Case supervision system and method based on artificial intelligence

Abstract

The invention discloses a case supervision system and method based on artificial intelligence, and the system comprises a data collection module, a case analysis module, an abnormity monitoring module, a risk early warning module and a man-machine interaction module, and improves the safety and efficiency of the case supervision system. A dynamically updated machine learning model is established through artificial intelligence, so that the accuracy and effectiveness of case supervision are ensured, the system adapts to the change and development of the judicial field, and the practicability and sustainability of the system are improved; the risk condition in the case handling process is comprehensively and objectively assessed through a comprehensive risk assessment and early warning mechanism, so that the potential risk is comprehensively found, and the case supervision effect and reliability are improved; and the usability and safety of the system are improved through man-machine interaction and visual display operation, so that decision making and case handling are assisted, comprehensive, intelligent and real-time supervision of cases is realized, and intelligent development in the judicial field is promoted.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based case supervision system and method. Background Art

[0002] Artificial intelligence is an interdisciplinary subject that integrates computer science, cybernetics, information theory, linguistics and other disciplines. It simulates human thinking and behavior through computers. Its core is machine learning algorithms. Based on a huge database, it can automatically link case facts with relevant legal provisions, provide accurate legal application guidance and similar case references, and generate drafts of legal documents based on case elements and given templates, and perform quality verification on the documents to improve the quality and efficiency of document writing.

[0003] However, the existing case supervision system has the defects of low case analysis efficiency, lack of systematic risk assessment, and the resulting untimely and fragmented early warning response;

[0004] Because historical case information and archive access information are scattered across different business systems, such as trial systems and archive management systems, and lack a cross-system data integration mechanism, data fragmentation occurs during supervisory analysis, making it impossible to form a complete business portrait. Furthermore, the assessment of evidence flaws relies on the experience of case handlers and lacks scientific modeling, making it easy to miss hidden risks. Abnormal archive access behavior cannot be detected in real time, relying instead on post-audit findings, such as unauthorized access and batch downloads. This results in delayed risk response and a single early warning response mechanism.

[0005] In view of the above technical defects, a solution is now proposed. Summary of the Invention

[0006] The purpose of the present invention is to solve the defects of the existing case supervision system, such as low case analysis efficiency, lack of systematic risk assessment, and the resulting untimely and fragmented early warning response.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A case supervision method based on artificial intelligence comprises the following steps:

[0009] Step 1: Collect cross-system integrated data: Cross-system integrated data includes historical case information and archive access information;

[0010] Step 2: Process historical case information and generate evidence defect risk signals: Analyze historical case information by establishing a case analysis model, identify case type distribution and high-incidence trends, and conduct in-depth analysis of the distribution of evidence defects and the probability of evidence validity, thereby assessing the degree of evidence defect risk and generating evidence defect risk signals;

[0011] Step 3: Processing file access information and generating abnormal behavior risk signals: By establishing an abnormal monitoring model to analyze file access information, identify abnormal file access operations, and thus assess the risk level of abnormal behavior and generate abnormal behavior risk signals;

[0012] Step 4: Establish a risk warning model, receive risk signal groups, and conduct warning processing: By combining the risk level of evidence flaws and the risk level of abnormal behavior, a comprehensive analysis of the potential case handling risk level is conducted to generate potential case handling risk signals. The evidence flaw risk signals, abnormal behavior risk signals, and potential case handling risk signals are integrated and marked as risk signal groups. By receiving the risk signal groups, corresponding real-time warnings are issued.

[0013] Step 5: Store and display warning information: The warning information includes the initial value of the cross-system integrated data and the processed data analysis results.

[0014] Furthermore, the specific process of collecting cross-system integrated data is as follows:

[0015] Calling historical case information through the standardized interface of the judicial business management system, which includes the trial system, procuratorial system, agency case handling system, and archive management system; and collecting archive access information through the archive access log of the archive management system;

[0016] Establish machine learning models to analyze and process cross-system integrated data. Machine learning models include case analysis models, anomaly monitoring models, and risk warning models.

