semantic feature value granularity response interaction coding vector to obtain a corrected project progress report - quality semantic feature granularity response interaction coding vector composed of multiple corrected project progress report - quality semantic feature granularity feature values and a corrected project progress report - quality semantic feature value granularity response interaction coding vector composed of multiple corrected project progress report - quality semantic feature value granularity feature values; fusing the corrected project progress report - quality semantic feature granularity response interaction coding vector and the corrected project progress report - quality semantic feature value granularity response interaction coding vector to obtain the progress report - quality detection record semantic interaction representation vector. The above process is expressed by the formula as follows:

[0056]

[0057]

[0058] V f ​= concat[E′ granular ; E′ value

[0059] where E gi is the eigenvalue at the i-th position in E granular , E vi is the eigenvalue at the i-th position in E value , cos is the cosine function, E′ gi is the corrected eigenvalue of E gi , E′ vi is the corrected eigenvalue of E vi , E′ granular is the corrected project progress report - quality semantic feature granularity response interaction coding vector, E′ value is the corrected project progress report - quality semantic eigenvalue granularity response interaction coding vector, V f is the semantic interaction representation vector of the progress report - quality detection record.

[0060] Specifically, here, considering the difference between the modeling representations of the project progress report - quality semantic feature granularity interaction and the project progress report - quality semantic eigenvalue granularity interaction may cause unstable perturbations in the feature manifold interface of the fused progress report - quality detection record semantic interaction representation vector with multi - granularity information expression. Taking the project progress report - quality semantic eigenvalue granularity interaction representation as an extensible constraint representation, for the overall feature interaction distribution growth under the eigenvalue - by - eigenvalue diffusion process, based on the growth index representation under the extensible constraint, model the interface shape perturbation deviation of the overall distribution of the project progress report - quality semantic feature granularity interaction. That is, taking the interaction interface gradient under each eigenvalue granularity as a perturbation contribution factor, determine the growth index stabilization contribution of the gradient diffusion under eigenvalue interdependence, so as to achieve the growth mode gradient correction dominated by perturbation stabilization and improve the manifold interface stability of the semantic interaction representation vector of the progress report - quality detection record.

[0061] ​In step S44, a multi-dimensional semantic comprehensive analysis is performed on the semantic interaction representation vectors of the multiple sub-project progress reports - quality inspection records to obtain the semantic interaction feature vectors of the project comprehensive progress report - quality inspection records. Specifically, in the embodiment of the present application, step S44 includes: passing the semantic interaction representation vectors of the multiple sub-project progress reports - quality inspection records through a sub-project multi-dimensional semantic comprehensive analyzer based on Transformer to obtain the semantic interaction feature vectors of the project comprehensive progress report - quality inspection records. Correspondingly, considering that each semantic interaction representation vector of the project progress report - quality inspection records contains the semantic information combining the progress and quality at the sub-project level, and there may be complex associations among the sub-projects, such as resource competition, sequential dependencies in the workflow, etc. Based on this, in the technical solution of the present application, the semantic interaction representation vectors of the multiple sub-project progress reports - quality inspection records are passed through a sub-project multi-dimensional semantic comprehensive analyzer based on Transformer to utilize the self-attention mechanism in the Transformer architecture to dynamically focus on different parts, further integrate this local information, and generate a global and multi-dimensional semantic interaction feature vector of the project comprehensive progress report - quality inspection records, providing a key basis for the overall evaluation of the project.

[0062] The following is a detailed elaboration of a specific implementation process of "passing the semantic interaction representation vectors of the multiple sub-project progress reports - quality inspection records through a sub-project multi-dimensional semantic comprehensive analyzer based on Transformer to obtain the semantic interaction feature vectors of the project comprehensive progress report - quality inspection records":

[0063] First is the data preparation stage. This is the foundation of the entire implementation process and is crucial for the accuracy and effectiveness of subsequent processing. It is necessary to collect the semantic interaction representation vectors of the progress reports - quality inspection records of multiple sub-projects and integrate them into a unified dataset to ensure the consistency of the data format. During this process, a comprehensive check of the data should be carried out to see if there are missing values, outliers, etc. For missing values, according to the characteristics of the data and business logic, methods such as mean filling and median filling can be used for processing; for outliers, they need to be corrected or directly removed according to the actual situation.

