Market supervision data asset multi-dimensional evaluation method and system based on hierarchy-network dynamic fusion model

Abstract

The invention discloses a market supervision data asset multi-dimensional evaluation method and system based on a hierarchy-network dynamic fusion model, and relates to the technical field of data asset evaluation. According to the method, a semantic association and index alignment relationship between data assets is comprehensively described by constructing a semantic map of the data assets; a hierarchy-network dynamic fusion model is formed by combining an analytic hierarchy process and a network analysis process, so that the determination of the index weight is more scientific and reasonable; a graph neural network is utilized to perform embedded representation learning on a data asset graph structure, comprehensive feature expression of each data asset is extracted, a multi-objective optimization method is combined, a Pareto optimal solution set is generated, scoring, sorting and grading are performed on the data assets, and reliability and interpretability of an evaluation result are ensured; the evaluation report is output in a structured data format and is provided for a market supervision system through an interface, thereby facilitating effective management and decision support for data assets by a supervision department, and having a wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of data asset evaluation, and in particular, to a multi-dimensional evaluation method and system for market supervision data assets based on a hierarchical-network dynamic fusion model. Background Art

[0002] With the explosive growth of data resources, data, as a new type of production factor, is playing an increasingly important role in the digital economy and intelligent supervision. Especially in the field of market supervision, regulatory authorities have a large amount of supervision data assets scattered in different systems, with heterogeneous formats and inconsistent structures, such as enterprise credit information, law enforcement records, transaction behavior data, etc.

[0003] However, the current technical means for data asset value evaluation are still relatively lagging, and there are mainly the following problems:

[0004] Lack of a unified data description and modeling method, resulting in difficulty in fusing data from different systems and lack of a unified caliber for evaluation;

[0005] The evaluation model structure is single, mostly using a static hierarchical index scoring method, ignoring the dependency relationship and interaction impact between indicators;

[0006] Lack of the ability to model the structural relationship between data, making it difficult to reveal the potential associations between data assets from an overall perspective;

[0007] The evaluation result dimension is limited, the weight setting is highly subjective, and there is a lack of optimization and balance for multi-objective and multi-dimensional indicators.

[0008] Therefore, there is an urgent need for a comprehensive evaluation method to achieve scientific, multi-angle analysis and intelligent evaluation of the value of market supervision data assets. Summary of the Invention

[0009] Aiming at the problems of the existing evaluation model with a single dimension, inability to dynamically adjust, and ignoring the complex network relationship between indicators, the present invention proposes a multi-dimensional evaluation method and system for market supervision data assets based on a hierarchical-network dynamic fusion model.

[0010] The present invention achieves the above object through the following technical solutions:

[0011] A multi-dimensional evaluation method for market supervision data assets based on a hierarchical-network dynamic fusion model, the method comprising:

[0012] Collect multi-source data assets from different market supervision business systems, and record meta-attribute information to construct an original data asset set;

[0013] Perform data cleaning, field unification, and type standardization on the original data asset collection, and establish a semantic metadata model using the Resource Description Framework. Map the attributes of each data asset to the dimension hierarchy or metric nodes in the knowledge graph to form a data asset semantic graph, which is used to depict the semantic associations and metric alignment relationships between data assets;

[0014] Construct a three-layer metric system based on the data asset semantic graph. Use the Analytic Hierarchy Process to calculate the structural weights of each layer of metrics, and use the Network Analysis Method to construct a metric dependence network to calculate the propagation influence weights. Integrate the structural weights and the propagation influence weights to form a hierarchical-network dynamic fusion model;

[0015] Based on the hierarchical-network dynamic fusion model and combined with the data asset semantic graph, construct a data asset graph structure. Use graph neural networks to perform embedded representation learning on the data asset graph structure, and extract the embedded vectors output by each data asset at the last layer of the graph neural network;

[0016] According to the embedded vectors and combined with the hierarchical-network dynamic fusion model, determine the fusion weights, construct a multi-objective optimization function, use the non-dominated sorting genetic algorithm to solve the multi-objective optimization function, generate a Pareto optimal solution set, score, rank, and classify each data asset, and form an evaluation report.

[0017] As a preferred embodiment of the present invention, collect multi-source data assets from different market supervision business systems, including structured, semi-structured, and unstructured data, and further include data source authentication, structure parsing, file batch import, and field mapping rule execution. Each collection task is attached with a unique task number for evaluation and traceability; the meta-attribute information includes at least data source, field, timestamp, institutional affiliation, and update frequency.