[0017] Historical case information includes case type, evidence type, and evidence defect type;

[0018] Archive access information includes access behavior characteristics, user rank, and departmental authority level.

[0019] Furthermore, the case analysis model includes a case type classification sub-model, a case high incidence trend prediction sub-model, an evidence flaw identification sub-model and an evidence validity assessment sub-model;

[0020] The case type classification sub-model is used to classify and label cases and obtain the distribution of case types;

[0021] The case high incidence trend prediction sub-model is used to predict the high incidence trend of case types;

[0022] The evidence flaw identification sub-model is used to identify the flaw patterns of case evidence and obtain the distribution of evidence flaws;

[0023] The evidence validity assessment sub-model is used to evaluate the validity probability of case evidence;

[0024] Further, the abnormality monitoring model comprises a sensitive case identification sub-model, an access mode learning sub-model, and an abnormal operation detection sub-model.

[0025] Further, the abnormality monitoring model comprises a sensitive case identification sub-model, an access mode learning sub-model, and an abnormal operation detection sub-model.

[0026] The sensitive case identification sub-model is configured to identify sensitive cases.

[0027] The access mode learning sub-model is configured to learn normal access modes.

[0028] The abnormal operation detection sub-model is configured to detect abnormal operations through normal access modes, wherein the abnormal operations include high-frequency retrieval of sensitive cases, unauthorized access, high-frequency access outside working hours, and cross-department unauthorized retrieval.

[0029] The abnormal operation detection sub-model is configured to detect abnormal operations through normal access modes, wherein the abnormal operations include high-frequency retrieval of sensitive cases, unauthorized access, high-frequency access outside working hours, and cross-department unauthorized retrieval.

[0030] Further, the risk early warning model comprises an evidence flaw early warning sub-model, an abnormal access early warning sub-model, and a comprehensive risk assessment sub-model.

[0031] The evidence flaw early warning sub-model is configured to receive the evidence flaw risk signal and perform real-time early warning on the evidence flaw risk.

[0032] The abnormal access early warning sub-model is configured to receive the abnormal behavior risk signal and perform real-time early warning on the abnormal access behavior.

[0033] The comprehensive risk assessment sub-model is configured to combine the evidence flaw risk degree and the abnormal behavior risk degree, analyze the potential case risk degree comprehensively, generate a potential case risk signal, and perform comprehensive risk early warning.

[0034] Further, the specific processing process of the case analysis model is as follows:

[0035] S2-1, the steps of the case type classification sub-model are: inputting case features, wherein the case features include case types, case description texts, evidence types, evidence flaw types, time stamps, and region codes;

[0036] Set and mark any case type as a, and mark the number of case types as N0;

[0037] Set a predefined vocabulary table and establish a word vector matrix, convert the case description text features into numerical vectors through word embedding, and then perform one-hot encoding on the case type features, analyze the mapping relationship between the case description text features and the case type features, and output the classification probability of N0 case types;

[0038] S2-2, the steps of the case high-incidence trend prediction sub-model are as follows: input time series data, which includes case type, timestamp and area code;

[0039] Count case types by timestamp and generate a case quantity fluctuation curve chart. From the case quantity fluctuation curve chart, obtain the historical baseline of the case quantity.

[0040] The LSTM time series prediction model is used to perform regression fitting on the case type quantity fluctuation curve to output the predicted quantity Ya for case type a. The deviation between the predicted quantity Ya for case type a and the historical quantity baseline is calculated to obtain the confidence level φa for the predicted quantity Ya for case type a. The predicted quantity Ya and confidence level φa for case type a are then output.