[0064] After completing data preparation, model selection and initialization need to be carried out. According to actual requirements and data characteristics, select a suitable Transformer-based model architecture, such as the original Transformer model, BERT and its variants, etc. They have powerful capabilities in processing sequence data and semantic information. If there is a suitable pre-trained model, load its parameters for initialization because the pre-trained model has been trained on large-scale text data and has learned rich language knowledge and semantic patterns, which can accelerate the convergence speed of the model and improve performance. If there is no suitable pre-trained model, the parameters of the model can also be randomly initialized. In addition, it is necessary to reasonably set the hyperparameters of the model, such as learning rate, batch size, number of training epochs, number of attention heads, etc. These hyperparameters will have an important impact on the effect and performance of the model, and the optimal values need to be determined through training and experience. Note that before the model is actually applied to data processing, it needs to be trained using the collected data to fine-tune the model so that it can effectively extract the key features of the input data.

[0065] Then enter the data input processing link. Multiple sub-project progress report-quality inspection record semantic interaction representation vectors need to be arranged in a certain order to construct an input sequence, which can be arranged according to the sub-project number, time sequence, etc. Since the Transformer model itself does not have position information, position encoding needs to be added to each vector in the input sequence. Position encoding can help the model capture the relative position relationship of vectors in the sequence, and thus better understand semantic information. Usually, a combination of sine and cosine functions can be used to generate it. At the same time, according to the length of the input sequence and task requirements, create a mask matrix to mask the positions that do not need to be concerned when calculating attention scores, such as padding positions or future time steps.

[0066] After the data input processing is completed, the multi-head self-attention mechanism calculation begins. For each vector in the input sequence, three vectors, namely query, key, and value, are generated through linear transformation. These three vectors are used to calculate the attention scores to reflect the correlation between vectors. When calculating the attention scores, for each query vector, its similarity with all key vectors needs to be calculated. A common calculation method is the dot product method. After obtaining the attention scores, the mask matrix is applied to them to mask the positions that do not need to be attended to, ensuring that the model only considers valid vectors when calculating the attention weights. Then, the softmax function is applied to the masked attention scores to convert them into attention weights, which represent the proportion of each vector in integrating information. Finally, according to the attention weights, a weighted sum is performed on the value vectors to obtain the output of each vector after being processed by the self-attention mechanism. To enable the model to capture semantic information from multiple different perspectives, the multi-head self-attention mechanism is adopted. The input sequence is projected into multiple low-dimensional subspaces respectively, and the self-attention calculation is independently performed in each subspace. Finally, the outputs of each subspace are concatenated together.

[0067] After being processed by the multi-head self-attention mechanism, the vectors need to enter the feed-forward neural network for further processing. The feed-forward neural network consists of two linear transformation layers, with a non-linear activation function (such as ReLU) in between. By introducing non-linearity through the non-linear activation function, the model can learn more complex semantic relationships and increase the model's expressive ability. After being processed by the feed-forward neural network, the vectors are mapped to a higher-dimensional feature space to further extract and integrate semantic information.

[0068] After the multi-head self-attention mechanism and the feed-forward neural network, layer normalization operations are performed respectively. Layer normalization can normalize the feature dimensions of each vector, enabling the model to avoid the problems of gradient vanishing or gradient explosion. At the same time, before layer normalization, a residual connection is used to add the input vector to the vector processed by the self-attention mechanism or the feed-forward neural network. The residual connection can help the model better learn the difference between the input and the output, alleviate the problem of gradient vanishing, and accelerate the convergence of the model.

[0069] Then, the above multi-head self-attention mechanism, feed-forward neural network, layer normalization, and residual connection operations are combined to form an encoding layer, and multiple encoding layers are stacked to form a deep neural network. Each encoding layer will further extract and transform the output of the previous layer, enabling the model to gradually learn higher-level and more abstract semantic features. The input sequence will be processed sequentially through multiple encoding layers. As the number of layers increases, the semantic information contained in the vector sequence will become richer and more abstract.

[0070] After being processed by multiple encoding layers, a sequence containing multiple vectors is obtained. At this time, according to the task requirements, a suitable aggregation method needs to be selected to aggregate these vectors into a project comprehensive progress report - quality inspection record semantic interaction feature vector. Common aggregation methods include taking the average value, the maximum value, pooling operations, etc. Process the vector sequence according to the selected aggregation method, and finally obtain the project comprehensive progress report - quality inspection record semantic interaction feature vector. This feature vector contains the comprehensive semantic information between multiple sub - project progress reports and quality inspection records, and can be used for the final risk classification task.