[0018] As a preferred embodiment of the present invention, the semantic metadata model is constructed based on the Resource Description Framework semantic namespace. Each data asset instance is modeled as a dl:Source type node, each attribute field in each data asset is modeled as a dl:Domain type node, and each attribute field is mapped to the kpi:Indicator or kpi:Level concept node in the knowledge graph through dl:mapTo;

[0019] The semantic metadata model records field type, null value ratio, and numerical distribution statistical information.

[0020] As a preferred embodiment of the present invention, the three-layer metric system consists of the following:

[0021] Target layer: Defined as "the comprehensive value of data assets in the market supervision system", which serves as the top-level node of the hierarchical analysis;

[0022] Criterion layer: Includes five dimensions: data quality, data benefits, data risks, data availability, and data stability;

[0023] Index layer: Each dimension of the criterion layer is broken down into at least two quantifiable index fields, and the index fields are from the bound attributes in the data asset semantic graph.

[0024] As a preferred solution of the present invention, the structural weights of each layer of indicators are calculated using the analytic hierarchy process, and the method includes:

[0025] Construct judgment matrices between the criterion layer and the target layer, and between the index layer and the criterion layer. Using the 1–9 scale method, pairwise comparison scores are made by at least five domain experts with supervision experience;

[0026] Use the eigenvalue method to calculate the maximum eigenvalue λ of the judgment matrix max , and obtain the corresponding weight eigenvector ω as the local weight of each node;

[0027] Conduct a consistency test on each judgment matrix and calculate the consistency ratio where CI is the consistency index and RI is the random consistency index. When the consistency ratio CR < 0.1, the consistency of the judgment matrix is acceptable. When the consistency ratio CR ≥ 0.1, reconstruct the judgment matrix;

[0028] Normalize and combine the local weights that pass the consistency test according to the hierarchical structure to obtain the structural weight w of each evaluation index AHP,i .

[0029] As a preferred solution of the present invention, the method of using network analysis to construct an index dependence network to calculate the propagation influence weight includes:

[0030] Take all the evaluation index nodes determined by the analytic hierarchy process as the node set I = {i1, i2,..., i n} of the index dependence network, where n is the total number of nodes;

[0031] Based on the business process logic, index causal chain relationship, and statistical correlation reflected in the data asset semantic graph, construct a set of directed dependence edges R = {r ij} between the indexes. Each edge r ij indicates that index i has an impact on index j, and the edge weight w ij ∈[0, 1];

[0032] Construct an adjacency matrix W ∈ R according to the edge weights n×n, the element W in the adjacency matrix ij = w ij , and the unconnected node pairs are assigned a value of 0;

[0033] Calculate the propagation influence matrix T = (I - W) -1 , which is used to quantify the indirect influence degree of each index on other indexes;

[0034] Normalize each row of the propagation influence matrix, and calculate the global propagation influence factor of the i-th index as the mean value of the corresponding row elements, denoted as the propagation influence weight w ANP,i ;

[0035] The fused hierarchical-network dynamic fusion model is expressed as:

[0036] W i * = α × w AHP,i + (1 - α) × w ANP,i ;

[0037] In the formula, W i * is the fusion weight of the i-th index; α ∈ [0, 1] is the fusion adjustment parameter.

[0038] As a preferred embodiment of the present invention, the method for constructing the data asset graph structure includes:

[0039] Take each data asset as a node in the graph, and the initial feature vector of the node consists of the numerical statistical features of the attribute fields of the data asset, the field semantic type encoding, and the index binding information;

[0040] Construct the edge set E = {e ij} between nodes according to the field similarity, graph connection path, and shared index information in the data asset semantic graph, and calculate the edge weight. The formula is:

[0041]

[0042] In the formula, x i , x j are the initial feature vectors of the nodes; τ is the bandwidth parameter;

[0043] Construct the adjacency matrix A and add self-loop terms to obtain Then calculate the normalized adjacency matrix where is 's degree matrix;

[0044] Use the graph neural convolutional network to perform embedding representation learning on the data asset graph structure. The node representation update process is:

[0045]

[0046] In the formula, H (l) represents the node feature representation of the l-th layer, and H (l+1) represents the node feature representation of the (l + 1)-th layer; W (l) is the trainable weight matrix of the l-th layer, and σ is the activation function;

[0047] The node embedding matrix H(L) of the final layer of the graph neural network is obtained, where the i-th row vector h i represents the comprehensive feature expression of the i-th data asset under the data asset graph structure and the index system.

[0048] As a preferred solution of the present invention, the method for constructing the multi-objective optimization function includes:

[0049] Taking the embedding vector h of each data asset i as the input of the feature expression of the data asset;

[0050] Combining the fusion weight vector determined in the hierarchical-network dynamic fusion model to construct the multi-objective optimization function F(x i ), and the expression is:

[0051] F(x i ) = [f1(x i ), f2(x i ),..., f m (x i )];

[0052] In the formula, m is the number of evaluation index dimensions in the multi-objective optimization;

[0053] Any objective function in the multi-objective optimization function F(x i ) is defined as:

[0054] f k (x i ) represents the weighted score of the i-th data asset under the k-th index dimension; is the k-th fusion vector; is the k-th dimensional component in the embedding vector.