[0041] S2-3, the steps of the evidence defect identification sub-model are as follows: input evidence features and case type features, where the evidence features include evidence type, evidence defect type, evidence description text, and evidence collection time;

[0042] Set and mark any evidence flaw pattern as b, and establish the cluster center μb of the evidence flaw pattern b through the K-means clustering algorithm;

[0043] Mark the feature vector of the evidence to be identified as xi, obtain and output the predicted probability Pb ​​of the evidence flaw pattern being determined to be b;

[0044] S2-4, the steps of the evidence validity assessment sub-model are: input evidence characteristics, including defect type, evidence reliability score, evidence submission time delay, and case type;

[0045] Establish a structured feature vector and obtain the probability of evidence validity Pc by combining the evidence reliability score C1 and the evidence submission time delay C2;

[0046] Then, the potential risk of evidence flaws is analyzed through the probability of evidence validity Pc, thereby evaluating the degree of evidence flaw risk and generating an evidence flaw risk signal;

[0047] Set the evaluation interval of the probability of evidence validity Pc, determine the risk of evidence flaws by comparing the intervals, thereby evaluating the degree of evidence flaw risk and generating an evidence flaw risk signal.

[0048] Furthermore, the specific processing process of the anomaly monitoring model is as follows:

[0049] S3-1, the steps of the sensitive case identification sub-model are as follows: input sensitive case characteristics, including the case confidentiality level, periodic visit volume and social attention, and obtain the sensitivity score D;

[0050] Set the threshold Df of the sensitivity score D: When the sensitivity score D is higher than the threshold Df, the case's sensitive label E is determined to be a sensitive case; conversely, when the sensitivity score D is lower than or equal to the threshold Df, the case's sensitive label E is determined to be a non-sensitive case;

[0051] S3-2, the steps of the access pattern learning sub-model are: input historical access requests, access behavior characteristics, and user profiles; access behavior characteristics include periodic access volume and the proportion of access during non-working hours; user profiles include user rank and departmental authority level;

[0052] Set a normal mode baseline for access behavior characteristics and obtain a confidence interval using the 3σ principle to determine the rationality of access requests.

[0053] S3-3, the steps of the abnormal operation detection sub-model are:

[0054] By accumulating the number of sensitive case calls within a unit period of working time and setting a threshold, when it exceeds the threshold, it is determined that sensitive case calls are frequently made;

[0055] By comparing the user profile with the permissions required for the case, if the user's rank and departmental permissions are lower than the permissions required for the case, it is determined to be an unauthorized access operation;

[0056] By accumulating the number of case calls in a unit cycle during non-working hours and setting a threshold, when it exceeds the threshold, it is determined to be a high-frequency access operation during non-working hours;

[0057] Compare the user's department with the case's department, and if the two do not match, it will be determined as an unauthorized cross-departmental access operation;

[0058] By combining the number of sensitive case operations, unauthorized access operations, non-working hours access operations, and cross-departmental unauthorized access operations, we can obtain a weighted abnormal behavior risk coefficient Gq, thereby assessing the degree of abnormal behavior risk and generating an abnormal behavior risk signal.

[0059] Set the evaluation interval of the abnormal behavior risk coefficient Gq, determine the abnormal behavior risk by comparing the intervals, thereby evaluating the degree of abnormal behavior risk and generating an abnormal behavior risk signal.

[0060] Furthermore, the specific processing process of the risk warning model is as follows:

[0061] S4-1, the steps of the evidence defect warning sub-model are: receiving the evidence defect risk signal and performing corresponding evidence defect warning processing according to the level of the evidence defect risk signal;

[0062] S4-2, the steps of the abnormal access warning sub-model are: receiving abnormal behavior risk signals and performing corresponding abnormal behavior warning processing according to the level of abnormal behavior risk signals;

[0063] S4-3, the steps of the comprehensive risk assessment sub-model are:

[0064] By combining the probability of evidence validity Pc and the abnormal behavior risk coefficient Gq, the case risk index Risk is obtained;

[0065] Set the evaluation range of the case risk index Risk, and comprehensively determine the degree of potential case handling risk through interval comparison, thereby generating a potential case handling risk signal of the corresponding level.