[0071] In step S45, based on the project comprehensive progress report - quality inspection record semantic interaction feature vector, a risk prediction result is obtained, and the risk prediction result is used to indicate whether there is a risk in the project. Specifically, in the embodiment of the present application, step S45 includes: passing the project comprehensive progress report - quality inspection record semantic interaction feature vector through a risk predictor based on a classifier to obtain the risk prediction result. That is, using the project comprehensive progress report - quality inspection record semantic interaction feature vector obtained by performing multi - dimensional semantic analysis on the multiple sub - project progress report - quality inspection record semantic interaction representation vectors to perform classification processing, so as to intelligently judge whether there is a risk in the project. It should be understood that the classifier has natural advantages in pattern recognition and classification tasks. It is trained based on a large amount of historical project data to enable it to learn the corresponding relationship between different feature vectors and risk states. The project comprehensive progress report - quality inspection record semantic interaction feature vector contains the comprehensive information of the project progress and quality. The classifier can accurately classify it into the "risk exists" or "risk does not exist" category according to the patterns of these feature vectors, so as to realize the prediction of project risks. That is, the risk prediction result can provide timely risk warnings for the project team, enabling the team to know in advance whether there is a risk in the project. Once a risk is identified, the team can immediately take corresponding preventive measures, such as adjusting resource allocation, optimizing the project process, strengthening quality monitoring, etc., so as to reduce the probability and impact of the risk occurrence.

[0072] In summary, step S4 is clearly described. It uses natural language processing technology based on deep learning to perform semantic understanding of each {project progress report and quality inspection record}, and then performs core semantic feature interaction on the {project progress report text semantic features and quality inspection record semantic features} of each sub - project. Based on the multi - dimensional analysis representation between the multiple sub - project progress report - quality inspection record semantic interactions after the interaction, it can intelligently judge whether there is a risk in the project. In this way, more accurate project risk prediction and early warning can be achieved.

[0073] In step S5, the human resource management module is used to manage recruitment, training, performance appraisal, and compensation management. It should be understood that the human resource management module plays a crucial role in the digital decision-making support platform. By integrating and optimizing core human resource functions such as recruitment, training, performance appraisal, and compensation management, it provides a comprehensive employee life cycle management solution for enterprises. The module starts with recruitment management, covering the whole process from job posting, resume screening to interview arrangement and employment notice, and uses artificial intelligence technology to analyze candidate profiles to ensure finding the best candidates. That is, during the recruitment process, the intelligent screening tool can quickly identify the talents most suitable for the job requirements, shortening the recruitment cycle and improving the recruitment quality. For the career development of employees, the module formulates personalized training plans, not only tracking the learning progress of employees, but also flexibly adjusting the course content according to feedback, thus supporting the integration of online learning platforms and improving the training efficiency and effect. At the same time, the performance appraisal part sets clear performance goals, regularly evaluates employee performance, and adopts an all-round feedback mechanism to collect evaluations from colleagues, subordinates, and superiors to ensure that the evaluation results are comprehensive and objective, so as to motivate employees to exert their maximum potential and provide a decision-making basis for management. In terms of compensation management, the module automatically processes salary calculation, tax declaration, and benefit distribution, supports flexible configuration of the compensation structure, and ensures the fairness and transparency of compensation. In addition, detailed payslips and financial statements are provided, simplifying the daily personnel management work, reducing human errors, and improving work efficiency. Generally speaking, the human resource management module can help enterprises simplify administrative affairs, enhance employees' growth opportunities, and establish a fair and effective performance evaluation system, and is an indispensable part of modern enterprises to achieve efficient human resource management.