[0055] As a preferred solution of the present invention, the evaluation report includes: the comprehensive score of the data asset, the single score of each evaluation index dimension, the Pareto rank to which it belongs, and application suggestions, including utilization priorities, risk warnings, or data governance strategy suggestions;

[0056] Output the evaluation in a structured data format and provide it to the market supervision system through an interface for integrated application.

[0057] A multi-dimensional evaluation system for market supervision data assets based on a hierarchical-network dynamic fusion model is applied to the multi-dimensional method for market supervision data assets based on the hierarchical-network dynamic fusion model as described above. The system includes:

[0058] A data collection module for collecting multi-source data assets from multiple market supervision business systems, including structured, semi-structured, and unstructured data, and recording meta-attribute information for the data assets to form a set of original data assets;

[0059] A semantic modeling module for performing data cleaning, field unification, and type standardization processing on the set of original data assets, and constructing a semantic metadata model based on the Resource Description Framework, mapping the attribute fields of each data asset to the dimension levels or index nodes in the knowledge graph, and generating a semantic graph of data assets for depicting the semantic associations and index alignment relationships between data assets;

[0060] An index modeling and fusion module for constructing a three-layer evaluation index system based on the semantic graph of data assets, calculating the structural weights of each layer of indexes using the Analytic Hierarchy Process, and constructing an index dependence network based on the semantic dependence relationships between the indexes, calculating the propagation influence weights through network analysis, and fusing the structural weights and the propagation influence weights to form a hierarchical-network dynamic fusion model;

[0061] An asset graph construction and embedding module for constructing a data asset graph structure based on the hierarchical-network dynamic fusion model and the semantic graph of data assets, modeling each data asset as a node in the graph, and setting edge weights according to field similarity and graph connection degree; performing embedding representation learning on the data asset graph structure using a graph neural network, and outputting the embedding vector of each data asset at the last layer of the graph neural network;

[0062] An optimization scoring and result generation module for constructing a multi-objective optimization function according to the embedding vectors and combining the fusion weights determined by the hierarchical-network dynamic fusion model, solving the function using the non-dominated sorting genetic algorithm, generating a Pareto optimal solution set, comprehensively scoring, calculating index dimension scores, sorting, and grading each data asset, and finally generating an evaluation report;

[0063] A result output module for outputting the evaluation report in a structured data format and integrating it with an external market supervision platform through a system interface for regulatory decision-making support and data governance optimization.

[0064] The beneficial effects of the present invention are as follows: By constructing a semantic graph of data assets, comprehensively depicting the semantic associations and index alignment relationships among data assets, realizing the all-round and multi-dimensional evaluation of market supervision data assets, and improving the comprehensiveness and accuracy of evaluation results; adopting a combination of the analytic hierarchy process and the network analytic method to calculate the structural weights and propagation influence weights of each layer of indicators, forming a hierarchical-network dynamic fusion model, making the determination of indicator weights more scientific and reasonable, and reflecting the true dependence relationship among indicators; using a graph neural network to perform embedded representation learning on the graph structure of data assets, extracting the comprehensive feature expressions of each data asset, and combining with a multi-objective optimization method to generate a Pareto optimal solution set, scoring, ranking, and grading data assets to ensure the reliability and interpretability of evaluation results; the evaluation report is output in a structured data format and provided to the market supervision system through an interface to achieve seamless integration with the existing system, facilitating the regulatory department to effectively manage data assets and provide decision support, and having broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0066] Among them:

[0067] Figure 1 is the method flow chart of the present invention;

[0068] Figure 2 is the schematic diagram of the system modular structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention fall within the scope of protection of the present invention.

[0070] As Figure 1 shown, it is an embodiment of the present invention. This embodiment provides a multi-dimensional evaluation method for market supervision data assets based on a hierarchical-network dynamic fusion model, including the following steps:

[0071] S1: Collect multi-source data assets from different market supervision business systems and record the meta-attribute information to construct an original data asset set;

[0072] The original data assets include, but are not limited to: enterprise basic data (industrial and commercial registration, credit code), transaction behavior data (invoice information, abnormal transactions), risk data (supervision penalties, blacklists), log data (in and out records, event streams), etc. The data collection process ensures that the source identification, timestamp, and attribute field information of the data are completely retained for subsequent modeling and analysis.

