A project screening intelligent supervision system and method based on data analysis

Abstract

The application discloses a project screening intelligent supervision system and method based on data analysis, relates to the technical field of data analysis, and deploys an intelligent quantitative recommendation platform in an enterprise, and stores all department demand data and all project parameters of projects in the intelligent quantitative recommendation platform when the project is screened; the intelligent quantitative recommendation platform carries out quantitative calculation on all data and outputs basic parameters; the basic matching degree and the subjective coefficient are iteratively adjusted according to the initial matching state of the remaining demand nodes for each demand node, and then convergence judgment is carried out; when the judgment converges, the enterprise matching degree calculation is carried out; when the judgment does not converge, the iterative adjustment is continuously carried out; conflict detection is carried out for each iteration, and when there is a conflict and the conflict continues for more than a preset number of rounds, an arbitration mechanism is executed; finally, a decision report is generated and output.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, specifically to an intelligent monitoring system and method for project screening based on data analysis. Background Technology

[0002] In the fields of intelligent recommendation, multi-source information fusion, and distributed data processing, it is often necessary to merge the evaluation results of multiple independent data sources on the same batch of objects to be screened in order to generate a global recommendation ranking or decision-making basis. For example, in personalized recommendation systems, different user behavior channels can be regarded as different demand nodes, requiring comprehensive scoring of products; in IoT edge computing, multiple sensor nodes evaluate the resource requirements of the same computing task, requiring the selection of the optimal task allocation strategy. Furthermore, in enterprise project screening, due to the large number of departments within an enterprise, it is also necessary to fuse and analyze different information. In existing technologies, data fusion often employs weighted average fusion and distributed consensus algorithms. However, existing weighted average fusion methods typically assume that the evaluation values ​​provided by different data sources have consistent dimensions and credibility. When there are significant conflicts between the evaluation values ​​of different request nodes for the same selection node, a simple weighted average will yield an intermediate value, but this value cannot reflect the actual degree of disagreement between nodes. Furthermore, when conflicts persist, the fusion result often deviates from the true situation.

[0003] Existing distributed consensus algorithms allow nodes to iteratively update their estimates by exchanging information with neighbors, eventually leading to a consensus across the entire network. However, these algorithms only handle one-dimensional values ​​and cannot address the differences in "preference directions" between nodes in a multi-dimensional feature space. When nodes have different preference directions in the multi-dimensional feature space, existing consensus algorithms require a large number of iterations to converge, and may even exhibit oscillations and non-convexity in non-convex preference spaces, severely impacting system real-time performance.

[0004] In summary, to address the technical shortcomings of existing technologies, such as the fact that the fusion results often deviate from reality and the inability to handle the differences in "preference direction" between nodes in a multi-dimensional feature space, there is an urgent need to provide a data analysis-based intelligent supervision system and method for project screening. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent monitoring system and method for project screening based on data analysis, so as to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A data-driven intelligent monitoring method for project screening includes the following steps: S100. Deploy an intelligent quantitative recommendation platform in the enterprise. When selecting projects, the enterprise inputs all departmental requirement data and all project parameters into the intelligent quantitative recommendation platform for storage. The intelligent quantitative recommendation platform performs quantitative calculations on all data and outputs basic parameters. Furthermore, the specific steps for outputting the basic parameters are as follows: S101. During project screening, the enterprise inputs all departmental demand data and all project parameters into the intelligent quantitative recommendation platform for storage. Each department is used as a demand node and each project is used as a screening node. When the enterprise starts the project screening task, the intelligent quantitative recommendation platform receives the task instruction and performs quantitative calculations on all the data stored in the intelligent quantitative recommendation platform. The basic parameters include basic matching degree, subjective coefficient, preference direction angle, and dynamic weight; The intelligent quantitative recommendation platform uses demand data from demand nodes and project parameters from filtering nodes to quantify the coverage rate of each filtering node. The coverage rate is defined as the number of covered parameters divided by the total number of demand data items. The number of covered parameters represents the number of project parameters whose values ​​are greater than the demand data values. By uniformly inputting departmental demand data and project parameters from different sources and in different formats into the intelligent quantitative recommendation platform, and using quantitative indicators such as coverage and completeness, unstructured data is transformed into computable numerical vectors, thus solving the technical problem of the inability to directly integrate multi-source heterogeneous data.

[0007] Calculate the parameter completeness of each filter node. The parameter completeness = number of valid fields ÷ total number of information fields. The number of valid fields represents the number of fields in the project parameters input to the project, excluding missing and outlier values ​​in the information fields themselves. Outlier values ​​represent field data that exceed the average value by 3 times. Weights were assigned to coverage and parameter completeness, and the weighted sum was used to calculate the department's basic matching degree with the project; the weights were set by the department staff. S102. Set a subjective comparison slider in the intelligent quantitative recommendation platform. Place two types of filtering nodes at both ends of the subjective comparison slider. The enterprise staff drag the slider according to their subjective intentions. Set the middle value of the subjective comparison slider to 0 and the two ends to 1; 0 means completely uncertain and 1 means very certain. Calculate the subjective coefficient of each filtering node based on the subjective dragging results of the staff. Repeat the calculation for all types of items. S103. Calculate the difference between all project parameters of different filtering nodes, extract the two project parameters with the largest difference in the filtering nodes as angle features, construct a plane rectangular coordinate system based on the angle features, extract the feature points of each project in the plane rectangular coordinate system, and extract the angle between the feature points and the x-axis in the plane rectangular coordinate system as the project type angle. The angle value range is 0 to 2π. Extract the historical filter item set of the demand node, calculate the two average angle feature values ​​of the historical filter items of the demand node, input them into the Cartesian coordinate system to obtain the demand feature point, and extract the angle between the demand feature point and the x-axis in the Cartesian coordinate system as the preference center angle. The angle value ranges from 0 to 2π. In each demand node, the preference direction angle is calculated using the item type angle and the preference center angle, where the preference direction angle = (item type angle - preference center angle) mod 2π. An initial matching state is constructed using the basic matching degree, subjective coefficient, and preference direction angle. A two-dimensional coordinate system is constructed by extracting the item parameters with the largest feature difference, compressing the high-dimensional item features into a single angle value, so that the preference difference between nodes can be measured by mathematical distance, providing a calculable geometric basis for subsequent conflict detection.