[0066] An artificial intelligence-based case supervision system includes a data acquisition module, a case analysis module, an anomaly monitoring module, a risk warning module, and a human-computer interaction module. When used, the system implements the above-mentioned artificial intelligence-based case supervision method;

[0067] The data acquisition module is used to collect cross-system integrated data:

[0068] The case analysis module is used to process historical case information to generate evidence flaw risk signals;

[0069] The anomaly monitoring module is used to process archive access information to generate abnormal behavior risk signals;

[0070] The risk warning module is used to group risk signals and perform warning processing;

[0071] The human-computer interaction module is used to store and display warning information.

[0072] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0073] The present invention establishes a dynamically updated machine learning model through artificial intelligence to ensure the accuracy and effectiveness of case supervision, adapt to changes and developments in the judicial field, and improve the practicality and sustainability of the system; furthermore, through a comprehensive risk assessment and early warning mechanism, it comprehensively and objectively assesses the risk status in the case handling process, thereby comprehensively discovering potential risks and improving the effectiveness and reliability of case supervision; and through human-computer interaction and visual display operations, it improves the usability and security of the system, thereby assisting decision-making and case handling, realizing comprehensive, intelligent, and real-time supervision of cases, and promoting the intelligent development of the judicial field.

[0074] The present invention uses a case analysis model to conduct cross-system data fusion and intelligent analysis to assess the risk level of evidence flaws. It can automatically discover hidden patterns and regularities from massive amounts of historical case data, providing data support for judicial decision-making and case handling process optimization, overcoming the shortcomings of existing technologies that rely on manual analysis, resulting in low efficiency and accuracy.

[0075] The present invention uses an anomaly monitoring model to perform multi-dimensional abnormal behavior identification and real-time warning, thereby identifying abnormal operations such as high-frequency retrieval of sensitive cases, unauthorized access, high-frequency access during non-working hours, and cross-departmental unauthorized retrieval. It can comprehensively and accurately detect abnormal behavior in file access, providing strong support for ensuring the security and compliance of judicial archives.

[0076] The present invention uses a risk warning model to provide real-time warning of potential case handling risks, realizes real-time monitoring and warning of the risks of evidence defects and file access risks in the case handling process, can timely discover potential case handling risks, provide decision-making support for judicial personnel, and improve the quality and efficiency of case handling. In addition, the system can continuously optimize model performance as judicial data continues to accumulate and judicial business continues to develop. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 shows a schematic diagram of the connection of the system modules of the present invention;

[0078] Figure 2 A schematic diagram showing the steps of the overall process of the present invention is shown;

[0079] Figure 3 A flow chart of the case analysis model of the present invention is shown. DETAILED DESCRIPTION

[0080] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0081] Example 1:

[0082] like Figure 1-3 As shown, an artificial intelligence-based case supervision system includes a data acquisition module, a case analysis module, an anomaly monitoring module, a risk warning module and a human-computer interaction module, wherein the data acquisition module, the case analysis module, the anomaly monitoring module, the risk warning module and the human-computer interaction module are communicatively connected;

[0083] The data acquisition module is used to collect cross-system integrated data;

[0084] The case analysis module is used to process historical case information to generate evidence flaw risk signals;

[0085] The anomaly monitoring module is used to process archive access information to generate abnormal behavior risk signals;

[0086] The risk warning module is used to group risk signals and perform warning processing;

[0087] The human-computer interaction module is used to store and display warning information.

[0088] The system working steps are as follows:

[0089] S1, collect cross-system integrated data: Cross-system integrated data includes historical case information and archive access information;

[0090] The specific process of collecting cross-system integration data is as follows:

[0091] Calling historical case information through the standardized interface of the judicial business management system, which includes the trial system, procuratorial system, agency case handling system, and archive management system; and collecting archive access information through the archive access log of the archive management system;

[0092] Establish machine learning models through artificial intelligence technology to analyze and process cross-system integrated data. Machine learning models include case analysis models, anomaly monitoring models, and risk warning models.