[0073] In one embodiment, multi-source data assets from different market supervision business systems (government affairs data platform, credit information system, industry supervision system, etc.) are collected, including structured (such as CSV / database), semi-structured (JSON / XML), and unstructured (documents) data, and further include data source authentication, structure parsing, file batch import, and execution of field mapping rules. Each collection task is attached with a unique task number for evaluation and traceability. The meta-attribute information includes data source, field, timestamp, institutional affiliation, update frequency, etc.

[0074] S2: Perform data cleaning, field unification, and type standardization processing on the original data asset set, and establish a semantic metadata model using the Resource Description Framework (RDF). Map the attributes of each data asset to the dimension levels or metric nodes in the knowledge graph to form a data asset semantic graph for depicting the semantic associations and metric alignment relationships between data assets;

[0075] Specifically, the semantic metadata model is constructed based on the semantic naming space of the Resource Description Framework. Each data asset instance is modeled as a dl:Source type node, each attribute field in the data asset is modeled as a dl:Domain type node, and each attribute field is mapped to the kpi:Indicator or kpi:Level concept node in the knowledge graph through dl:mapTo;

[0076] The semantic metadata model records field types, null value ratios, and numerical distribution statistical information.

[0077] S3: Construct a three-layer metric system based on the data asset semantic graph, calculate the structural weights of each layer of metrics using the Analytic Hierarchy Process, and use the Network Analysis Method to construct a metric dependence network to calculate the propagation influence weights. Integrate the structural weights and propagation influence weights to form a hierarchical-network dynamic fusion model;

[0078] The three-layer metric system consists of the following:

[0079] Goal layer: Defined as "the comprehensive value of data assets in the market supervision system", which is the top-level node of the hierarchical analysis;

[0080] Criterion layer: It includes five dimensions: data quality, data benefit, data risk, data availability, and data stability;

[0081] Index layer: Each dimension of the criterion layer is decomposed into at least two quantifiable index fields, and the index fields come from the bound attributes in the semantic graph of data assets;

[0082] For example, the index fields corresponding to the data quality dimension include integrity, accuracy, consistency, etc., and "field missing rate" and "proportion of illegal values" can be set under them; the index fields corresponding to the data benefit dimension include output rate, utilization rate, etc., and the index fields corresponding to the data risk dimension include sensitivity, leakage risk, etc., and "historical anomaly rate" and "proportion of sensitive fields" can be set under them; the index fields corresponding to the data availability dimension include accessibility, circulation, etc., and the index fields corresponding to the data stability dimension include life cycle, change frequency, etc. These indexes all come from the statistical feature fields associated with the dl:Domain node in the semantic graph.

[0083] The analytic hierarchy process is used to calculate the structural weights of each layer of indexes. The methods include:

[0084] Construct judgment matrices between the criterion layer and the target layer, and between the index layer and the criterion layer. The 1–9 ratio scale method is adopted, and pairwise comparison scores are made by at least five domain experts with regulatory experience;

[0085] Use the eigenvalue method to calculate the maximum eigenvalue λ of the judgment matrix max and obtain the corresponding weight eigenvector ω as the local weight of each node;

[0086] Conduct a consistency test on each judgment matrix and calculate the consistency ratio where CI is the consistency index and RI is the random consistency index;

[0087] The consistency index CI reflects the degree of deviation between the currently constructed judgment matrix and the completely consistent matrix. The calculation formula is: When λ max = n, it means that the judgment matrix is completely consistent, and at this time CI = 0;

[0088] The random consistency index RI represents the expected value of the consistency of each order matrix under the condition of completely randomly constructing the judgment matrix, and can be obtained by Saaty's statistics of a large number of random matrices;

[0089] When the consistency ratio CR < 0.1, the consistency of the judgment matrix is acceptable. When the consistency ratio CR ≥ 0.1, reconstruct the judgment matrix;

[0090] Normalize and combine the local weights that pass the consistency test according to the hierarchical structure to obtain the structural weight w of each evaluation index AHP,i .