[0008] S104. Extract the demand data volume and demand time of each demand node in the intelligent quantitative recommendation platform. The demand data volume represents the total amount of demand data information input by the demand node, and the demand time represents the duration of each type of demand data input by each demand node. After normalizing the demand data volume and demand time of each department, a weighted sum is obtained to obtain a dynamic weight. Enterprise staff can customize the parameters used in the system, including learning rate, coupling coefficient, angle penalty factor, confidence decay coefficient, convergence matching degree threshold, and convergence subjective coefficient threshold.

[0009] S200. For each demand node, the basic matching degree and subjective coefficient are iteratively adjusted based on the initial matching state of the other demand nodes. Then, a convergence judgment is made. When convergence is determined, the enterprise matching degree is calculated. If convergence is not achieved, iterative adjustment continues. Furthermore, the specific steps for iteratively adjusting the basic matching degree and subjective coefficient for each demand node based on the initial matching state of the other demand nodes are as follows: S201. For each demand node, its own basic matching degree is iteratively adjusted using the demand node's own preference direction angle in the initial matching state, the preference direction angles of other demand nodes, the basic matching degree, and the dynamic weight. The formula is as follows: ; In the formula, s i，p （t+1） s represents the matching degree between the demand node i and the selection node p at iteration t+1. i，p （t） s represents the basic matching degree of demand node i to filter node p before adjustment at time t. j，p （t） θ represents the basic matching degree of the remaining demand nodes j to the filter node p at time t before adjustment. i，pθ represents the directional angle of preference of demand node i for selection node p. j，p w represents the preference angle of the remaining demand node j towards the selection node p. j （t） d(θ) represents the dynamic weight of the remaining demand node j before adjustment t, β represents the coupling coefficient, γ represents the angle penalty factor, the sigmoid function makes the result between (0, 1), and d(θ) i，p θ j，p The angular frequency difference represents the preference direction of demand node i and department j towards screening node p. The matching degree of each demand node is no longer fixed, but is iteratively updated based on the matching degree of other nodes, the preference direction angle, and dynamic weights. This solves the fundamental defect of traditional weighted fusion methods that only change the weights without changing the original values ​​and cannot eliminate evaluation discrepancies between nodes.

[0010] For the same filtering node p, the specific differences in the preference direction angles between demand nodes i and j are as follows: ; In the iterative formula, the intensity of mutual influence between nodes is modulated by the difference in preference direction angle. Nodes with similar directions have a greater influence, while those with opposite directions have a smaller or even negative influence. This is equivalent to automatically identifying groups of nodes with similar opinions and those with opposing opinions during the fusion process, thus avoiding fusion distortion caused by blind averaging.

[0011] S202. Adjust the subjective coefficient using the basic matching degree and matching degree of demand node i before and after iterative adjustment, using the following formula: ; In the formula, c i，p （t+1） c represents the subjective coefficient of demand node i relative to selection node p at time t+1 after adjustment. i，p （t） This represents the subjective coefficient of demand node i towards selection node p at time t before adjustment, and λ represents the confidence decay coefficient; if the calculation result c i，p （t+1） <0, then c i，p （t+1） Set to 0.01; S203. After adjusting the matching degree and subjective coefficient of all demand nodes in each round, calculate the difference between the matching degree and subjective coefficient of all demand nodes before and after the adjustment, and extract the maximum value △s of the difference in matching degree of all demand nodes. max When △s maxWhen St and all subjective coefficient differences are less than Ct, it is determined that convergence is achieved and enterprise matching degree calculation is performed; otherwise, iterative adjustment of matching degree and subjective coefficients is continued, where St represents the convergence matching degree threshold, and Ct represents the convergence subjective coefficient threshold. At the same time, the maximum change of the matching degree and all changes of the subjective coefficients are monitored, and convergence is determined only when both are lower than the preset threshold. The dual-condition convergence mechanism is stricter than a single threshold, ensuring that the system reaches a real stable state instead of local oscillation, and avoiding early stopping or infinite iteration.

[0012] S300. During iterative adjustment of demand nodes, weight evolution is performed every R rounds to update the dynamic weight of each demand node; Further, the specific steps of updating the dynamic weight of each demand node are as follows: S301. When the iterative adjustment of demand nodes has been performed for R=5 rounds, weight evolution is performed once. The weight evolution refers to calculating the dynamic weight change for each demand node by using the matching degree and preference direction angle of the latest round, with the formula as follows: ; In the formula, δw i represents the dynamic weight change of demand node i, s i，p represents the matching degree of demand node i for the screening node p in the latest round of adjustment, s j，p represents the matching degree of the remaining demand node j for the screening node p in the latest round of adjustment, N represents the total number of demand nodes in the intelligent quantitative recommendation platform, and α represents the learning rate; S302. Update the dynamic weight for each demand node by using the dynamic weight change, with the formula as follows: w i new =w i old +δw i ; in the formula, w i new represents the updated dynamic weight of demand node i, w i old represents the dynamic weight of demand node i before update; if w i new is less than 0, it is set to 0.01; after the update, the sum of the dynamic weights of all demand nodes is made to be 1. Performing evolution and update on the dynamic weights of demand nodes at fixed rounds, and calculating the weight change based on the matching degree and preference direction angle of the latest round, enables the dynamic demand weights to be optimized in real time along with the iterative process of screening by the screening node, no longer limited to the initial set values, always adapting to the actual situation of current screening by the screening node, and correcting the problem that the initial weights do not match the actual demand; S400. During the iterative adjustment of demand nodes, conflict detection is performed for each round of iteration, and when a conflict exists and the conflict duration exceeds the preset number of rounds, an arbitration mechanism is executed; Furthermore, the specific steps for implementing the arbitration mechanism when a conflict exists and the number of rounds of conflict exceeds a preset round threshold are as follows: S401, The preset round number threshold is T conflict When iteratively adjusting all demand nodes, conflict detection is performed after each round of iterative adjustment. The conflict detection is determined using the preference direction angle and the difference in matching degree. Specifically, when d(θ) i，p θ j，p ) > θ angle And |s i，p -s j，p When |>△s, it is determined that there is a conflict between demand node i and demand node j regarding the selection node p, θ angle Δs represents the preference orientation angle conflict threshold, and Δs represents the matching degree conflict threshold. When a conflict is detected and the duration exceeds T rounds... conflict When a conflict occurs, an arbitration mechanism is executed; the arbitration mechanism includes global injection adjustment, local annealing, and manual intervention; arbitration is only triggered when the conflict state continues for more than a certain number of rounds, thus avoiding unnecessary arbitration intervention caused by instantaneous disturbances and improving the system's noise resistance.