[0074] In step S6, the cultural management module is utilized to analyze the corporate culture and related activities, and conduct cultural publicity and training. It should be understood that the cultural management module first evaluates the current corporate culture status by collecting and analyzing multi-source data such as employee feedback, social media data, and internal communication records, and identifies its strengths and weaknesses. Based on these analysis results, the enterprise can more accurately understand its own cultural status quo and plan corresponding improvement measures accordingly. At the same time, the module provides tools to support the planning and execution of various cultural activities, such as team-building activities, celebrations, workshops, etc., to enhance the connection among employees and their sense of belonging to the enterprise. To strengthen the cultural dissemination, multiple channels such as the intranet, emails, and social media can be used to release content about the corporate values, mission, and vision, ensuring that all employees can receive consistent and positive cultural information. In addition, the cultural management module has developed targeted cultural training courses to help new and old employees understand and integrate into the corporate culture, covering aspects such as leadership development, teamwork skills, and communication ability improvement, providing support for the growth and development of employees. Finally, the effectiveness of cultural promotion activities can be monitored by setting key performance indicators (KPIs), such as participation rate, satisfaction score, etc., and the cultural management strategy can be adjusted according to the feedback to continuously optimize the cultural construction process. Generally speaking, the cultural management module can help the enterprise establish and maintain a healthy and positive corporate culture through systematic analysis, planning, and execution. This can not only attract and retain talents, but also be conducive to stimulating the creativity and teamwork spirit of employees.

[0075] In summary, the data management method for a digital decision support platform based on the embodiments of the present application is elucidated. It formulates and adjusts strategic plans through the strategic management module to ensure the matching of resource allocation and strategic goals, monitors and analyzes financial data through the financial management module to dynamically control the financial budget and capital security, plans and evaluates the effectiveness of marketing activities through the market management module, tracks the project progress and quality through the project management module, and conducts risk early warning based on the progress reports and quality inspection records of multiple sub-projects, processes recruitment, training, performance appraisal, and salary management through the human resource management module, and analyzes the corporate culture and conducts cultural publicity and training activities through the cultural management module. In this way, through the digital and intelligent management of each module, the decision-making efficiency and management level of the enterprise can be effectively improved.

Claims

1. A data management method for a digital decision support platform, the digital decision support platform including a strategic management module, a financial management module, a market management module, a project management module, a human resources management module, and a cultural management module, characterized in that, The described data management method includes: Using the strategic management module to formulate strategic plans and monitor strategic processes, and analyzing the matching between strategic goals and resource allocation to dynamically adjust the strategic goals; Using the financial management module to monitor and analyze financial data to dynamically control financial budgets and fund security; Using the market management module to execute the planning and effectiveness evaluation of marketing activities; Using the project management module to track the progress and quality of projects and conduct risk warnings. Among them, using the project management module to conduct risk warnings includes conducting risk warnings based on the project-quality semantic core association information between the {project progress reports and quality inspection records} of multiple sub-projects; Using the human resource management module for management recruitment, training, performance appraisal, and salary management; Using the culture management module to analyze corporate culture and related activities, and conduct cultural publicity and training.

2. The data management method for a digital decision support platform according to claim 1, characterized in that, Conducting the risk warning based on the project-quality semantic core association information between the {project progress reports and quality inspection records} of multiple sub-projects includes: Obtaining the {project progress reports and quality inspection records} of the multiple sub-projects; Respectively performing text semantic understanding on each of the {project progress reports and quality inspection records} in the {project progress reports and quality inspection records} of the multiple sub-projects to obtain the {project progress report text semantic coding vectors and quality inspection record semantic coding vectors} of the multiple sub-projects; Performing project-quality core semantic feature interaction representation on each group of corresponding project progress report text semantic coding vectors and quality inspection record semantic coding vectors in the {project progress report text semantic coding vectors and quality inspection record semantic coding vectors} of the multiple sub-projects to obtain multiple sub-project progress report-quality inspection record semantic interaction representation vectors; Performing multi-dimensional semantic comprehensive analysis on the multiple sub-project progress report-quality inspection record semantic interaction representation vectors to obtain project comprehensive progress report-quality inspection record semantic interaction feature vectors; Based on the project comprehensive progress report-quality inspection record semantic interaction feature vectors, obtaining a risk prediction result, and the risk prediction result is used to indicate whether there is a risk in the project.

3. The data management method for a digital decision support platform according to claim 2, wherein Respectively performing text semantic understanding on each of the {project progress reports and quality inspection records} in the {project progress reports and quality inspection records} of the multiple sub-projects to obtain the {project progress report text semantic coding vectors and quality inspection record semantic coding vectors} of the multiple sub-projects, including: respectively passing each of the {project progress reports and quality inspection records} in the {project progress reports and quality inspection records} of the multiple sub-projects through a text semantic understanding device based on the DeBERTa model to obtain the {project progress report text semantic coding vectors and quality inspection record semantic coding vectors} of the multiple sub-projects.

