Big data law supervision model establishment method and device, equipment and storage medium

Abstract

The invention discloses a big data legal supervision model establishment method, device and equipment and a storage medium, and the method comprises the steps: importing legal supervision data, cleaning missing values and abnormal values, and carrying out the feature dimension reduction of cleaned target legal data based on a principal component analysis algorithm; extracting a core original feature column in the feature data after dimension reduction through principal component load analysis; performing group clustering by using the core original feature column based on a Kmeans clustering algorithm to obtain an optimal clustering result; establishing a preset law supervision model according to the optimal clustering result, and outputting a suspect group analysis result; the problems of poor subjectivity and adaptability and high maintenance cost caused by manual rule presetting can be effectively avoided, the objectivity, efficiency and accuracy of legal supervision are remarkably improved, quantifiable group scale distribution, core feature tags and risk level assessment results are provided for inspection organizations, and the risk level assessment efficiency is improved. And the supervision decision is converted from experience driving to data driving.

Description

Technical Field

[0001] This invention relates to the field of unsupervised learning technology, and in particular to a method, apparatus, device, and storage medium for establishing a big data legal supervision model. Background Technology

[0002] Traditional methods for building big data legal supervision models are based on prosecutors summarizing model rules, and then technical personnel use existing logical rules to correlate, compare, and contrast multi-source data to establish legal supervision model applications and generate expected legal supervision clues.

[0003] This approach relies heavily on model rules derived from prosecutors' case-handling experience. It is impossible to establish a legal oversight model for deep-seated data risks and unpredictable data collision rules.

[0004] With the development of artificial intelligence and machine learning technologies, based on unsupervised learning techniques, without relying on predefined rules, but based on massive amounts of data, computers can autonomously summarize patterns and discover the inherent patterns and characteristics of data through mathematical algorithms such as cluster analysis, autonomously summarize data collision rules, and establish new big data legal supervision models.

[0005] The traditional big data legal supervision model system is constructed in two parts: data source and collision rules.

[0006] First, based on past case-handling experience, the prosecutor knew that there might be clues to problems in a certain area, and initiated a big data legal supervision model.

[0007] The prosecutor first summarizes the data collision rules for the model, then coordinates the data sources required by the model. Finally, based on the existing data sources and the collision rules summarized by the prosecutor, a big data legal supervision model application is established.

[0008] The data source and collision rules need to work closely together to complete the model building.

[0009] See Figure 1 , Figure 1 A schematic diagram of a traditional big data legal supervision model system, such as Figure 1 As shown, given the existing data sources, a big data legal supervision model can only be successfully established by relying on the collision rules summarized by prosecutors.

[0010] This method relies heavily on existing case-handling experience and cannot fully tap the potential of the data. If new types of cases or new scenarios are encountered, potential unknown risks in the data may be missed.

[0011] Traditional methods for establishing big data legal supervision models mainly suffer from the following problems: 1) It relies heavily on the prosecutor's case-handling experience and cannot make deeper use of data to find unknown potential risk points.

[0012] 2) The establishment of a legal supervision model requires human intervention throughout the entire process, from conception and research to management. It is costly to build and consumes a lot of manpower and time. Summary of the Invention

[0013] The main objective of this invention is to provide a method, apparatus, device, and storage medium for establishing a big data legal supervision model, aiming to solve the technical problems in the prior art where legal supervision models rely on prosecutors to manually preset rules, resulting in strong subjectivity, poor adaptability, high maintenance costs, and the inability to automatically identify complex groups of suspects.

[0014] In a first aspect, the present invention provides a method for establishing a big data legal supervision model, the method comprising the following steps: Import legal supervision data and clean up missing and outlier values. Perform feature dimensionality reduction on the cleaned target legal data based on principal component analysis algorithm, and extract the core original feature columns from the dimensionality-reduced feature data through principal component loading analysis. The optimal clustering result is obtained by using the core original feature columns based on the Kmeans clustering algorithm to perform population clustering. Based on the optimal clustering results, a pre-defined legal supervision model is established, and the analysis results of the suspect group are output.

[0015] Optionally, the process of importing legal supervision data and cleaning missing and outlier values, performing feature dimensionality reduction on the cleaned target legal data based on principal component analysis, and extracting the core original feature columns from the dimensionality-reduced feature data through principal component loading analysis includes: Import the legal supervision dataset to be analyzed, and detect the legal supervision data in the dataset to obtain missing values ​​and outliers; The missing and outlier values ​​in the legal supervision data are removed to obtain the target legal data after data cleaning. The target legal data is dimensionality reduced using principal component analysis algorithm to obtain the dimensionality-reduced feature data. The core original feature columns in the feature data are extracted through principal component loading analysis.

[0016] Optionally, the step of performing dimensionality reduction on the target legal data using principal component analysis to obtain dimensionality-reduced feature data includes: The target legal data is standardized using the principal component analysis algorithm according to the following formula to obtain standardized data:

[0017] in, To standardize data, For the first part of the original data The sample at the th Values ​​on each feature For the data involved The crime is Weight values ​​on features For the first The mean of each feature, For the first Standard deviation of each feature; The covariance matrix of the standardized data is calculated using the following formula:

[0018] in, Let covariance matrix be the variance matrix. For sample size, For the standardized data matrix, This is the matrix transpose. Perform eigenvalue decomposition on the covariance matrix and select the eigenvectors corresponding to the k largest eigenvalues; The target legal data is mapped to a dimensionality-reduced space composed of the feature vectors to obtain the dimensionality-reduced feature data.

[0019] Optionally, the step of extracting the core original feature columns from the feature data through principal component loading analysis includes: The loading value of each original feature column in the feature data on the principal components is calculated using the following formula:

[0020] in, For load values, Main component and original variables The correlation coefficient, For the first Eigenvalues ​​of each principal component; The original feature columns whose absolute value of the load value is greater than a preset load value threshold are taken as the core original feature columns.

[0021] Optionally, the K-means clustering algorithm uses the core original feature columns to perform group clustering to obtain the optimal clustering result, including: Based on the K-means clustering algorithm, using the core original feature columns as input data, the Euclidean distance between the data object and each cluster center of a preset number of cluster centers is calculated iteratively using the following formula:

[0022] in, For data objects With the Cluster centers Euclidean distance, For the dimensions of data objects, For data objects The Each attribute value For the first Cluster centers The Each attribute value; Clusters are assigned based on the Euclidean distance, and the average distance between each cluster is calculated. If the average distance satisfies the following formula, the current cluster is discarded:

[0023] in, For the dimensions of data objects, For data objects The Each attribute value The average distance threshold; Repeat the cluster assignment and average distance calculation process until the average distance of all clusters is less than the average distance threshold. The mean vector of the cluster is updated by the following formula, and the iteration continues until a preset number of iterations is reached or the mean vector no longer changes.