[0013] The global adjustment includes constructing a virtual enterprise global perspective demand node A. The preference direction angle of the enterprise global perspective demand node is set to the average of the preference direction angles of all demand nodes. The dynamic weight is normalized to 1.2 times the maximum value of the dynamic weights of all demand nodes. The staff uses the proportion of the same field in the total fields of all demand data and project parameters in the intelligent quantitative recommendation platform as the basic matching degree of the enterprise global perspective demand node. The subjective coefficient is the maximum value among all demand nodes. The enterprise global perspective department A is added to all demand nodes in the intelligent quantitative recommendation platform and iteratively adjusted. Arbitration is considered complete when the conflict disappears within 10 rounds. If the conflict does not disappear within 10 rounds, local annealing is performed. The local annealing process involves halving the coupling coefficient and doubling the subjective coefficient for two conflicting requirement nodes. After iterative adjustment, arbitration is considered complete when the conflict disappears within 5 rounds; otherwise, manual intervention is performed. The manual intervention includes direct intervention by enterprise staff at conflicting demand nodes. The staff sets the preference direction angle and matching degree of conflicting demand nodes i and j to the median value of the two, then eliminates the conflict and iterates and adjusts again.

[0014] The three-tiered arbitration mechanism follows the principle of prioritizing automation and providing a backup of human intervention. Global injection adjustments resolve routine conflicts automatically through virtual global demand nodes, allowing for rapid recovery and iteration without human intervention. Local annealing specifically weakens the coupling relationship between conflict demand nodes, efficiently resolving moderate disagreements. Human intervention serves as the final safeguard, handling extreme disputes and ensuring uninterrupted screening throughout the process. This achieves a perfect balance between intelligent automation and human flexibility, ensuring that all conflicts are resolved reasonably.

[0015] S500: Calculate the enterprise matching degree of each screening node using the matching degree of each demand node and the final iterative value of the subjective coefficient after the iteration adjustment. Furthermore, the specific steps for calculating the enterprise matching degree for each filtering node are as follows: S501. After the demand node iterative adjustment converges, the enterprise matching degree is calculated using the converged matching degree and the final iterative value of the subjective coefficient. The formula is as follows: ; In the formula, s i，p （final） c represents the final iterative value of the matching degree between demand node i and filter node p. i，p （final） E represents the final iterative value of the subjective coefficient of demand node i relative to selection node p. p This represents the matching degree of enterprises to the filter node p; it outputs the matching degree of enterprises to each filter node and the number of convergence rounds.

[0016] S600: Sort all filtering nodes according to enterprise matching degree, mark items for different arbitration results in the arbitration mechanism, and finally generate a decision report for output.

[0017] Furthermore, the specific steps for generating and outputting the decision report are as follows: S601. Sort all screening nodes in descending order of enterprise matching degree. If there are projects that eventually converge through global injection adjustment and local annealing in the arbitration mechanism, mark the corresponding screening node as "low consensus screening node". If there are screening nodes that eventually converge through manual intervention in the arbitration mechanism, mark the corresponding screening node as "disputed screening node". Construct a sorting table for filtering nodes and a conflict resolution log. The sorting table for filtering nodes includes enterprise matching degree and number of convergence rounds. The conflict resolution log includes conflict requirement nodes and arbitration methods. The top K uncontroversial screening nodes in the screening node ranking table are selected for recommendation. Among the top K uncontroversial screening nodes, risk warnings are issued for screening nodes with low consensus. K is set by the staff. The screening node ranking table, conflict resolution log, and recommended screening nodes constitute the output of a decision report.

[0018] A data analysis-based intelligent project screening and supervision system includes an intelligent quantitative recommendation platform module, a convergence iteration module, a weight adjustment module, a conflict detection module, an enterprise matching module, and a decision reporting module. The intelligent quantitative recommendation platform module is used to deploy the intelligent quantitative recommendation platform in enterprises. When screening projects, enterprises input all departmental demand data and project parameters of all projects into the intelligent quantitative recommendation platform for storage; the intelligent quantitative recommendation platform performs quantitative calculations on all data and outputs basic parameters. The convergence iteration module is used to iteratively adjust the basic matching degree and subjective coefficient for each demand node based on the initial matching state of the other demand nodes, and then perform a convergence judgment. When convergence is judged, the enterprise matching degree is calculated; if convergence is not judged, iterative adjustment continues. The weight adjustment module is used to perform weight evolution once every R rounds during the iterative adjustment of demand nodes, and update the dynamic weight of each demand node. The conflict detection module is used to perform conflict detection for each iteration when the demand node is iteratively adjusted. When a conflict exists and the conflict lasts for more than a preset number of rounds, an arbitration mechanism is executed. The enterprise matching module is used to calculate the enterprise matching degree of each screening node by using the matching degree of each demand node and the final iterative value of the subjective coefficient after the iteration adjustment. The decision report module is used to sort all the filtering nodes according to the enterprise matching degree, mark the items for different arbitration results in the arbitration mechanism, and finally generate a decision report for output.

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1. This application iteratively adjusts the original evaluation values ​​of each node, enabling mutual calibration among nodes, resulting in a final convergence state where the matching degree is closer to the group consensus center. Simultaneously, a confidence decay mechanism automatically identifies and weakens the contributions of unstable nodes. Overall, under the same input conditions, the deviation between the enterprise matching degree output by this scheme and the ideal true value is significantly reduced.