4. The data management method for a digital decision support platform according to claim 3, characterized in that, Performing project-quality core semantic feature interaction representation on each corresponding pair of project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors in the {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects to obtain multiple sub-project progress report-quality inspection record semantic interaction representation vectors, including: Extracting the {project progress report text semantic encoding vector and quality inspection record semantic encoding vector} of the first sub-project from the {project progress report text semantic encoding vectors and quality inspection record semantic encoding vectors} of the multiple sub-projects; Calculating the feature granularity features and eigenvalue granularity features between the project progress report text semantic encoding vector and the quality inspection record semantic encoding vector of the first sub-project respectively to obtain a project progress report-quality semantic feature granularity response interaction encoding vector and a project progress report-quality semantic eigenvalue granularity response interaction encoding vector; Fusing the project progress report-quality semantic feature granularity response interaction encoding vector and the project progress report-quality semantic eigenvalue granularity response interaction encoding vector to obtain the progress report-quality inspection record semantic interaction representation vector of the first sub-project.

5. The data management method for a digital decision support platform according to claim 4, wherein Calculating the feature granularity features and eigenvalue granularity features between the project progress report text semantic encoding vector and the quality inspection record semantic encoding vector of the first sub-project respectively to obtain a project progress report-quality semantic feature granularity response interaction encoding vector and a project progress report-quality semantic eigenvalue granularity response interaction encoding vector, including: Constructing a semantic autocorrelation association matrix of the project progress report text semantic encoding vector and the quality inspection record semantic encoding vector of the first sub-project to obtain a project progress report text semantic autocorrelation association matrix and a quality inspection record semantic autocorrelation association matrix; Performing autocorrelation core information decoupling and anchoring on the project progress report text semantic autocorrelation association matrix and the quality inspection record semantic autocorrelation association matrix respectively to obtain a project progress report text semantic core information anchoring encoding vector and a quality inspection record semantic core information anchoring encoding vector; Performing project progress report-quality feature granularity response interaction encoding on the project progress report text semantic core information anchoring encoding vector and the quality inspection record semantic core information anchoring encoding vector to obtain the project progress report-quality semantic feature granularity response interaction encoding vector; Performing project progress report-quality eigenvalue granularity response interaction encoding on the project progress report text semantic core information anchoring encoding vector and the quality inspection record semantic core information anchoring encoding vector to obtain the project progress report-quality semantic eigenvalue granularity response interaction encoding vector.

6. The data management method for a digital decision support platform according to claim 5, characterized in that, Fusing the project progress report-quality semantic feature granularity response interaction encoding vector and the project progress report-quality semantic eigenvalue granularity response interaction encoding vector to obtain the progress report-quality inspection record semantic interaction representation vector of the first sub-project, including: Perform project progress report-quality growth distribution perturbation correction on the project progress report-quality semantic feature granularity response interaction coding vector and the project progress report-quality semantic feature value granularity response interaction coding vector to obtain a corrected project progress report-quality semantic feature granularity response interaction coding vector composed of multiple corrected project progress report-quality semantic feature granularity feature values and a corrected project progress report-quality semantic feature value granularity response interaction coding vector composed of multiple corrected project progress report-quality semantic feature value granularity feature values; Fuse the corrected project progress report-quality semantic feature granularity response interaction coding vector and the corrected project progress report-quality semantic feature value granularity response interaction coding vector to obtain the progress report-quality detection record semantic interaction representation vector.

7. The data management method for a digital decision support platform according to claim 6, characterized in that Perform sub-project multi-dimensional semantic comprehensive analysis on the multiple sub-project progress report-quality detection record semantic interaction representation vectors to obtain a project comprehensive progress report-quality detection record semantic interaction feature vector, including: passing the multiple sub-project progress report-quality detection record semantic interaction representation vectors through a sub-project multi-dimensional semantic comprehensive analyzer based on Transformer to obtain the project comprehensive progress report-quality detection record semantic interaction feature vector.

8. The data management method for a digital decision support platform according to claim 7, characterized in that Based on the project comprehensive progress report-quality detection record semantic interaction feature vector, obtain a risk prediction result, including: passing the project comprehensive progress report-quality detection record semantic interaction feature vector through a risk predictor based on a classifier to obtain the risk prediction result.

Images