[0091] Use the network analysis method to construct an index dependence network to calculate the propagation influence weight. The method includes:

[0092] Take all the evaluation index nodes determined by the analytic hierarchy process as the node set of the index dependence network \(I = \{i_1, i_2, \ldots, i_n\}\), where \(n\) is the total number of nodes; n}

[0093] Construct a set of directed dependence edges \(R = \{r_{ij}\}\) between indexes based on the business process logic, index causal chain relationship, and statistical correlation reflected in the data asset semantic graph. Each edge \(r_{ij}\) indicates that index \(i\) has an impact on index \(j\), and the edge weight \(w_{ij} \in [0, 1]\); ij} ij where each edge \(r_{ij}\) ij ∈[0,1];

[0094] Construct an adjacency matrix \(W \in R^{n\times n}\) according to the edge weights. The element \(W_{ij}\) in the adjacency matrix is \(W_{ij} = w_{ij}\), and the unconnected node pairs are assigned 0; n×n where the adjacency matrix \(W\in R^{n\times n}\), ij and the element \(W_{ij}\) in the adjacency matrix ij = w_{ij},

[0095] Calculate the propagation influence matrix \(T=(I - W)^{-1}\), which is used to quantify the indirect influence degree of each index on other indexes; -1 where \((I - W)^{-1}\) is used to quantify the indirect influence degree of each index on other indexes;

[0096] Normalize each row of the propagation influence matrix, and calculate the global propagation influence factor of the \(i\)-th index as the mean value of the corresponding row elements, denoted as the propagation influence weight \(w_i\); ANP,i ;

[0097] The fused analytic hierarchy-network dynamic fusion model is expressed as:

[0098] W_i = \alpha\times w_i+(1 - \alpha)\times w_i^h; i * where \(W_i\) is the fusion weight of the \(i\)-th index; \(\alpha\in[0, 1]\) is the fusion adjustment parameter, and its value can be set according to the specific business scenario to achieve the dynamic adjustment of the weight structure and propagation relationship. AHP,i +(1 - \alpha)\times w_i^h; ANP,i ;

[0099] In the formula, \(W_i\) i * is the fusion weight of the \(i\)-th index; \(\alpha\in[0, 1]\) is the fusion adjustment parameter, and its value can be set according to the specific business scenario to achieve the dynamic adjustment of the weight structure and propagation relationship.

[0100] S4: Based on the analytic hierarchy-network dynamic fusion model and combined with the data asset semantic graph, construct a data asset graph structure, and use a graph neural network to perform embedded representation learning on the data asset graph structure, and extract the embedded vector output by the graph neural network at the last layer for each data asset;

[0101] The construction method of the data asset graph structure includes:

[0102] Regarding each data asset as a node in the graph, the initial feature vector of the node consists of the numerical statistical features of the attribute fields of the data asset, the field semantic type coding, and the metric binding information;

[0103] Construct the edge set E = {e ij} between nodes according to the field similarity, graph connection path, and shared metric information in the data asset semantic graph, and calculate the edge weights. The formula is:

[0104]

[0105] In the formula, x i , x j are the initial feature vectors of the nodes, τ is the bandwidth parameter, and all edge weights are used for the propagation calculation of the graph neural network;

[0106] Construct the adjacency matrix A and add self-loop terms to obtain Then calculate the normalized adjacency matrix where is 's degree matrix;

[0107] Use the graph neural convolutional network to perform embedding representation learning on the data asset graph structure. The node representation update process is:

[0108]

[0109] In the formula, H (l) represents the node feature representation of the l-th layer, and H (l+1) represents the node feature representation of the l+1-th layer; W (l) is the trainable weight matrix of the l-th layer, and σ is the activation function;

[0110] Obtain the node embedding matrix H(L) output by the last layer of the graph convolutional neural network. The i-th row embedding vector h i represents the comprehensive feature expression of the i-th data asset under the data asset graph structure and the metric system.

[0111] S5: According to the embedding vectors, determine the fusion weights in combination with the hierarchical-network dynamic fusion model, construct a multi-objective optimization function, use the non-dominated sorting genetic algorithm (NSGA-II) to solve the multi-objective optimization function, generate a Pareto optimal solution set, score, rank, and classify each data asset, and form an evaluation report.

[0112] In one specific embodiment, the method for constructing the multi-objective optimization function includes:

[0113] Take the embedding vector h i of each data asset as the input of the feature expression of the data asset;

[0114] Combined with the fusion weight vector determined in the hierarchical-network dynamic fusion model Construct a multi-objective optimization function F(x i ), and the expression is:

[0115] F(x i ) = [f1(x i ), f2(x i ),..., f m (x i )];

[0116] In the formula, m is the number of dimensions of the evaluation indexes in multi-objective optimization. For example, the evaluation indexes include quality, benefit, risk, liquidity, etc., and there are a total of m = 5 objective functions;

[0117] Any objective function in the multi-objective optimization function F(x i ) is defined as:

[0118] f k (x i ) represents the weighted score of the i-th data asset under the k-th index dimension; is the k-th fusion vector; is the k-th dimensional component in the embedding vector.

[0119] The evaluation report includes: the comprehensive score of the data asset, the single scores of each evaluation index dimension, the Pareto level to which it belongs, and application suggestions, including utilization priorities, risk warnings, or data governance strategy suggestions;

[0120] An example is as follows:

[0121] Comprehensive score: 92.6;

[0122] Index dimension scores: such as data benefit: 93, risk control: 87;

[0123] Pareto level: Grade I;

[0124] Suggestion: Prioritize inclusion in the joint supervision data resource pool.