[0020] 2. Existing distributed average consensus algorithms require a large number of iterations to converge when dealing with scenarios where node preference directions differ significantly in a multi-dimensional feature space, and may even experience oscillations. This application introduces the preference direction angle as an influence modulation factor, enabling nodes with similar directions to quickly aggregate and nodes with opposite directions to weaken each other. Essentially, this constructs a directed weighted consensus network, significantly reducing the number of iterations.

[0021] 3. The existing system in this application, upon detecting an evaluation conflict between nodes, either discards the data or directly triggers an alarm requiring manual intervention. This solution innovatively employs a three-level automatic escalation arbitration mechanism: global injection → local annealing → manual interface. In most conflict scenarios, the first two levels can resolve the conflict automatically without manual intervention, maintaining the system's automated operation; manual intervention is only triggered in extreme cases. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the steps of an intelligent monitoring method for project screening based on data analysis according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example: Figure 1 As shown, the present invention provides a technical solution. A data-driven intelligent monitoring method for project screening includes the following steps: S100. Deploy an intelligent quantitative recommendation platform in the enterprise. When selecting projects, the enterprise inputs all departmental requirement data and all project parameters into the intelligent quantitative recommendation platform for storage. The intelligent quantitative recommendation platform performs quantitative calculations on all data and outputs basic parameters. Furthermore, the specific steps for outputting the basic parameters are as follows: S101. During project screening, the enterprise inputs all departmental demand data and all project parameters into the intelligent quantitative recommendation platform for storage. Each department is used as a demand node and each project is used as a screening node. When the enterprise starts the project screening task, the intelligent quantitative recommendation platform receives the task instruction and performs quantitative calculations on all the data stored in the intelligent quantitative recommendation platform. The basic parameters include basic matching degree, subjective coefficient, preference direction angle, and dynamic weight; The intelligent quantitative recommendation platform uses demand data from demand nodes and project parameters from filtering nodes to quantify the coverage rate of each filtering node. The coverage rate is defined as the number of covered parameters divided by the total number of demand data items. The number of covered parameters represents the number of project parameters whose values ​​are greater than the demand data values. By uniformly inputting departmental demand data and project parameters from different sources and in different formats into the intelligent quantitative recommendation platform, and using quantitative indicators such as coverage and completeness, unstructured data is transformed into computable numerical vectors, thus solving the technical problem of the inability to directly integrate multi-source heterogeneous data.

[0025] Calculate the parameter completeness of each filter node. The parameter completeness = number of valid fields ÷ total number of information fields. The number of valid fields represents the number of fields in the project parameters input to the project, excluding missing and outlier values ​​in the information fields themselves. Outlier values ​​represent field data that exceed the average value by 3 times. Weights were assigned to coverage and parameter completeness, and the weighted sum was used to calculate the department's basic matching degree with the project; the weights were set by the department staff. S102. Set a subjective comparison slider in the intelligent quantitative recommendation platform. Place two types of filtering nodes at both ends of the subjective comparison slider. The enterprise staff drag the slider according to their subjective intentions. Set the middle value of the subjective comparison slider to 0 and the two ends to 1; 0 means completely uncertain and 1 means very certain. Calculate the subjective coefficient of each filtering node based on the subjective dragging results of the staff. Repeat the calculation for all types of items. S103. Calculate the difference between all project parameters of different filtering nodes, extract the two project parameters with the largest difference in the filtering nodes as angle features, construct a plane rectangular coordinate system based on the angle features, extract the feature points of each project in the plane rectangular coordinate system, and extract the angle between the feature points and the x-axis in the plane rectangular coordinate system as the project type angle. The angle value range is 0 to 2π. Extract the historical filter item set of the demand node, calculate the two average angle feature values ​​of the historical filter items of the demand node, input them into the Cartesian coordinate system to obtain the demand feature point, and extract the angle between the demand feature point and the x-axis in the Cartesian coordinate system as the preference center angle. The angle value ranges from 0 to 2π. In each demand node, the preference direction angle is calculated using the item type angle and the preference center angle, where the preference direction angle = (item type angle - preference center angle) mod 2π. An initial matching state is constructed using the basic matching degree, subjective coefficient, and preference direction angle. A two-dimensional coordinate system is constructed by extracting the item parameters with the largest feature difference, compressing the high-dimensional item features into a single angle value, so that the preference difference between nodes can be measured by mathematical distance, providing a calculable geometric basis for subsequent conflict detection.

[0026] S104. Extract the demand data volume and demand time of each demand node in the intelligent quantitative recommendation platform. The demand data volume represents the total amount of demand data information input by the demand node, and the demand time represents the duration of each type of demand data input by each demand node. After normalizing the demand data volume and demand time of each department, a weighted sum is obtained to obtain a dynamic weight. Enterprise staff can customize the parameters used in the system, including learning rate, coupling coefficient, angle penalty factor, confidence decay coefficient, convergence matching degree threshold, and convergence subjective coefficient threshold.

[0027] S200. For each demand node, the basic matching degree and subjective coefficient are iteratively adjusted based on the initial matching state of the other demand nodes. Then, a convergence judgment is made. When convergence is determined, the enterprise matching degree is calculated. If convergence is not achieved, iterative adjustment continues. Furthermore, the specific steps for iteratively adjusting the basic matching degree and subjective coefficient for each demand node based on the initial matching state of the other demand nodes are as follows: S201. For each demand node, its own basic matching degree is iteratively adjusted using the demand node's own preference direction angle in the initial matching state, the preference direction angles of other demand nodes, the basic matching degree, and the dynamic weight. The formula is as follows: ; In the formula, s i，p （t+1） s represents the matching degree between the demand node i and the selection node p at iteration t+1. i，p （t） s represents the basic matching degree of demand node i to filter node p before adjustment at time t. j，p （t） θ represents the basic matching degree of the remaining demand nodes j to the filter node p at time t before adjustment. i，p θ represents the directional angle of preference of demand node i for selection node p. j，p w represents the preference angle of the remaining demand node j towards the selection node p. j （t） d(θ) represents the dynamic weight of the remaining demand node j before adjustment t, β represents the coupling coefficient, γ represents the angle penalty factor, the sigmoid function makes the result between (0, 1), and d(θ) i，p θ j，p The angular frequency difference represents the preference direction of demand node i and department j towards screening node p. The matching degree of each demand node is no longer fixed, but is iteratively updated based on the matching degree of other nodes, the preference direction angle, and dynamic weights. This solves the fundamental defect of traditional weighted fusion methods that only change the weights without changing the original values ​​and cannot eliminate evaluation discrepancies between nodes.