[0024] in, Cluster The mean vector, For the first Cluster centers, For data objects; The K-means model with different numbers of clusters is systematically scored, and the clustering scheme with the highest score is dynamically selected as the optimal clustering result.

[0025] Optionally, the step of assigning clusters based on the Euclidean distance, systematically scoring K-means models with different numbers of clusters, and dynamically selecting the clustering scheme with the highest score as the optimal clustering result includes: Clusters are assigned based on the Euclidean distance. For K-means models with different numbers of clusters, the ratio of inter-cluster dispersion to intra-cluster compactness is calculated using the following formula:

[0026] in, The ratio of inter-cluster dispersion to intra-cluster compactness is the Calinski-Harabasz index, which is the score for a cluster size of K. Inter-cluster dispersion Cluster compactness The total number of samples, The number of clusters; The clustering scheme with the largest ratio is selected as the optimal clustering result.

[0027] Optionally, the step of establishing a preset legal supervision model based on the optimal clustering results and outputting the analysis results of the suspect group includes: The suspect group formed by the optimal clustering results will be used as the basic data. Based on the aforementioned basic data, a pre-defined legal supervision model based on unsupervised learning is generated through statistical analysis. This model includes group size distribution, core feature labels, and risk level assessment. The analysis results of the suspect group are then output through the pre-defined legal supervision model.

[0028] Secondly, to achieve the above objectives, the present invention also proposes a big data legal supervision model establishment device, the big data legal supervision model establishment device comprising: The core feature extraction module is used to import legal supervision data and clean missing and outlier values. It performs feature dimensionality reduction on the cleaned target legal data based on principal component analysis algorithm, and extracts the core original feature columns from the dimensionality-reduced feature data through principal component loading analysis. The clustering optimization module is used to perform population clustering based on the K-means clustering algorithm using the core original feature columns to obtain the optimal clustering result; The model building output module is used to build a preset legal supervision model based on the optimal clustering results and output the analysis results of the suspect group.

[0029] Thirdly, to achieve the above objectives, the present invention also proposes a big data legal supervision model establishment device, which includes: a memory, a processor, and a big data legal supervision model establishment program stored in the memory and capable of running on the processor. The big data legal supervision model establishment program is configured to implement the steps of the big data legal supervision model establishment method described above.

[0030] Fourthly, to achieve the above objectives, the present invention also proposes a storage medium storing a big data legal supervision model establishment program, wherein when the big data legal supervision model establishment program is executed by a processor, it implements the steps of the big data legal supervision model establishment method described above.

[0031] The big data legal supervision model establishment method proposed in this invention involves importing legal supervision data and cleaning missing and outlier values. Principal component analysis (PCA) is used to perform feature dimensionality reduction on the cleaned target legal data, and PCA is used to extract core original feature columns from the dimensionality-reduced feature data. K-means clustering algorithm is then used to perform group clustering based on these core original feature columns to obtain the optimal clustering result. A pre-set legal supervision model is established based on the optimal clustering result, outputting the suspect group analysis results. This method effectively avoids the problems of subjectivity, poor adaptability, and high maintenance costs caused by manually pre-set rules, significantly improving the objectivity, efficiency, and accuracy of legal supervision. It provides procuratorial organs with quantifiable group size distribution, core feature labels, and risk level assessment results, enabling a shift in supervisory decision-making from experience-driven to data-driven. It eliminates the need for prosecutors to pre-set model rules, reduces the time cost of manual data analysis, lowers the threshold for building big data legal supervision models, improves equipment utilization, reduces the cost of building big data legal supervision models, fully explores data value, avoids overlooking criminal facts, and effectively combats crime. Attached Figure Description

[0032] Figure 1 A schematic diagram of a traditional big data legal supervision model system; Figure 2 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention; Figure 3 This is a flowchart illustrating the first embodiment of the big data legal supervision model establishment method of the present invention; Figure 4 This is a flowchart illustrating the second embodiment of the big data legal supervision model establishment method of the present invention; Figure 5 This is a flowchart illustrating the third embodiment of the big data legal supervision model establishment method of the present invention; Figure 6 This is a schematic diagram of the system for establishing a big data legal supervision model based on unsupervised learning in the big data legal supervision model establishment method of the present invention; Figure 7 This is a schematic diagram of the data table in the big data legal supervision model establishment method of the present invention; Figure 8 This is a schematic diagram of the data after feature extraction in the big data legal supervision model establishment method of the present invention; Figure 9 This is a schematic diagram of the Calinski Harabasz Score used in the big data legal supervision model establishment method of this invention; Figure 10 This is a schematic diagram of the suspect feature analysis radar in the big data legal supervision model establishment method of the present invention; Figure 11This is a schematic diagram of the suspect feature comparison radar in the big data legal supervision model establishment method of the present invention; Figure 12 This is a functional block diagram of the first embodiment of the big data legal supervision model establishment device of the present invention.

[0033] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0034] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0035] The solution of this invention mainly involves: importing legal supervision data and cleaning missing and outlier values; performing feature dimensionality reduction on the cleaned target legal data based on principal component analysis; extracting core original feature columns from the dimensionality-reduced feature data through principal component loading analysis; performing group clustering using the core original feature columns based on the K-means clustering algorithm to obtain the optimal clustering result; establishing a preset legal supervision model based on the optimal clustering result and outputting the suspect group analysis result; effectively avoiding the problems of subjectivity, poor adaptability, and high maintenance costs caused by manually preset rules, significantly improving the objectivity, efficiency, and accuracy of legal supervision, providing procuratorial organs with quantifiable group size distribution, core feature labels, and risk level assessment results, realizing the transformation of supervision decision-making from experience-driven to data-driven, eliminating the need for prosecutors to preset model rules in advance, reducing the time cost of manual data analysis, lowering the threshold for building big data legal supervision models, improving equipment utilization, reducing the cost of building big data legal supervision models, fully exploring data value, avoiding omissions of criminal facts, effectively combating crime, and solving the technical problems of strong subjectivity, poor adaptability, high maintenance costs, and inability to automatically identify complex suspect groups caused by the reliance on prosecutors' manually preset rules in existing legal supervision models.

[0036] Reference Figure 2 , Figure 2 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention.

[0037] like Figure 2As shown, the device may include: a processor 1001, such as a CPU; a communication bus 1002; a user interface 1003; a network interface 1004; and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0038] Those skilled in the art will understand that Figure 2 The device structure shown does not constitute a limitation on the device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0039] like Figure 2 As shown, the memory 1005, which serves as a storage medium, may include an operating device, a network communication module, a user interface module, and a big data legal supervision model establishment program.