[0125] The evaluation results are output in a structured data format and provided to the market supervision system through an interface for integrated application.

[0126] As Figure 2 shown, this is another embodiment of the present invention. This embodiment provides a multi-dimensional evaluation system for market supervision data assets based on a hierarchical-network dynamic fusion model, which is applied to the multi-dimensional method for market supervision data assets based on the hierarchical-network dynamic fusion model as described above, and includes:

[0127] A data collection module, which is used to collect multi-source data assets from multiple market supervision business systems, including structured, semi-structured, and unstructured data, and record meta-attribute information for the data assets to form a collection of original data assets;

[0128] A semantic modeling module, which is used to perform data cleaning, field unification, and type standardization processing on the collection of original data assets, and construct a semantic metadata model based on the Resource Description Framework, map the attribute fields of each data asset to the dimension levels or metric nodes in the knowledge graph, and generate a data asset semantic graph for depicting the semantic associations and metric alignment relationships between data assets;

[0129] An index modeling and fusion module, which is used to construct a three-layer evaluation index system based on the data asset semantic graph, calculate the structural weights of each layer of indexes using the Analytic Hierarchy Process, construct an index dependence network based on the semantic dependence relationships between indexes, calculate the propagation influence weights through network analysis, and fuse the structural weights and propagation influence weights to form a hierarchical-network dynamic fusion model;

[0130] An asset graph construction and embedding module, which is used to construct a data asset graph structure based on the hierarchical-network dynamic fusion model and the data asset semantic graph, model each data asset as a node in the graph, and set edge weights according to field similarity and graph connection degree; use a graph neural network to perform embedding representation learning on the data asset graph structure, and output the embedding vector of each data asset at the last layer of the graph neural network;

[0131] An optimization scoring and result generation module, which is used to construct a multi-objective optimization function according to the embedding vectors and in combination with the fusion weights determined by the hierarchical-network dynamic fusion model, solve the function using the non-dominated sorting genetic algorithm, generate a Pareto optimal solution set, perform comprehensive scoring, index dimension scoring, ranking, and grade division on each data asset, and finally generate an evaluation report;

[0132] A result output module, which is used to output the evaluation report in a structured data format and integrate it with an external market supervision platform through a system interface for regulatory decision-making support and data governance optimization.

[0133] In summary, the present invention constructs a unified data asset semantic metadata model, supports the integration of structured, semi-structured, and unstructured data, and performs unified abstraction and semantic binding on field attributes, effectively improving the description ability and scalability of data assets, and solving the problem that it is difficult to fuse and model existing regulatory data; uses the Resource Description Framework (RDF) to construct a data asset semantic graph, and maps field attributes to metric nodes in the regulatory knowledge graph, thereby realizing the unified representation of data content and metric meanings in the evaluation system, and improving the accuracy and interpretability of the evaluation model.

[0134] In the process of index modeling, by combining the Analytic Hierarchy Process (AHP) and the Analytic Network Process (ANP), on the one hand, the index structure weights are constructed based on expert knowledge, and on the other hand, the propagation influence network is constructed based on the dependence relationship between indexes, realizing the dynamic fusion of structural and dependence weights, effectively improving the adaptability and expression ability of the index system, and overcoming the limitation of the traditional single scoring model on the independence assumption of indexes.

[0135] By constructing a data asset graph structure based on field semantic similarity and graph connection relationship, and introducing a Graph Convolutional Network (GCN) to perform embedding learning on asset nodes, a comprehensive feature representation reflecting asset semantic features, index structure, and global influence is extracted, enhancing the perception ability of the evaluation model for complex structural relationships and realizing high-dimensional modeling of data asset value. Based on the asset feature representation output by the graph neural network and combined with the index weights obtained in the fusion model, a multi-objective optimization function is constructed, and the Non-dominated Sorting Genetic Algorithm (NSGA-II) is used to generate and evaluate the solution set for sorting, realizing value trade-off and Pareto optimal classification under multiple index dimensions, and the evaluation results are more scientific and objective, which can be used to support the intelligent scheduling and risk control of data resources.

[0136] The finally generated comprehensive score of data assets, index dimension scores, Pareto levels, and evaluation suggestions can be output in a structured form and connected to the supervision platform through an interface to realize real-time invocation and application feedback of the evaluation model, supporting various actual application scenarios such as data governance, supervision decision-making, and asset optimization.

[0137] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, it can be implemented in whole or in part in the form of a computer program product, and the computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another.