[0028] For the same filtering node p, the specific differences in the preference direction angles between demand nodes i and j are as follows: ; In the iteration formula, the mutual influence strength between nodes is modulated by the difference in preference direction angles. Nodes with similar directions have great influence, while nodes with opposite directions have little or even negative influence, which is equivalent to automatically identifying node groups with "similar opinions" and "opposite opinions" during the fusion process, and avoids fusion distortion caused by blind averaging.

[0029] S202. Adjust the subjective coefficient by using the basic matching degrees and matching degrees of the demand node i before and after iterative adjustment, where the formula is as follows: ; In the formula, c i，p （t+1） represents the adjusted subjective coefficient of the demand node i for the screening node p at time t+1, c i，p （t） represents the subjective coefficient of the demand node i for the screening node p at time t before adjustment, and λ represents the confidence attenuation coefficient; if the calculation result c i，p （t+1） <0, then c i，p （t+1） is set to 0.01; S203. After adjusting the matching degrees and subjective coefficients of all demand nodes in each round, calculate the differences of the matching degrees and subjective coefficients of all demand nodes before and after the adjustment respectively, and extract the maximum value △s of the matching degree differences of all demand nodes max ; when △s max <St and all the differences of the subjective coefficients are less than Ct, it is determined that convergence is achieved and enterprise matching degree calculation is executed; otherwise, iterative adjustment of the matching degrees and subjective coefficients is continued, where St represents the convergence matching degree threshold, and Ct represents the convergence subjective coefficient threshold. The maximum change of the matching degrees and all changes of the subjective coefficients are monitored at the same time. Convergence is determined only when both are lower than the preset thresholds. The dual-condition convergence mechanism is stricter than a single threshold, which ensures that the system reaches a real stable state rather than local oscillation, and avoids early stopping or infinite iteration.

[0030] S300. During the iterative adjustment of the demand nodes, weight evolution is executed every R rounds to update the dynamic weight of each demand node; Further, the specific steps of updating the dynamic weight of each demand node are as follows: S301. When the iterative adjustment of the demand nodes is executed for R=5 rounds, weight evolution is executed once. The weight evolution refers to calculating the dynamic weight variation for each demand node by using the matching degree and the preference direction angle of the latest round, where the formula is as follows: ; In the formula, δw i represents the dynamic weight variation of the demand node i, s i，p represents the matching degree of the demand node i for the screening node p during the latest round of adjustment, sj，p α represents the matching degree of the remaining demand node j to the screening node p in the most recent round of adjustment, N represents the total number of demand nodes in the intelligent quantitative recommendation platform, and α represents the learning rate. S302. For each demand node, update the dynamic weight using the dynamic weight change amount. The formula is: w i new =w i old +δw i ;In the formula, w i new w represents the dynamic weight of the demand node i after the update. i old This represents the dynamic weight of the demand node i before the update; if w i new If the value is less than 0, the value is copied to 0.01; after the update, the sum of the dynamic weights of all demand nodes is 1. The dynamic weights of demand nodes are evolved and updated in fixed rounds, and the weight change is calculated based on the matching degree and preference direction angle of the latest round. This allows the dynamic weights of demand nodes to be optimized in real time as the filtering nodes iterate, no longer limited to the initial set value, and always adapted to the actual situation of the current filtering node, correcting the problem of mismatch between the initial weights and the actual demand. S400: When adjusting the demand node iteratively, perform conflict detection for each iteration. If a conflict exists and the conflict continues for more than a preset number of iterations, execute the arbitration mechanism. Furthermore, the specific steps for implementing the arbitration mechanism when a conflict exists and the number of rounds of conflict exceeds a preset round threshold are as follows: S401, The preset round number threshold is T conflict When iteratively adjusting all demand nodes, conflict detection is performed after each round of iterative adjustment. The conflict detection is determined using the preference direction angle and the difference in matching degree. Specifically, when d(θ) i，p θ j，p ) > θ angle And |s i，p -s j，p When |>△s, it is determined that there is a conflict between demand node i and demand node j regarding the selection node p, θ angle Δs represents the preference orientation angle conflict threshold, and Δs represents the matching degree conflict threshold. When a conflict is detected and the duration exceeds T rounds... conflict When a conflict occurs, an arbitration mechanism is executed; the arbitration mechanism includes global injection adjustment, local annealing, and manual intervention; arbitration is only triggered when the conflict state continues for more than a certain number of rounds, thus avoiding unnecessary arbitration intervention caused by instantaneous disturbances and improving the system's noise resistance.

[0031] The global adjustment includes constructing a virtual enterprise global perspective demand node A. The preference direction angle of the enterprise global perspective demand node is set to the average of the preference direction angles of all demand nodes. The dynamic weight is normalized to 1.2 times the maximum value of the dynamic weights of all demand nodes. The staff uses the proportion of the same field in the total fields of all demand data and project parameters in the intelligent quantitative recommendation platform as the basic matching degree of the enterprise global perspective demand node. The subjective coefficient is the maximum value among all demand nodes. The enterprise global perspective department A is added to all demand nodes in the intelligent quantitative recommendation platform and iteratively adjusted. Arbitration is considered complete when the conflict disappears within 10 rounds. If the conflict does not disappear within 10 rounds, local annealing is performed. The local annealing process involves halving the coupling coefficient and doubling the subjective coefficient for two conflicting requirement nodes. After iterative adjustment, arbitration is considered complete when the conflict disappears within 5 rounds; otherwise, manual intervention is performed. The manual intervention includes direct intervention by enterprise staff at conflicting demand nodes. The staff sets the preference direction angle and matching degree of conflicting demand nodes i and j to the median value of the two, then eliminates the conflict and iterates and adjusts again.