[0040] The device of this invention calls the big data legal supervision model establishment program stored in the memory 1005 through the processor 1001, and performs the following operations: Import legal supervision data and clean up missing and outlier values. Perform feature dimensionality reduction on the cleaned target legal data based on principal component analysis algorithm, and extract the core original feature columns from the dimensionality-reduced feature data through principal component loading analysis. The optimal clustering result is obtained by using the core original feature columns based on the Kmeans clustering algorithm to perform population clustering. Based on the optimal clustering results, a pre-defined legal supervision model is established, and the analysis results of the suspect group are output.

[0041] The device of this invention, through processor 1001 calling the big data legal supervision model establishment program stored in memory 1005, also performs the following operations: Import the legal supervision dataset to be analyzed, and detect the legal supervision data in the dataset to obtain missing values ​​and outliers; The missing and outlier values ​​in the legal supervision data are removed to obtain the target legal data after data cleaning. The target legal data is dimensionality reduced using principal component analysis algorithm to obtain the dimensionality-reduced feature data. The core original feature columns in the feature data are extracted through principal component loading analysis.

[0042] The device of this invention, through processor 1001 calling the big data legal supervision model establishment program stored in memory 1005, also performs the following operations: The target legal data is standardized using the principal component analysis algorithm according to the following formula to obtain standardized data:

[0043] in, To standardize data, For the first part of the original data The sample at the th Values ​​on each feature For the data involved The crime is Weight values ​​on features For the first The mean of each feature, For the first Standard deviation of each feature; The covariance matrix of the standardized data is calculated using the following formula:

[0044] in, Let covariance matrix be the variance matrix. For sample size, For the standardized data matrix, This is the matrix transpose. Perform eigenvalue decomposition on the covariance matrix and select the eigenvectors corresponding to the k largest eigenvalues; The target legal data is mapped to a dimensionality-reduced space composed of the feature vectors to obtain the dimensionality-reduced feature data.

[0045] The device of this invention, through processor 1001 calling the big data legal supervision model establishment program stored in memory 1005, also performs the following operations: The loading value of each original feature column in the feature data on the principal components is calculated using the following formula:

[0046] in, For load values, Main component and original variables The correlation coefficient, For the first Eigenvalues ​​of each principal component; The original feature columns whose absolute value of the load value is greater than a preset load value threshold are taken as the core original feature columns.

[0047] The device of this invention, through processor 1001 calling the big data legal supervision model establishment program stored in memory 1005, also performs the following operations: Based on the K-means clustering algorithm, using the core original feature columns as input data, the Euclidean distance between the data object and each cluster center of a preset number of cluster centers is calculated iteratively using the following formula:

[0048] in, For data objects With the Cluster centers Euclidean distance, For the dimensions of data objects, For data objects The Each attribute value For the first Cluster centers The Each attribute value; Clusters are assigned based on the Euclidean distance, and the average distance between each cluster is calculated. If the average distance satisfies the following formula, the current cluster is discarded:

[0049] in, For the dimensions of data objects, For data objects The Each attribute value The average distance threshold; Repeat the cluster assignment and average distance calculation process until the average distance of all clusters is less than the average distance threshold. The mean vector of the cluster is updated by the following formula, and the iteration continues until a preset number of iterations is reached or the mean vector no longer changes.

[0050] in, Cluster The mean vector, For the first Cluster centers, For data objects; The K-means model with different numbers of clusters is systematically scored, and the clustering scheme with the highest score is dynamically selected as the optimal clustering result.

[0051] The device of this invention, through processor 1001 calling the big data legal supervision model establishment program stored in memory 1005, also performs the following operations: Clusters are assigned based on the Euclidean distance. For K-means models with different numbers of clusters, the ratio of inter-cluster dispersion to intra-cluster compactness is calculated using the following formula:

[0052] in, The ratio of inter-cluster dispersion to intra-cluster compactness is the Calinski-Harabasz index, which is the score for a cluster size of K. Inter-cluster dispersion Cluster compactness The total number of samples, The number of clusters; The clustering scheme with the largest ratio is selected as the optimal clustering result.

[0053] The device of this invention, through processor 1001 calling the big data legal supervision model establishment program stored in memory 1005, also performs the following operations: The suspect group formed by the optimal clustering results will be used as the basic data. Based on the aforementioned basic data, a pre-defined legal supervision model based on unsupervised learning is generated through statistical analysis. This model includes group size distribution, core feature labels, and risk level assessment. The analysis results of the suspect group are then output through the pre-defined legal supervision model.

[0054] This embodiment, through the above-described scheme, imports legal supervision data and cleans missing and outlier values. Principal component analysis (PCA) is used to perform feature dimensionality reduction on the cleaned target legal data, and PCA extracts the core original feature columns from the dimensionality-reduced feature data. K-means clustering is then used to perform group clustering based on these core original feature columns to obtain the optimal clustering result. A pre-defined legal supervision model is established based on the optimal clustering result, outputting the suspect group analysis results. This effectively avoids the problems of subjectivity, poor adaptability, and high maintenance costs caused by manually pre-defined rules, significantly improving the objectivity, efficiency, and accuracy of legal supervision. It provides procuratorial organs with quantifiable group size distribution, core feature labels, and risk level assessment results, realizing the transformation of supervision decision-making from experience-driven to data-driven. It eliminates the need for prosecutors to pre-define model rules, reduces the time cost of manual data analysis, lowers the threshold for building big data legal supervision models, improves equipment utilization, reduces the cost of building big data legal supervision models, fully explores data value, avoids overlooking criminal facts, and effectively combats crime.

[0055] Based on the above hardware structure, an embodiment of the big data legal supervision model establishment method of the present invention is proposed.

[0056] Reference Figure 3 , Figure 3 This is a flowchart illustrating the first embodiment of the big data legal supervision model establishment method of the present invention.

[0057] In the first embodiment, the method for establishing the big data legal supervision model includes the following steps: Step S10: Import legal supervision data and clean missing and outlier values. Perform feature dimensionality reduction on the cleaned target legal data based on principal component analysis algorithm, and extract the core original feature columns from the dimensionality-reduced feature data through principal component loading analysis.

[0058] It should be noted that by importing the legal supervision dataset and cleaning up missing and outlier values, the cleaned data can be dimensionality reduced using the Principal Component Analysis (PCA) algorithm. The core original feature columns in the dimensionality-reduced feature data can be extracted through principal component loading analysis. These feature columns can effectively represent the key business dimensions in the original legal supervision data, providing a high-quality, low-noise input foundation for subsequent clustering analysis.

[0059] Step S20: Using the core original feature columns based on the K-means clustering algorithm, perform group clustering to obtain the optimal clustering result.