[0093] The machine learning model uses deep learning algorithms, including one or more of convolutional neural networks, recurrent neural networks, and graph neural networks. By updating and optimizing the machine learning model data, the accuracy and adaptability of the model are improved.

[0094] Historical case information includes case type, evidence type, and evidence defect type;

[0095] Archive access information includes access behavior characteristics, user rank, and departmental authority level.

[0096] S2, Processing Historical Case Information and Generating Evidence Flaw Risk Signals: Analyzing historical case information through the establishment of a case analysis model, identifying case type distribution and high-incidence trends, and conducting in-depth analysis of the distribution of evidence flaws and the probability of evidence validity, thereby assessing the degree of evidence flaw risk and generating evidence flaw risk signals;

[0097] The case analysis model includes a case type classification sub-model, a case high incidence trend prediction sub-model, an evidence flaw identification sub-model and an evidence validity assessment sub-model;

[0098] The case type classification sub-model is used to classify and label cases and obtain the distribution of case types;

[0099] The case high incidence trend prediction sub-model is used to predict the high incidence trend of case types;

[0100] The evidence flaw identification sub-model is used to identify the flaw patterns of case evidence and obtain the distribution of evidence flaws;

[0101] The evidence validity assessment sub-model is used to evaluate the validity probability of case evidence;

[0102] Then, through the validity probability of the case evidence, the potential risk of evidence flaws is analyzed to evaluate the degree of evidence flaw risk and generate an evidence flaw risk signal.

[0103] The specific processing process of the case analysis model is as follows:

[0104] S2-1, the steps of the case type classification sub-model are: input case features, which include case type, case description text, evidence type, evidence defect type, timestamp and region code;

[0105] Among them, case types include contract disputes, case description text includes case summary and keywords; evidence types include physical evidence, documentary evidence and electronic evidence; evidence defect types include tampering, missing and unknown source; timestamp refers to the time when the case occurred; area code refers to the code of the location where the case occurred;

[0106] Set and mark any case type as a, and mark the number of case types as N0;

[0107] Set a predefined vocabulary and build a word vector matrix. Convert the case description text features into numerical vectors through word embedding. Then perform one-hot encoding on the case type features. By analyzing the mapping relationship between the case description text features and the case type features, output the classification probabilities of N0 case types. Specifically,

[0108] The maximum posterior probability estimate of case type a is Pa: Pa = argmax a P(c|fes), where fes is the case description text feature;

[0109] S2-2, the steps of the case high-incidence trend prediction sub-model are as follows: input time series data, which includes case type, timestamp and area code;

[0110] Count case types by timestamp and generate a case quantity fluctuation curve chart. From the case quantity fluctuation curve chart, obtain the historical baseline of the case quantity.

[0111] The LSTM time series prediction model is used to perform regression fitting on the case type quantity fluctuation curve to output the predicted quantity Ya of case type a. The deviation between the predicted quantity Ya of case type a and the historical quantity baseline is calculated to obtain the confidence level φa of the predicted quantity Ya of case type a. The predicted quantity Ya and confidence level φa of case type a are output as follows: Y0a is the data value of case type a in the historical baseline;

[0112] S2-3, the steps of the evidence defect identification sub-model are as follows: input evidence features and case type features, where the evidence features include evidence type, evidence defect type, evidence description text, and evidence collection time;

[0113] Set and mark any evidence flaw pattern as b, and establish the cluster center μb of the evidence flaw pattern b through the K-means clustering algorithm;

[0114] Mark the feature vector of the evidence to be identified as xi, obtain and output the predicted probability Pb ​​of the evidence flaw pattern being b: Pb = argmin b ∑||xi-μb|| 2 ;

[0115] S2-4, the steps of the evidence validity assessment sub-model are: input evidence characteristics, including defect type, evidence reliability score, evidence submission time delay, and case type;

[0116] Establish a structured feature vector, in which the evidence reliability score is obtained by assigning values ​​based on the collector's qualifications and manual experience;

[0117] The probability of evidence validity Pc is obtained by combining the evidence reliability score C1 and the evidence submission time delay C2: Pc = sigmoid(λ1*C1-λ2*C2), where λ1 and λ2 are the weight coefficients of the evidence reliability score C1 and the evidence submission time delay C2, respectively. The weight coefficients are used for dimensionless calculations and are obtained by pre-setting and assigning values ​​after calculations based on a large amount of experimental data.