[0032] The three-tiered arbitration mechanism follows the principle of prioritizing automation and providing a backup of human intervention. Global injection adjustments resolve routine conflicts automatically through virtual global demand nodes, allowing for rapid recovery and iteration without human intervention. Local annealing specifically weakens the coupling relationship between conflict demand nodes, efficiently resolving moderate disagreements. Human intervention serves as the final safeguard, handling extreme disputes and ensuring uninterrupted screening throughout the process. This achieves a perfect balance between intelligent automation and human flexibility, ensuring that all conflicts are resolved reasonably.

[0033] S500: Calculate the enterprise matching degree of each screening node using the matching degree of each demand node and the final iterative value of the subjective coefficient after the iteration adjustment. Furthermore, the specific steps for calculating the enterprise matching degree for each filtering node are as follows: S501. After the demand node iterative adjustment converges, the enterprise matching degree is calculated using the converged matching degree and the final iterative value of the subjective coefficient. The formula is as follows: ; In the formula, s i，p （final） c represents the final iterative value of the matching degree between demand node i and filter node p. i，p （final） E represents the final iterative value of the subjective coefficient of demand node i relative to selection node p. p This represents the matching degree of enterprises to the filter node p; it outputs the matching degree of enterprises to each filter node and the number of convergence rounds.

[0034] S600: Sort all filtering nodes according to enterprise matching degree, mark items for different arbitration results in the arbitration mechanism, and finally generate a decision report for output.

[0035] Furthermore, the specific steps for generating and outputting the decision report are as follows: S601. Sort all screening nodes in descending order of enterprise matching degree. If there are projects that eventually converge through global injection adjustment and local annealing in the arbitration mechanism, mark the corresponding screening node as "low consensus screening node". If there are screening nodes that eventually converge through manual intervention in the arbitration mechanism, mark the corresponding screening node as "disputed screening node". Construct a sorting table for filtering nodes and a conflict resolution log. The sorting table for filtering nodes includes enterprise matching degree and number of convergence rounds. The conflict resolution log includes conflict requirement nodes and arbitration methods. The top K uncontroversial screening nodes in the screening node ranking table are selected for recommendation. Among the top K uncontroversial screening nodes, risk warnings are issued for screening nodes with low consensus. K is set by the staff. The screening node ranking table, conflict resolution log, and recommended screening nodes constitute the output of a decision report.

[0036] A data analysis-based intelligent project screening and supervision system includes an intelligent quantitative recommendation platform module, a convergence iteration module, a weight adjustment module, a conflict detection module, an enterprise matching module, and a decision reporting module. The intelligent quantitative recommendation platform module is used to deploy the intelligent quantitative recommendation platform in enterprises. When screening projects, enterprises input all departmental demand data and project parameters of all projects into the intelligent quantitative recommendation platform for storage; the intelligent quantitative recommendation platform performs quantitative calculations on all data and outputs basic parameters. The convergence iteration module is used to iteratively adjust the basic matching degree and subjective coefficient for each demand node based on the initial matching state of the other demand nodes, and then perform a convergence judgment. When convergence is judged, the enterprise matching degree is calculated; if convergence is not judged, iterative adjustment continues. The weight adjustment module is used to perform weight evolution once every R rounds during the iterative adjustment of demand nodes, and update the dynamic weight of each demand node. The conflict detection module is used to perform conflict detection for each iteration when the demand node is iteratively adjusted. When a conflict exists and the conflict lasts for more than a preset number of rounds, an arbitration mechanism is executed. The enterprise matching module is used to calculate the enterprise matching degree of each screening node by using the matching degree of each demand node and the final iterative value of the subjective coefficient after the iteration adjustment. The decision report module is used to sort all the filtering nodes according to the enterprise matching degree, mark the items for different arbitration results in the arbitration mechanism, and finally generate a decision report for output. Example:

[0037] A company deployed an intelligent quantitative recommendation platform, setting the R&D department, marketing department, and operations department as demand nodes, and three candidate projects as screening nodes P1, P2, and P3.

[0038] The R&D department requires 8 data items, and P1 meets 6 of them, with a coverage rate of 6 / 8 = 0.75. P1 has 12 parameters, with 1 missing, resulting in a parameter completeness of 11 / 12 ≈ 0.92. Assigning a weight of 0.6 to coverage and 0.4 to completeness, we get the basic matching degree as 0.75 × 0.6 + 0.92 × 0.4 = 0.818.

[0039] Staff members slide the subjective comparison slider on the platform. The certainty of P1 is 0.9, therefore the subjective coefficient c=0.9. A coordinate system is constructed using "input cost" and "benefit cycle" as angular features; the project type angle of P1 is π / 3; the historical preference center angle of the R&D department is π / 4; the preference direction angle = (π / 3 − π / 4) mod 2π = π / 12, where mod keeps the calculation result between 0 and 2π; π=3.14 is substituted into the iterative adjustment. The dynamic weights for the three departments are calculated as follows: R&D Department 0.4, Marketing Department 0.35, and Operations Department 0.25.

[0040] Taking the R&D department's selection of node P1 as an example: The matching degree is adjusted according to the iterative formula: the matching degree s=0.818 in the t-th round. After the preference direction angle and dynamic weight coupling adjustment by the marketing department and the operations department, the matching degree in the t+1 round is updated to 0.832.

[0041] Adjust the subjective coefficient: Calculated based on the attenuation coefficient λ, the subjective coefficient is updated from 0.9 to 0.89.

[0042] Convergence criterion: The maximum difference Δs between the matching degrees of all demand nodes. max =0.02<St=0.03, all subjective coefficient differences<Ct=0.05, the system determines that the iteration has converged, and proceeds to calculate the enterprise matching degree.