[0060] It should be understood that, based on the K-means clustering algorithm, core original feature columns are used as input data to perform group clustering, obtain the optimal clustering result, and dynamically select the clustering scheme with the highest ratio as the optimal clustering result. This achieves objective and automated division of the suspect group, avoids subjective bias caused by manually preset rules, and ensures the scientificity and repeatability of group division.

[0061] Step S30: Establish a preset legal supervision model based on the optimal clustering results, and output the analysis results of the suspect group.

[0062] Understandably, based on the optimal clustering results, a pre-defined legal supervision model can be established, thereby outputting the analysis results of the suspect group, providing the procuratorial organs with objective and quantitative basis for supervision decisions, and realizing the transformation from experience-driven to data-driven supervision model.

[0063] This embodiment, through the above-described scheme, imports legal supervision data and cleans missing and outlier values. Principal component analysis (PCA) is used to perform feature dimensionality reduction on the cleaned target legal data, and PCA extracts the core original feature columns from the dimensionality-reduced feature data. K-means clustering is then used to perform group clustering based on these core original feature columns to obtain the optimal clustering result. A pre-defined legal supervision model is established based on the optimal clustering result, outputting the suspect group analysis results. This effectively avoids the problems of subjectivity, poor adaptability, and high maintenance costs caused by manually pre-defined rules, significantly improving the objectivity, efficiency, and accuracy of legal supervision. It provides procuratorial organs with quantifiable group size distribution, core feature labels, and risk level assessment results, realizing the transformation of supervision decision-making from experience-driven to data-driven. It eliminates the need for prosecutors to pre-define model rules, reduces the time cost of manual data analysis, lowers the threshold for building big data legal supervision models, improves equipment utilization, reduces the cost of building big data legal supervision models, fully explores data value, avoids overlooking criminal facts, and effectively combats crime.

[0064] Furthermore, Figure 4 This is a flowchart illustrating the second embodiment of the big data legal supervision model establishment method of the present invention, as shown below. Figure 4 As shown, based on the first embodiment, a second embodiment of the big data legal supervision model establishment method of the present invention is proposed. In this embodiment, step S10 specifically includes the following steps: Step S11: Import the legal supervision dataset to be analyzed, and detect the legal supervision data in the legal supervision dataset to obtain missing values ​​and outliers.

[0065] It should be noted that after importing the original datasets in the field of legal supervision (such as structured data such as case records, crime types, amounts involved, and criminal details) into the analysis system, data detection technologies (such as statistical verification or business rule matching) can automatically identify missing values ​​(such as empty or blank fields) and outliers (such as negative amounts involved, logical errors such as contradictions between crime classification and case type). This provides accurate input basis for subsequent data cleaning, ensures the integrity and reliability of the input data, and avoids the accuracy of feature extraction and cluster analysis being affected by data quality issues.

[0066] Step S12: Remove the missing values ​​and outliers from the legal supervision data to obtain the cleaned target legal data.

[0067] Understandably, by removing missing values ​​(such as empty or blank fields) and outliers (such as negative amounts involved, logical contradictions between the crime and the case type, and other outliers outside of business rules) detected in the legal supervision dataset, we can ensure the integrity and business rationality of the data. This will result in high-quality cleaned target legal data, providing a reliable and noise-free input basis for subsequent feature reduction and clustering analysis, and avoiding model bias or analysis failure due to data quality issues.

[0068] Step S13: Perform dimensionality reduction on the target legal data using the principal component analysis algorithm to obtain the dimensionality-reduced feature data.

[0069] It should be understood that applying principal component analysis (PCA) to reduce the dimensionality of cleaned legal supervision data can yield dimensionality-lower but more information-rich reduced feature data, providing a high-quality, low-noise input basis for subsequent principal component load analysis to extract core original feature columns.

[0070] Furthermore, step S13 specifically includes the following steps: The target legal data is standardized using the principal component analysis algorithm according to the following formula to obtain standardized data:

[0071] in, To standardize data, For the first part of the original data The sample at the th Values ​​on each feature For the data involved The crime is Weight values ​​on features For the first The mean of each feature, For the first Standard deviation of each feature; The covariance matrix of the standardized data is calculated using the following formula:

[0072] in, Let covariance matrix be the variance matrix. For sample size, For the standardized data matrix, This is the matrix transpose. Perform eigenvalue decomposition on the covariance matrix and select the eigenvectors corresponding to the k largest eigenvalues; The target legal data is mapped to a dimensionality-reduced space composed of the feature vectors to obtain the dimensionality-reduced feature data.

[0073] Understandably, by introducing a standardized formula for crime feature weights, the target legal data is processed for business adaptability, where crime weights quantify the importance of different crimes in the feature dimension. Subsequently, the covariance matrix of the standardized data is calculated to capture the correlation between features. Then, the covariance matrix is ​​decomposed to select the top k feature vectors with the highest contribution. Finally, the original data is projected onto this low-dimensional feature space to obtain dimensionality-reduced feature data that retains the variance information of the core business dimension, providing an accurate input basis for the subsequent principal component loading analysis to extract high-information feature columns.

[0074] Step S14: Extract the core original feature columns from the feature data through principal component loading analysis.

[0075] It is understandable that principal component loading analysis can extract the core original feature columns from the feature data, retain the data variance information to the greatest extent and eliminate redundant features, thereby providing a high-information, low-noise, high-quality input foundation for subsequent cluster analysis.

[0076] Furthermore, step S14 specifically includes the following steps: The loading value of each original feature column in the feature data on the principal components is calculated using the following formula:

[0077] in, For load values, Main component and original variables The correlation coefficient, For the first Eigenvalues ​​of each principal component; The original feature columns whose absolute value of the load value is greater than a preset load value threshold are taken as the core original feature columns.

[0078] It should be understood that by calculating the correlation coefficient between the original feature columns and the principal components and multiplying it by the square root of the principal component eigenvalues, the loading value is obtained to quantify the contribution strength of the original feature columns to the principal components. Subsequently, the original feature columns whose absolute loading values ​​are greater than a preset threshold (e.g., 0.8 or other values, which are not limited in this embodiment) are selected as core feature columns to ensure that these features can effectively characterize the key business dimensions (such as crime elements, amount involved, etc.) in the original legal supervision data, retain the data variance information to the greatest extent and eliminate redundant features, so as to provide a high-quality and high-information input basis for subsequent cluster analysis.