[0138] In addition, each functional unit in various embodiments of the present application can be integrated in one processing module, or each unit can exist physically alone, or two or more units can be integrated in one module. The above integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk, or an optical disc, etc.

[0139] As described above, it is only the specific implementation manner of the present application. However, the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various changes or substitutions, and these should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A multi-dimensional evaluation method for market supervision data assets based on a hierarchical-network dynamic fusion model, characterized in that The method includes: Collecting multi-source data assets from different market supervision business systems, and recording meta-attribute information to construct a collection of original data assets; performing data cleaning, field unification, and type standardization processing on the collection of original data assets, and establishing a semantic metadata model using the Resource Description Framework, mapping the attributes of each data asset to the dimension levels or indicator nodes in the knowledge graph to form a data asset semantic graph for depicting the semantic associations and indicator alignment relationships between data assets; Constructing a three-layer indicator system based on the data asset semantic graph, calculating the structural weights of each layer of indicators using the Analytic Hierarchy Process, and constructing an indicator dependence network using the Network Analysis Method to calculate the propagation influence weights, and fusing the structural weights and the propagation influence weights to form a hierarchical-network dynamic fusion model; Based on the hierarchical-network dynamic fusion model and combined with the data asset semantic graph, constructing a data asset graph structure, and using a graph neural network to perform embedding representation learning on the data asset graph structure to extract the embedding vectors output by each data asset at the last layer of the graph neural network; According to the embedding vectors and combined with the hierarchical-network dynamic fusion model to determine the fusion weights, constructing a multi-objective optimization function, and using the Non-dominated Sorting Genetic Algorithm to solve the multi-objective optimization function to generate a Pareto optimal solution set, scoring, ranking, and classifying each data asset, and forming an evaluation report.

2. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 1, characterized in that The collecting of multi-source data assets from different market supervision business systems includes structured, semi-structured, and unstructured data, and further includes data source authentication, structure parsing, file batch import, and field mapping rule execution, and each collection task is attached with a unique task number for evaluation and traceability; the meta-attribute information includes at least data source, field, timestamp, institutional ownership, and update frequency.

3. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 1, wherein, The semantic metadata model is constructed based on the Resource Description Framework semantic namespace, each data asset instance is modeled as a dl:Source type node, each attribute field in each data asset is modeled as a dl:Domain type node, and each attribute field is mapped to the kpi:Indicator or kpi:Level concept node in the knowledge graph through dl:mapTo; The semantic metadata model records field type, null value ratio, and numerical distribution statistical information.

4. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 1, wherein The three-layer indicator system consists of the following: Goal layer: Defined as "the comprehensive value of data assets in the market supervision system", serving as the top-level node of the Analytic Hierarchy Process; Criterion layer: Includes five dimensions of data quality, data benefit, data risk, data availability, and data stability; Indicator layer: Each dimension of the criterion layer is disassembled into at least two quantifiable indicator fields, and the indicator fields come from the bound attributes in the data asset semantic graph.

5. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 4, characterized in that The method of calculating the structural weights of each layer of indicators using the Analytic Hierarchy Process includes: Constructing judgment matrices between the criterion layer and the goal layer, and between the indicator layer and the criterion layer, using the 1–9 scale method, and performing pairwise comparison scoring by at least five domain experts with supervision experience; Calculate the maximum eigenvalue λ of the judgment matrix using the characteristic root method max , and obtain the corresponding weight eigenvector ω as the local weight of each node; Perform a consistency test on each judgment matrix and calculate the consistency ratio where CI is the consistency index and RI is the random consistency index. When the consistency ratio CR < 0.1, the consistency of the judgment matrix is acceptable. When the consistency ratio CR ≥ 0.1, reconstruct the judgment matrix; The local weights that pass the consistency test are normalized and combined hierarchically to obtain the structural weight w of each evaluation index AHP,i .

6. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 5, characterized in that, The method of constructing an index dependence network using network analysis to calculate the propagation influence weight includes: taking all the evaluation index nodes determined by the analytic hierarchy process as the node set I = {i1, i2,..., i n}, where n is the total number of nodes; Based on the business process logic, indicator causal chain relationships, and statistical correlations reflected in the data asset semantic graph, construct a set of directed dependency edges R = {r ij} between indicators. Each edge r ij indicates that indicator i has an impact on indicator j, and the edge weight w ij ∈ [0, 1]; Construct the adjacency matrix \(W\in\mathbb{R}\) according to the edge weights n×n , where the element \(W\) in the adjacency matrix ij \(= w\) ij , and assign 0 to the unconnected node pairs; Calculate the propagation influence matrix \(T=(I - W)\) -1 , which is used to quantify the degree of indirect influence of each indicator on other indicators; Normalize each row of the propagation influence matrix, and calculate the global propagation influence factor of the \(i\)-th index as the mean of the elements in the corresponding row, denoted as the propagation influence weight \(w\). ANP,i ; The fused hierarchical-network dynamic fusion model is expressed as: In the formula, is the fusion weight of the i-th index; α ∈ [0, 1] is the fusion adjustment parameter.

7. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 1, characterized in that The method for constructing the data asset graph structure includes: Regarding each data asset as a node in the graph, the initial feature vector of the node consists of the numerical statistical features of the attribute fields of the data asset, the field semantic type encoding, and the index binding information; Construct the edge set E = {e ij} between nodes based on field similarity, graph connection paths, and shared metric information in the data asset semantic graph, and calculate the edge weights. The formula is as follows: where x i , x j are the initial feature vectors of the nodes; τ is the bandwidth parameter; Construct the adjacency matrix A and add self-loop terms to obtain Then calculate the normalized adjacency matrix where is the degree matrix of; Using a graph neural convolutional network to perform embedded representation learning on the data asset graph structure, and the node representation update process is as follows: where, H (l) represents the node feature representation of the l-th layer, and H (l+1) represents the node feature representation of the (l + 1)-th layer; W (l) is the trainable weight matrix of the l-th layer, and σ is the activation function; Obtain the node embedding matrix H(L) of the output of the last layer of the graph neural network, where the i-th row vector h i represents the comprehensive feature expression of the i-th data asset under the data asset graph structure and the index system.

8. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 7, wherein The method for constructing the multi-objective optimization function includes: Use the embedding vector h of each data asset i as the feature expression input of this data asset; Combined with the fusion weight vector determined in the hierarchical-network dynamic fusion model Construct a multi-objective optimization function F(x i ), and the expression is as follows: F(x i ) = [f1(x i ), f2(x i ),..., f m (x i )]; In the formula, m is the number of dimensions of the evaluation indicators in the multi-objective optimization; Multi-objective optimization function F(x i ) is defined as any one of the objective functions in: f k (x i ) represents the weighted score of the i-th data asset under the k-th metric dimension; is the k-th fusion vector; is the k-th component in the embedding vector.

9. The multi-dimensional evaluation method for market supervision data assets based on the hierarchical-network dynamic fusion model according to claim 8, characterized in that, The evaluation report includes: the comprehensive score of the data asset, the individual scores of each evaluation indicator dimension, the Pareto level to which it belongs, and application suggestions, including utilization priorities, risk warnings, or data governance strategy suggestions; Output the evaluation in a structured data format and provide it to the market supervision system through an interface for integrated application.

10. A multi-dimensional evaluation system for market supervision data assets based on a hierarchical-network dynamic fusion model, which is applied to the multi-dimensional method for market supervision data assets based on the hierarchical-network dynamic fusion model according to any one of claims 1-9, characterized in that The system includes: A data collection module, which is used to collect multi-source data assets from multiple market supervision business systems, including structured, semi-structured, and unstructured data, and record the meta-attribute information of the data assets to form a set of original data assets; A semantic modeling module, which is used to perform data cleaning, field unification, and type standardization processing on the set of original data assets, and construct a semantic metadata model based on the Resource Description Framework, map the attribute fields of each data asset to the dimension levels or index nodes in the knowledge graph, and generate a data asset semantic graph for depicting the semantic association and index alignment relationship between data assets; An index modeling and fusion module, which is used to construct a three-layer evaluation index system based on the data asset semantic graph, calculate the structural weights of each layer of indicators using the analytic hierarchy process, construct an index dependency network based on the semantic dependency relationship between the indicators, calculate the propagation influence weights through network analysis, and fuse the structural weights and the propagation influence weights to form a hierarchical-network dynamic fusion model; An asset graph construction and embedding module, which is used to construct a data asset graph structure based on the hierarchical-network dynamic fusion model and the data asset semantic graph, model each data asset as a node in the graph, and set the edge weights according to the field similarity and the graph connection degree; use a graph neural network to perform embedded representation learning on the data asset graph structure, and output the embedding vector of each data asset at the last layer of the graph neural network; An optimization scoring and result generation module, which is used to construct a multi-objective optimization function according to the embedding vector and the fusion weights determined by the hierarchical-network dynamic fusion model, solve the function using the non-dominated sorting genetic algorithm, generate a Pareto optimal solution set, perform comprehensive scoring, index dimension scoring, sorting, and level division on each data asset, and finally generate an evaluation report; A result output module, which is used to output the evaluation report in a structured data format and integrate it with an external market supervision platform through a system interface for regulatory decision support and data governance optimization.

Images