[0043] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A data analysis-based intelligent supervision method for project screening, characterized in that: The method includes the following steps: S100. Deploy an intelligent quantitative recommendation platform in the enterprise. When selecting projects, the enterprise inputs all departmental requirement data and all project parameters into the intelligent quantitative recommendation platform for storage. The intelligent quantitative recommendation platform performs quantitative calculations on all data and outputs basic parameters. S200. For each demand node, the basic matching degree and subjective coefficient are iteratively adjusted based on the initial matching state of the other demand nodes. Then, a convergence judgment is made. When convergence is determined, the enterprise matching degree is calculated. If convergence is not achieved, iterative adjustment continues. S300. During the iterative adjustment of demand nodes, a weight evolution is performed once every R rounds to update the dynamic weight of each demand node. S400: When adjusting the demand node iteratively, perform conflict detection for each iteration. If a conflict exists and the conflict continues for more than a preset number of iterations, execute the arbitration mechanism. S500: Calculate the enterprise matching degree of each screening node using the matching degree of each demand node and the final iterative value of the subjective coefficient after the iteration adjustment. S600: Sort all filtering nodes according to enterprise matching degree, mark items for different arbitration results in the arbitration mechanism, and finally generate a decision report for output.

2. The intelligent supervision method for project screening based on data analysis according to claim 1, characterized in that: The basic parameters in S100 include basic matching degree, subjective coefficient, preference direction angle and dynamic weight; The intelligent quantitative recommendation platform uses demand data for demand nodes and project parameters for filtering nodes to quantify the coverage of each filtering node. The coverage is calculated as: coverage = number of coverage parameters ÷ total number of demand data. The number of coverage parameters refers to the number of project parameters whose parameter values ​​are greater than the required data values; Calculate the parameter completeness of each filter node. The parameter completeness = number of valid fields ÷ total number of information fields. The number of valid fields represents the number of fields in the project parameters input to the project, excluding missing and outlier values ​​in the information fields themselves. Outlier values ​​represent field data that exceed the average value by 3 times. Weights were assigned to coverage and parameter completeness, and the weighted sum was used to calculate the department's basic matching degree with the project; the weights were set by the department staff. The subjective coefficient includes setting a subjective comparison slider in the intelligent quantitative recommendation platform, placing two types of filtering nodes at both ends of the subjective comparison slider, and having enterprise staff drag the slider according to their subjective intentions. The middle value of the subjective comparison slider is set to 0, and the two ends are set to 1; 0 represents complete uncertainty, and 1 represents high certainty. The subjective coefficient of each filtering node is calculated based on the subjective dragging results of the staff. The calculation is repeated for all types of items. The preferred orientation angle includes calculating the difference between all project parameters of different filtering nodes, extracting the two project parameters with the largest difference in the filtering nodes as angle features, constructing a Cartesian coordinate system based on the angle features, extracting the feature points of each project in the Cartesian coordinate system, and extracting the angle between the feature points and the x-axis in the Cartesian coordinate system as the project type angle, with the angle value ranging from 0 to 2π. Extract the historical filter item set of the demand node, calculate the two average angle feature values ​​of the historical filter items of the demand node, input them into the Cartesian coordinate system to obtain the demand feature point, and extract the angle between the demand feature point and the x-axis in the Cartesian coordinate system as the preference center angle. The angle value ranges from 0 to 2π. In each demand node, the preference direction angle is calculated using the project type angle and the preference center angle, where the preference direction angle = (project type angle - preference center angle) mod 2π; the initial matching state is constructed using the basic matching degree, subjective coefficient and preference direction angle.

3. The intelligent supervision method for project screening based on data analysis according to claim 1, characterized in that: The specific steps for initializing the dynamic weights of each department in S100 are as follows: Extract the demand data volume and demand time of each demand node in the intelligent quantitative recommendation platform. The demand data volume represents the total amount of demand data information input by the demand node, and the demand time represents the duration of each type of demand data input by each demand node. The dynamic weight is obtained by normalizing the amount of data and the time required for each department's needs and then summing them by weight.

4. The intelligent supervision method for project screening based on data analysis according to claim 1, characterized in that: The convergence determination in S200 specifically involves: For each demand node, its own basic matching degree is iteratively adjusted using the demand node's own preference direction angle in the initial matching state, the preference direction angles of other demand nodes, the basic matching degree, and the dynamic weight. The formula is as follows: ； In the formula, s i，p （t+1） s represents the matching degree between the demand node i and the selection node p at iteration t+1. i，p （t） s represents the basic matching degree of demand node i to filter node p before adjustment at time t. j，p （t） θ represents the basic matching degree of the remaining demand nodes j to the filter node p at time t before adjustment. i，p θ represents the directional angle of preference of demand node i for selection node p. j，p w represents the preference angle of the remaining demand node j towards the selection node p. j （t） d(θ) represents the dynamic weight of the remaining demand node j before adjustment t, β represents the coupling coefficient, γ represents the angle penalty factor, the sigmoid function makes the result between (0, 1), and d(θ) i，p θ j，p The difference in the angular frequency of the preference direction between demand node i and department j for screening node p is represented by ). For the same filtering node p, the specific differences in the preference direction angles between demand nodes i and j are as follows: ； S202. Adjust the subjective coefficient using the basic matching degree and matching degree of demand node i before and after iterative adjustment, using the following formula: ； In the formula, c i，p （t+1） c represents the subjective coefficient of demand node i relative to selection node p at time t+1 after adjustment. i，p （t） This represents the subjective coefficient of demand node i towards selection node p at time t before adjustment, and λ represents the confidence decay coefficient; if the calculation result c i，p （t+1） <0, then c i，p （t+1） Set to 0.01; S203. After adjusting the matching degree and the subjective coefficient of all demand nodes in each round, respectively calculate the difference values of the matching degree and the subjective coefficient of all demand nodes before and after the adjustment, and extract the maximum value Δs of the matching degree difference values of all demand nodes max ; when Δs max <St and all difference values of the subjective coefficients are less than Ct, it is determined that convergence is achieved and enterprise matching degree calculation is performed; otherwise, iterative adjustment of the matching degree and the subjective coefficient is continued, where St represents the convergence matching degree threshold, and Ct represents the convergence subjective coefficient threshold.