[0079] Accordingly, step S20 specifically includes the following steps: Based on the K-means clustering algorithm, using the core original feature columns as input data, the Euclidean distance between the data object and each cluster center of a preset number of cluster centers is calculated iteratively using the following formula:

[0080] in, For data objects With the Cluster centers Euclidean distance, For the dimensions of data objects, For data objects The Each attribute value For the first Cluster centers The Each attribute value; Clusters are assigned based on the Euclidean distance, and the average distance between each cluster is calculated. If the average distance satisfies the following formula, the current cluster is discarded:

[0081] in, For the dimensions of data objects, For data objects The Each attribute value The average distance threshold; Repeat the cluster assignment and average distance calculation process until the average distance of all clusters is less than the average distance threshold. The mean vector of the cluster is updated by the following formula, and the iteration continues until a preset number of iterations is reached or the mean vector no longer changes.

[0082] in, Cluster The mean vector, For the first Cluster centers, For data objects; The K-means model with different numbers of clusters is systematically scored, and the clustering scheme with the highest score is dynamically selected as the optimal clustering result.

[0083] It should be noted that, based on the core original feature columns, the distance between the data object and the cluster center is calculated using the Euclidean distance formula, and the data is assigned to the nearest cluster. Then, the average distance of each cluster is calculated. If it exceeds a preset threshold, the cluster is discarded. The assignment and average distance calculation are repeated until all clusters meet the average distance. Next, the cluster mean vector is updated, and iterative optimization is performed until a preset number of times or the mean vector converges. Finally, the models with different numbers of clusters are systematically scored, and the clustering scheme with the highest score is dynamically selected as the optimal result. By filtering sparse clusters, the clustering quality is effectively improved, the interference of sparse features is avoided, and the objectivity and accuracy of the suspect group classification are ensured.

[0084] Furthermore, the step of cluster assignment based on the Euclidean distance, systematically scoring the K-means models with different numbers of clusters, and dynamically selecting the clustering scheme with the highest score as the optimal clustering result, specifically includes the following steps: Clusters are assigned based on the Euclidean distance. For K-means models with different numbers of clusters, the ratio of inter-cluster dispersion to intra-cluster compactness is calculated using the following formula:

[0085] in, The ratio of inter-cluster dispersion to intra-cluster compactness is the Calinski-Harabasz index, which is the score for a cluster size of K. Inter-cluster dispersion Cluster compactness The total number of samples, The number of clusters; The clustering scheme with the largest ratio is selected as the optimal clustering result.

[0086] It should be understood that after completing the Euclidean distance cluster assignment, for K-means models with different numbers of clusters (K values, such as K=2,3,4...), the Calinski-Harabasz index CH(K) is calculated as a clustering quality evaluation index. By systematically comparing the CH(K) scores corresponding to different K values, the clustering scheme corresponding to the maximum value of CH(K) is selected as the optimal clustering result, thereby achieving objective and automated selection of the number of clusters and avoiding subjective bias caused by manually preset K values.

[0087] This embodiment, through the above-described scheme, imports the legal supervision dataset to be analyzed, detects missing and outlier values ​​in the legal supervision data within the dataset, removes the missing and outlier values ​​to obtain cleaned target legal data, performs dimensionality reduction on the target legal data using principal component analysis (PCA) to obtain dimensionality-reduced feature data, and extracts core original feature columns from the feature data through PCA. This accurately extracts high-information features and eliminates redundant noise, providing a high-quality input foundation for subsequent clustering analysis. This effectively avoids subjective bias caused by manually preset rules, significantly improves the data processing efficiency and feature representation capability of the legal supervision model, and provides reliable support for the objective quantitative analysis of suspect groups.

[0088] Furthermore, Figure 5 This is a flowchart illustrating the third embodiment of the big data legal supervision model establishment method of the present invention, as shown below. Figure 5 As shown, based on the first embodiment, a third embodiment of the big data legal supervision model establishment method of the present invention is proposed. In this embodiment, step S30 specifically includes the following steps: Step S31: Use the clustered suspect group as the basic data based on the optimal clustering result.

[0089] It should be noted that the suspect groups (such as groups with concentrated amounts of money involved and similar types of crimes) segmented by the optimal clustering scheme (i.e., the clustering result with the highest Calinski-Harabasz index score) in the clustering analysis are used as the basis of structured data to provide accurate group segmentation input for the legal supervision model. This ensures that the model analysis is directly based on objective clustering results rather than human assumptions, thereby improving the pertinence and reliability of subsequent analyses such as group feature label generation and risk level assessment.

[0090] Step S32: Based on the basic data, generate a preset legal supervision model based on unsupervised learning through statistical analysis, which includes group size distribution, core feature labels and risk level assessment, and output the suspect group analysis results through the preset legal supervision model.

[0091] Understandably, the suspect groups formed based on the optimal clustering results serve as the foundational data. Through statistical analysis methods (such as frequency distribution calculation, feature aggregation, and risk weight assessment), structured analysis results are automatically generated. These results include group size distribution (sample size and proportion of each suspect group), core feature labels (such as the range of amounts involved, crime type clustering, and attributes of the involved parties), and risk level assessment (quantitative scoring based on feature contribution). This constructs and outputs the analysis results of a pre-set legal supervision model based on unsupervised learning, providing the procuratorate with objective and quantifiable evidence for group supervision. This significantly improves the accuracy, efficiency, and scientific nature of legal supervision, achieving a substantial shift from manual experience-based judgment to a data-driven supervision model.

[0092] In specific implementations, such as Figure 6 As shown, Figure 6 This is a schematic diagram of the system for establishing a big data legal supervision model based on unsupervised learning in the big data legal supervision model establishment method of the present invention. See also Figure 6 This method leverages massive amounts of data of various types and utilizes unsupervised learning techniques in autonomous machine learning to allow computers to autonomously identify abnormal data, summarize and deduce data collision rules, and thus establish a big data legal supervision model. This approach fully utilizes the data analysis and prediction capabilities of computers to better understand and mine data information.

[0093] It should be noted that the above-described solution in this embodiment has the following beneficial effects: 1) Prosecutors no longer need to pre-set model rules, reducing the time cost of manual data analysis and lowering the threshold for building big data legal supervision models. The work that prosecutors and data analysis engineers used to need more than 10 days to complete has been shortened to more than 10 hours.

[0094] 2) Based on selected data features, the computer automatically performs data analysis and classification, extracts model rules, and creates a big data legal supervision model without human intervention. This improves equipment utilization and reduces the cost of building big data legal supervision models.

[0095] 3) Based on unsupervised learning technology, computers can autonomously learn and summarize data patterns, discover abnormal data features and build models, which can fully explore the value of data, avoid missing criminal facts, and effectively combat crime.

[0096] In practical implementation, taking the data of suspects of dangerous driving as an example, and using unsupervised learning to build a big data legal supervision model, we can use Scikit-learn's clustering algorithm to divide the suspect data into groups, and analyze and give the type of each suspect group based on the suspect indicators.

[0097] Step 1: Data import.