[0118] Then, the potential risk of evidence flaws is analyzed through the probability of evidence validity Pc, thereby evaluating the degree of evidence flaw risk and generating an evidence flaw risk signal;

[0119] Set the evaluation interval of the probability of evidence validity Pc, determine the risk of evidence flaws by comparing the intervals, thereby evaluating the degree of evidence flaw risk and generating an evidence flaw risk signal;

[0120] Among them, when the probability of evidence validity Pc is lower, the risk degree of evidence flaw is determined to be higher, and the level of evidence flaw risk signal generated is higher.

[0121] S3, processing archive access information and generating abnormal behavior risk signals: By establishing an abnormal monitoring model to analyze archive access information, identify abnormal archive access operations, thereby assessing the risk level of abnormal behavior and generating abnormal behavior risk signals;

[0122] The anomaly monitoring model includes a sensitive case identification sub-model, an access pattern learning sub-model and an abnormal operation detection sub-model;

[0123] Sensitive case identification sub-model, used to identify sensitive cases;

[0124] Access pattern learning sub-model, used to learn normal access patterns;

[0125] The abnormal operation detection sub-model is used to identify abnormal operations through normal access pattern detection. Abnormal operations include frequent access to sensitive cases, unauthorized access, frequent access during non-working hours, and cross-departmental unauthorized access.

[0126] By identifying the results of abnormal operations, the risk level of abnormal behavior is assessed and abnormal behavior risk signals are generated.

[0127] The specific processing process of the anomaly monitoring model is as follows:

[0128] S3-1, the steps of the sensitive case identification sub-model are as follows: input sensitive case characteristics, including the case confidentiality level, periodic visit volume, and social attention; the case confidentiality level L includes public, secret, confidential, and top secret, and the case confidentiality level L of public, secret, confidential, and top secret cases is assigned values ​​of 0, 1, 2, and 3 respectively;

[0129] Get the sensitivity score D: D = (1 + η * M) * (ω1 * L + ω2 * W);

[0130] Where M is the degree of public attention, and M∈[0,1]. Public attention is weighted and comprehensively evaluated by the number of media publicity views and discussions. W is the number of visits to the case period. ω1 and ω2 are the weighting factors of the case's confidentiality level L and the number of visits to the case period W, respectively. η is the adjustment coefficient, and the adjustment coefficient and weighting factor are obtained by default. Ultimately, the sensitivity score D is in the range of [0,1].

[0131] Set the threshold Df of the sensitivity score D: When the sensitivity score D is higher than the threshold Df, the case's sensitive label E is determined to be a sensitive case; conversely, when the sensitivity score D is lower than or equal to the threshold Df, the case's sensitive label E is determined to be a non-sensitive case;

[0132] S3-2, the steps of the access pattern learning sub-model are: input historical access requests, access behavior characteristics, and user profiles; access behavior characteristics include periodic access volume and the proportion of access during non-working hours; user profiles include user rank and departmental authority level;

[0133] Set a normal mode baseline for access behavior characteristics and obtain a confidence interval using the 3σ principle to determine the rationality of access requests.

[0134] S3-3, the steps of the abnormal operation detection sub-model are:

[0135] Abnormal operations include frequent access to sensitive cases, unauthorized access, frequent access during non-working hours, and cross-departmental unauthorized access.