5. The intelligent supervision method for project screening based on data analysis according to claim 1, characterized in that: The specific steps for updating the dynamic weight of each demand node in S300 are as follows: When the demand node iterative adjustment is executed for R=5 rounds, a weight evolution is performed once. The weight evolution means that for each demand node, the dynamic weight change is calculated using the matching degree and preference direction angle of the most recent round. For each demand node, the dynamic weight is updated using the dynamic weight change, with the formula: w i new =w i old +δw i ;In the formula, w i new w represents the dynamic weight of the demand node i after the update. i old This represents the dynamic weight of the demand node i before the update; If w i new If the value is less than 0, the value is copied to 0.01; after the update, the sum of the dynamic weights of all demand nodes is 1.

6. The intelligent supervision method for project screening based on data analysis according to claim 4, characterized in that: The collision detection in S400 specifically includes: The preset round threshold is T conflict When iteratively adjusting all demand nodes, conflict detection is performed after each round of iterative adjustment. The conflict detection is determined using the preference direction angle and the difference in matching degree. Specifically, when d(θ) i，p θ j，p ) > θ angle And |s i，p -s j，p When |>△s, it is determined that there is a conflict between demand node i and demand node j regarding the selection node p, θ angle Δs represents the preference orientation angle conflict threshold, and Δs represents the matching degree conflict threshold. When a conflict is detected and the duration exceeds T rounds... conflict In such cases, an arbitration mechanism shall be implemented.

7. The intelligent supervision method for project screening based on data analysis according to claim 1, characterized in that: The arbitration mechanism in S400 includes global injection adjustment, local annealing, and manual intervention; The global adjustment includes constructing a virtual enterprise global perspective demand node A. The preference direction angle of the enterprise global perspective demand node is set to the average of the preference direction angles of all demand nodes. The dynamic weight is normalized to 1.2 times the maximum value of the dynamic weights of all demand nodes. The staff uses the proportion of the same field in the total fields of all demand data and project parameters in the intelligent quantitative recommendation platform as the basic matching degree of the enterprise global perspective demand node. The subjective coefficient is the maximum value among all demand nodes. The enterprise global perspective department A is added to all demand nodes in the intelligent quantitative recommendation platform and iteratively adjusted. Arbitration is considered complete when the conflict disappears within 10 rounds. If the conflict does not disappear within 10 rounds, local annealing is performed. The local annealing process involves halving the coupling coefficient and doubling the subjective coefficient for two conflicting requirement nodes. After iterative adjustment, arbitration is considered complete when the conflict disappears within 5 rounds; otherwise, manual intervention is performed. The manual intervention includes direct intervention by enterprise staff at conflicting demand nodes. The staff sets the preference direction angle and matching degree of conflicting demand nodes i and j to the median value of the two, then eliminates the conflict and iterates and adjusts again.

8. The intelligent supervision method for project screening based on data analysis according to claim 1, characterized in that: The specific steps for calculating the enterprise matching degree of each screening node in S500 are as follows: Once the iterative adjustment of the demand node has converged, the enterprise matching degree is calculated using the converged matching degree and the final iterative value of the subjective coefficient. The formula is as follows: ； In the formula, s i，p （final） c represents the final iterative value of the matching degree between demand node i and filter node p. i，p （final） E represents the final iterative value of the subjective coefficient of demand node i relative to selection node p. p This represents the matching degree of enterprises to the filter node p; it outputs the matching degree of enterprises to each filter node and the number of convergence rounds.

9. The intelligent supervision method for project screening based on data analysis according to claim 1, characterized in that: The decision report in S600 specifically includes: All screening nodes are sorted in descending order of enterprise matching degree. If there are projects that eventually converge through global injection adjustment and local annealing in the arbitration mechanism, the corresponding screening node is marked as a "low consensus screening node". If there are screening nodes that eventually converge through manual intervention in the arbitration mechanism, the corresponding screening node is marked as a "disputed screening node". Construct a sorting table for filtering nodes and a conflict resolution log. The sorting table for filtering nodes includes enterprise matching degree and number of convergence rounds. The conflict resolution log includes conflict requirement nodes and arbitration methods. The top K uncontroversial screening nodes in the screening node ranking table are selected for recommendation. Among the top K uncontroversial screening nodes, risk warnings are issued for screening nodes with low consensus. K is set by the staff. The screening node ranking table, conflict resolution log, and recommended screening nodes constitute the output of a decision report.

10. A data analysis-based intelligent monitoring system for project screening, characterized in that: The intelligent monitoring system for project screening includes an intelligent quantitative recommendation platform module, a convergence iteration module, a weight adjustment module, a conflict detection module, an enterprise matching module, and a decision reporting module. The intelligent quantitative recommendation platform module is used to deploy the intelligent quantitative recommendation platform in enterprises. When screening projects, enterprises input all departmental demand data and project parameters of all projects into the intelligent quantitative recommendation platform for storage. The intelligent quantitative recommendation platform performs quantitative calculations on all data and outputs basic parameters; The convergence iteration module is used to iteratively adjust the basic matching degree and subjective coefficient for each demand node based on the initial matching state of the other demand nodes, and then perform a convergence judgment. When convergence is judged, the enterprise matching degree is calculated; if convergence is not achieved, iterative adjustment continues. The weight adjustment module is used to perform weight evolution once every R rounds during the iterative adjustment of demand nodes, and update the dynamic weight of each demand node. The conflict detection module is used to perform conflict detection for each iteration when the demand node is iteratively adjusted. When a conflict exists and the conflict lasts for more than a preset number of rounds, an arbitration mechanism is executed. The enterprise matching module is used to calculate the enterprise matching degree of each screening node by using the matching degree of each demand node and the final iterative value of the subjective coefficient after the iteration adjustment. The decision report module is used to sort all the filtering nodes according to the enterprise matching degree, mark the items for different arbitration results in the arbitration mechanism, and finally generate a decision report for output.

Images