[0098] See Figure 7 , Figure 7 This is a schematic diagram of a data table in the big data legal supervision model establishment method of the present invention, as shown below. Figure 7 As shown, import the data of suspects for dangerous driving from Excel into the system.

[0099] The data contains more than 60,000 entries, each with 327 features.

[0100] The characteristics include the department's case number, the name of the handling unit, the name, gender, ethnicity, nationality, place of household registration, place of residence, work unit / school, rank, position, occupation, alcohol content, sentence pronounced in the first instance, sentence term pronounced in the first instance, and age at the time of the crime.

[0101] Step 2: Data preprocessing.

[0102] The suspect's original data contained a small number of missing and outlier values, and needed to be cleaned before it could be used for analysis.

[0103] For example, records such as: alcohol content is blank, first-instance fine amount is blank, first-instance sentence term is blank, first-instance sentence term needs to be converted into a uniform number, and fine amount less than 10 needs to be uniformly multiplied by 10000, etc.

[0104] Step 3: Feature dimensionality reduction.

[0105] Because the raw data has too many features to be directly used for suspect data analysis, it is necessary to perform dimensionality reduction on the features for subsequent cluster analysis.

[0106] Considering that the weights of characteristic elements differ among different crimes in actual case handling during the prosecution process, how can we combine the element weights to accurately perform feature dimensionality reduction without causing the loss of important indicator data features?

[0107] Therefore, the principal component analysis algorithm based on prosecutorial data solves the problem of feature loss of important indicator data after feature dimensionality reduction.

[0108] The steps are as follows: 1) Define the weight values ​​of each characteristic element for different crimes:

[0109] in, For the first part of the original data The charges against the first Values ​​on each feature; For example: , wait 2) Data standardization.

[0110] To avoid the influence of feature dimensions on the results, all raw data needs to be standardized, taking into account the influence of different element weights for different crimes.

[0111] 3) Covariance matrix.

[0112] The covariance matrix of standardized data is calculated using the above formula, where n is the sample size, i.e., the number of observations, and T is the matrix transpose, which means swapping the rows and columns of the matrix.

[0113] 4) Eigenvalue decomposition.

[0114] Perform eigenvalue decomposition on the covariance matrix Σ:

[0115] in, The features after decomposition The eigenvalue represents the variance of the data along the direction of the corresponding eigenvector. is the eigenvector, representing the direction of the principal component.

[0116] 5) Principal Component Selection. Sort the eigenvalues ​​in descending order and select the eigenvectors corresponding to the k largest eigenvalues ​​(the k principal components) to form a matrix. .

[0117] 6) Data dimensionality reduction.

[0118] Mapping the original data X to a lower-dimensional space:

[0119] in, For the dimensionality-reduced data, It is a matrix composed of the first k eigenvectors.

[0120] Step 4: Feature Extraction. Since the final cluster analysis requires analysis based on the feature columns of the original data, it aims to determine which features of the original data are abnormal, summarize legal supervision rules, and establish a legal supervision model.

[0121] Therefore, after feature dimensionality reduction in step 3, it is necessary to find the corresponding core original feature columns. This invention uses principal component loading analysis algorithm to find the core feature columns of the original data features. The specific steps are as follows: 1) Calculate the principal component loadings.

[0122] For each principal component, its loading represents the weights of the original variables after standardization.

[0123] Principal component loadings can be obtained by calculating the covariance between the principal components and the original variables.

[0124] Specifically, for the i-th principal component and the i-th original variable, the loading is:

[0125] in, For load values, Main component and original variables The correlation coefficient, For the first The eigenvalues ​​of each principal component.

[0126] The specific implementation steps are as follows: First, for the original variables... Centralization and normalization are performed:

[0127] Where is the mean and σj is the standard deviation, the covariance matrix or correlation coefficient matrix of the standardized data is calculated, and its eigenvalues ​​are solved. and the corresponding unit eigenvector (Requires ||ei||=1), finally calculate the original features For principal components load It is obtained by multiplying the eigenvectors by the square roots of their eigenvalues:

[0128] Assuming the first principal component eigenvalues , corresponding feature vector The load is: , indicating the first primitive feature and The load is 0.76, and the second original feature is... The load is -1.36.

[0129] 2) Feature extraction.

[0130] Principal components after dimensionality reduction of features from step 3 and original eigenvalues Substituting these values ​​into step 1) above, we can obtain the Loadings(i,j) values, which represent the loadings of each original variable j on the principal component i. By selecting original variables j with Loadings(i,j) values ​​greater than 0.8 for each principal component i, we can finally obtain the core feature values ​​j in the original data.

[0131] Data after feature extraction, such as Figure 8 As shown, Figure 8This is a schematic diagram of the data after feature extraction in the big data legal supervision model establishment method of this invention. Taking the data of the suspect in this dangerous driving crime case as an example, the NJXE core features of the suspect are finally obtained, which are: N ZASNL: The suspect's age at the time of the crime.

[0132] J JJHL: Suspect's blood alcohol content.

[0133] X YSXGXXQ: The suspect's sentence was announced in the first instance.

[0134] E YSFJSE_NUM: The amount of the first-instance fine imposed on the suspect (in ten thousand yuan).

[0135] Step 5: Cluster analysis.

[0136] The traditional K-means algorithm iteratively assigns data points to the nearest centroid and minimizes the sum of squares of the distances from all points to their centroids by updating the centroids.

[0137] Considering the sparsity of data in actual case handling during the prosecutorial process due to various reasons (untimely case card filling, lack of external data acquisition, etc.), this invention implements an improved K-means algorithm that uses a novel distance metric function to solve the problem of sparsity in prosecutorial data. The specific steps are as follows: 1) Randomly select K objects from the data as cluster centers for K classes. Considering the sparsity of the data, a relatively large number of initial classes are pre-defined, and the initial K value is usually chosen to be 20% of the total number of feature values.

[0138] 2) Assign other data to the class closest to it.

[0139] Based on the randomly selected center point C in step 1 i Calculate the relationship between the remaining data objects and the cluster center C. i Using Euclidean distance, find the cluster center C that is closest to the target data object. i And assign the data objects to cluster center C. i From the corresponding clusters, K cluster groups are obtained.

[0140]

[0141] in, For data objects With the Cluster centers Euclidean distance, For the dimensions of data objects, For data objects The Each attribute value For the first Cluster centers The Each attribute value; 3) Calculate the average distance in each class. Given an average distance threshold Q in a class, discard classes whose average distance exceeds the threshold Q.

[0142]

[0143] in, For the dimensions of data objects, For data objects The Each attribute value The average distance threshold; This filters out sparse data categories, reduces reliance on sparse features, and ensures that the data ultimately used for classification are all core feature data.

[0144] 4) Repeat step 3) until the average distance in each class is less than the threshold Q. This results in N clusters, which are: .

[0145] Update the cluster mean vector based on the cluster to which each sample belongs.

[0146]

[0147] 5) Among them, Cluster The mean vector, For the first Cluster centers, For data objects.

[0148] 6) Continue to divide and classify based on the new centroids, repeating the iteration (repeating 2 to 5). When the set number of iterations is reached or the mean vector of the clusters no longer changes, the model construction is complete, and the clustering algorithm results are output.

[0149] Step 6: Model scoring.

[0150] See Figure 9 , Figure 9 This is a schematic diagram of the Calinski-Harabasz Score used in the big data legal supervision model establishment method of this invention, as shown below. Figure 9As shown in the figure, the horizontal axis represents the number of clusters (K value, ranging from 2 to 14), and the vertical axis represents the Calinski-Harabasz Score (the ratio of inter-cluster dispersion to intra-cluster compactness). As can be seen from the graph, when K=8, the Score reaches a peak of >8500, which is significantly higher than other K values ​​(such as about 7500 when K=14 and about 8000 when K=6), indicating that the 10-cluster scheme performs optimally under the condition of maximum inter-cluster dispersion and minimum intra-cluster dispersion.

[0151] Clustering algorithms are used to cluster the original feature data to obtain the type analysis results of suspects of dangerous driving crimes, but it is difficult to determine how many classes to cluster them into.

[0152] Therefore, a loop method is needed, using the number of clusters as the loop variable (2 to 14), to train K-means models with different numbers of clusters in sequence, and using the calinski_harabasz_score method to score the results of each K-means model.

[0153] Step 7: Results Analysis.

[0154] See Figure 10 , Figure 10 This is a schematic diagram of the suspect feature analysis radar in the big data legal supervision model establishment method of the present invention, as shown below. Figure 10 As shown, the feature radar charts of five suspect groups (clusters 0 to 4) generated based on the clustering results are shown. Each sub-chart in the chart corresponds to one group. The radar dimensions include feature terms such as J, E, X, NX, and N. The definition of the feature terms can be customized according to the actual application scenario. For example: J (Joint suspicious contacts): Frequency of contact with suspicious persons, referring to the number of times the suspect has recently had contact with persons with criminal records or sensitive identities. The higher the value, the closer the connection. E (Economic abnormal fluctuations): The degree of abnormal economic fluctuations refers to the degree of abnormality in the suspect's recent financial transactions (such as large amounts of unexplained income, frequent transfers); X (eXceptional movement trajectory): Abnormality of movement trajectory, which refers to the frequency with which a suspect deviates from their daily activity range and goes to sensitive locations (such as the vicinity of the crime scene); NX (Network complexity): The degree of social complexity refers to the proportion of different industries / risk groups in the suspect's social network. The higher the value, the more complex the social network. N (Prior record nexus): Prior record correlation, which refers to the degree of matching between the suspect / related persons and past cases and prior convictions.

[0155] The different distributions of the blue areas among the different groups reflect the prominence of each group in the corresponding feature dimensions (for example, Cluster 0 is more prominent in features X, E, and J, while Cluster 2 is only prominent in feature E). This figure allows for a visual comparison of the feature differences among the suspect groups, aiding in the analysis of the characteristic features of each group. Based on the clustering results, a radar chart of suspect characteristics is drawn, and the characteristics of each suspect group are analyzed based on the information displayed in the radar chart.

[0156] The analysis results are shown in Table 1 below: Table 1. Examples of Suspect Characteristics and Number of Cases Involved:

[0157] See Figure 11 , Figure 11 This is a schematic diagram of the suspect feature comparison radar in the big data legal supervision model establishment method of the present invention, as shown below. Figure 11 As shown, a radar chart integrating the features of five suspect groups (Cluster 0-4) is presented. The dimensions correspond to J (frequency of contact with suspicious persons), E (degree of abnormal economic fluctuations), X (degree of abnormality in whereabouts), NX (degree of complexity of social relationships), and N (degree of correlation with prior criminal records) in a criminal investigation scenario. Different colored areas in the chart represent different suspect groups, and the coverage of each area in the corresponding dimension intuitively reflects the strength and weakness distribution of different groups in various suspicious features. This chart can be used to quickly compare the feature differences of multiple suspect groups and help identify key suspect groups.

[0158] Step 8: Establishment of a big data legal supervision model.

[0159] Based on the anomaly categories obtained from the cluster analysis in step 6, a big data legal supervision model is established.

[0160] If two abnormal categories, suspect group 2 and suspect group 4, are obtained in step 6 above, an "abnormal legal supervision model for suspects of dangerous driving crimes" can be established.

[0161] The first monitoring point of the model is "the suspect is young at the time of the crime, has a low blood alcohol content, and a short sentence, but the amount of the fine is particularly high". The second monitoring point of the model is "the suspect is young at the time of the crime, has a low blood alcohol content, and a relatively low amount of the fine, but the sentence is particularly high".

[0162] This embodiment, through the above-described scheme, uses the clustered suspect group formed by the optimal clustering results as the basic data. Based on the basic data, a pre-set legal supervision model based on unsupervised learning is generated through statistical analysis. This model includes group size distribution, core feature labels, and risk level assessment. The analysis results of the suspect group are output through the pre-set legal supervision model. This effectively avoids the problems of subjectivity, poor adaptability, and high maintenance costs caused by manually pre-set rules. It significantly improves the objectivity, efficiency, and accuracy of legal supervision, providing the procuratorate with quantifiable group size distribution, core feature labels, and risk level assessment results. This enables the transformation of supervision decision-making from experience-driven to data-driven, eliminating the need for prosecutors to pre-set model rules, reducing the time cost of manual data analysis, lowering the threshold for building big data legal supervision models, improving equipment utilization, reducing the cost of building big data legal supervision models, fully exploring the value of data, avoiding the omission of criminal facts, and effectively combating crime.

[0163] Accordingly, the present invention further provides a device for establishing a big data legal supervision model.

[0164] Reference Figure 12 , Figure 12 This is a functional block diagram of the first embodiment of the big data legal supervision model establishment device of the present invention.

[0165] In the first embodiment of the big data legal supervision model establishment device of the present invention, the big data legal supervision model establishment device includes: The core feature extraction module 10 is used to import legal supervision data and clean missing and outlier values. Based on the principal component analysis algorithm, it performs feature dimensionality reduction on the cleaned target legal data and extracts the core original feature columns from the dimensionality-reduced feature data through principal component load analysis.

[0166] Clustering optimization module 20 is used to perform group clustering based on the Kmeans clustering algorithm using the core original feature columns to obtain the optimal clustering result.

[0167] The model building output module 30 is used to build a preset legal supervision model based on the optimal clustering results and output the analysis results of the suspect group.

[0168] The steps for implementing each functional module of the big data legal supervision model establishment device can be referred to in the various embodiments of the big data legal supervision model establishment method of the present invention, and will not be repeated here.

[0169] Furthermore, this embodiment of the invention also proposes a storage medium storing a big data legal supervision model establishment program. When the big data legal supervision model establishment program is executed by a processor, it implements the operations described in the above-described big data legal supervision model establishment method embodiment.

[0170] Those skilled in the art will understand that all or part of the steps in the methods described above can be implemented by a program instructing related hardware. The program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium is a computer-readable storage medium, including: USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program code.

[0171] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0172] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0173] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for establishing a big data legal supervision model, characterized in that the apparatus, equipment, and storage medium are: The method for establishing the big data legal supervision model includes: Import legal supervision data and clean up missing and outlier values. Perform feature dimensionality reduction on the cleaned target legal data based on principal component analysis algorithm, and extract the core original feature columns from the dimensionality-reduced feature data through principal component loading analysis. The optimal clustering result is obtained by using the core original feature columns based on the Kmeans clustering algorithm to perform population clustering. Based on the optimal clustering results, a pre-defined legal supervision model is established, and the analysis results of the suspect group are output.

2. The method for establishing a big data legal supervision model as described in claim 1, characterized in that, The process involves importing legal supervision data and cleaning missing and outlier values. Principal component analysis (PCA) is used to perform feature dimensionality reduction on the cleaned target legal data. Principal component loading analysis is then used to extract the core original feature columns from the dimensionality-reduced feature data, including: Import the legal supervision dataset to be analyzed, and detect the legal supervision data in the dataset to obtain missing values ​​and outliers; The missing and outlier values ​​in the legal supervision data are removed to obtain the target legal data after data cleaning. The target legal data is dimensionality reduced using principal component analysis algorithm to obtain the dimensionality-reduced feature data. The core original feature columns in the feature data are extracted through principal component loading analysis.

3. The method for establishing a big data legal supervision model as described in claim 2, characterized in that, The step of performing dimensionality reduction on the target legal data using principal component analysis to obtain dimensionality-reduced feature data includes: The target legal data is standardized using the principal component analysis algorithm according to the following formula to obtain standardized data: in, To standardize data, For the first part of the original data The sample at the th Values ​​on each feature For the data involved The crime is Weight values ​​on features For the first The mean of each feature, For the first Standard deviation of each feature; The covariance matrix of the standardized data is calculated using the following formula: in, Let covariance matrix be the variance matrix. For sample size, For the standardized data matrix, This is the matrix transpose. Perform eigenvalue decomposition on the covariance matrix and select the eigenvectors corresponding to the k largest eigenvalues; The target legal data is mapped to a dimensionality-reduced space composed of the feature vectors to obtain the dimensionality-reduced feature data.

4. The method for establishing a big data legal supervision model as described in claim 2, characterized in that, The extraction of core original feature columns from the feature data through principal component loading analysis includes: The loading value of each original feature column in the feature data on the principal components is calculated using the following formula: in, For load values, Main component and original variables The correlation coefficient, For the first Eigenvalues ​​of each principal component; The original feature columns whose absolute value of the load value is greater than a preset load value threshold are taken as the core original feature columns.

5. The method for establishing a big data legal supervision model as described in claim 1, characterized in that, The K-means clustering algorithm uses the core original feature columns to perform group clustering and obtain the optimal clustering result, including: Based on the K-means clustering algorithm, using the core original feature columns as input data, the Euclidean distance between the data object and each cluster center of a preset number of cluster centers is calculated iteratively using the following formula: in, For data objects With the Cluster centers Euclidean distance, For the dimensions of data objects, For data objects The Each attribute value For the first Cluster centers The Each attribute value; Clusters are assigned based on the Euclidean distance, and the average distance between each cluster is calculated. If the average distance satisfies the following formula, the current cluster is discarded: in, For the dimensions of data objects, For data objects The Each attribute value The average distance threshold; Repeat the cluster assignment and average distance calculation process until the average distance of all clusters is less than the average distance threshold. The mean vector of the cluster is updated by the following formula, and the iteration continues until a preset number of iterations is reached or the mean vector no longer changes. in, Cluster The mean vector, For the first Cluster centers, For data objects; The K-means model with different numbers of clusters is systematically scored, and the clustering scheme with the highest score is dynamically selected as the optimal clustering result.

6. The method for establishing a big data legal supervision model as described in claim 5, characterized in that, The process of cluster assignment based on the Euclidean distance, systematically scoring K-means models with different numbers of clusters, and dynamically selecting the clustering scheme with the highest score as the optimal clustering result includes: Clusters are assigned based on the Euclidean distance. For K-means models with different numbers of clusters, the ratio of inter-cluster dispersion to intra-cluster compactness is calculated using the following formula: in, The ratio of inter-cluster dispersion to intra-cluster compactness is the Calinski-Harabasz index, which is the score for a cluster size of K. Inter-cluster dispersion Cluster compactness The total number of samples, The number of clusters; The clustering scheme with the largest ratio is selected as the optimal clustering result.

7. The method for establishing a big data legal supervision model as described in claim 1, characterized in that, The step of establishing a preset legal supervision model based on the optimal clustering results and outputting the analysis results of the suspect group includes: The suspect group formed by the optimal clustering results will be used as the basic data. Based on the aforementioned basic data, a pre-defined legal supervision model based on unsupervised learning is generated through statistical analysis. This model includes group size distribution, core feature labels, and risk level assessment. The analysis results of the suspect group are then output through the pre-defined legal supervision model.

8. A device for establishing a big data legal supervision model, characterized in that, The device for establishing the big data legal supervision model includes: The core feature extraction module is used to import legal supervision data and clean missing and outlier values. It performs feature dimensionality reduction on the cleaned target legal data based on principal component analysis algorithm, and extracts the core original feature columns from the dimensionality-reduced feature data through principal component loading analysis. The clustering optimization module is used to perform population clustering based on the K-means clustering algorithm using the core original feature columns to obtain the optimal clustering result; The model building output module is used to build a preset legal supervision model based on the optimal clustering results and output the analysis results of the suspect group.

9. A device for establishing a big data legal supervision model, characterized in that, The big data legal supervision model establishment device includes: a memory, a processor, and a big data legal supervision model establishment program stored on the memory and capable of running on the processor. The big data legal supervision model establishment program is configured to implement the steps of the big data legal supervision model establishment method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a big data legal supervision model establishment program, which, when executed by a processor, implements the steps of the big data legal supervision model establishment method as described in any one of claims 1 to 7.

Images