[0136] By accumulating the number of sensitive case calls within a unit period of working time and setting a threshold, when it exceeds the threshold, it is determined that sensitive case calls are frequently made;

[0137] By comparing the user profile with the permissions required for the case, if the user's rank and departmental permissions are lower than the permissions required for the case, it is determined to be an unauthorized access operation;

[0138] By accumulating the number of case calls in a unit cycle during non-working hours and setting a threshold, when it exceeds the threshold, it is determined to be a high-frequency access operation during non-working hours;

[0139] Compare the user's department with the case's department, and if the two do not match, it will be determined as an unauthorized cross-departmental access operation;

[0140] By combining the number of sensitive case operations, unauthorized access operations, non-working hours access operations, and cross-departmental unauthorized access operations, we can obtain a weighted abnormal behavior risk coefficient Gq, thereby assessing the degree of abnormal behavior risk and generating an abnormal behavior risk signal.

[0141] Set the evaluation interval of the abnormal behavior risk coefficient Gq, determine the abnormal behavior risk by comparing the intervals, thereby evaluating the abnormal behavior risk level and generating an abnormal behavior risk signal;

[0142] Among them, when the abnormal behavior risk coefficient Gq is higher, the abnormal behavior risk degree is determined to be higher, and the level of the generated abnormal behavior risk signal is higher.

[0143] S4, establish a risk warning model, receive risk signal groups and conduct warning processing: by combining the risk level of evidence defects and the risk level of abnormal behavior, comprehensively analyze the potential case handling risk level, generate potential case handling risk signals, and integrate the evidence defect risk signals, abnormal behavior risk signals and potential case handling risk signals into a risk signal group. By receiving the risk signal group, corresponding real-time warnings can be issued;

[0144] The risk early warning model comprises an evidence flaw early warning sub-model, an abnormal access early warning sub-model, and a comprehensive risk assessment sub-model.

[0145] The evidence flaw early warning sub-model is configured to receive evidence flaw risk signals and perform real-time early warning on evidence flaw risks.

[0146] The abnormal access early warning sub-model is configured to receive abnormal behavior risk signals and perform real-time early warning on abnormal access behaviors.

[0147] The comprehensive risk assessment sub-model is configured to combine the evidence flaw risk degree and the abnormal behavior risk degree to comprehensively analyze the potential case handling risk degree, generate potential case handling risk signals, and perform comprehensive risk early warning.

[0148] The specific processing procedure of the risk early warning model is as follows:

[0149] S4-1, the procedure of the evidence flaw early warning sub-model is to receive evidence flaw risk signals and perform corresponding evidence flaw early warning processing according to the levels of the evidence flaw risk signals.

[0150] S4-2, the procedure of the abnormal access early warning sub-model is to receive abnormal behavior risk signals and perform corresponding abnormal behavior early warning processing according to the levels of the abnormal behavior risk signals.

[0151] S4-3, the procedure of the comprehensive risk assessment sub-model is:

[0152] The case risk index Risk is obtained by combining the evidence effectiveness probability Pc and the abnormal behavior risk coefficient Gq. The case risk index Risk is higher when the evidence effectiveness probability Pc is lower and the abnormal behavior risk coefficient Gq is higher.

[0153] The evaluation interval of the case risk index Risk is set, the potential case handling risk degree is comprehensively determined by interval comparison, and the potential case handling risk signals of corresponding levels are generated.

[0154] When the case risk index Risk is higher, it is determined that the potential case handling risk degree is higher, and the level of the potential case handling risk signals is set to be higher.

[0155] S5, the early warning information is stored and displayed, the early warning information comprises initial values and processed data analysis results of cross-system integrated data, the data analysis results comprise the evidence flaw risk degree, the abnormal behavior risk degree, and the potential case handling risk degree, and background professional management personnel perform corresponding emergency plan configuration according to different risk categories and degrees.

[0156] In summary, the present invention establishes a dynamically updated machine learning model through artificial intelligence to ensure the accuracy and effectiveness of case supervision, adapt to changes and developments in the judicial field, and improve the practicality and sustainability of the system; then, through a comprehensive risk assessment and early warning mechanism, it comprehensively and objectively evaluates the risk status in the case handling process, thereby comprehensively discovering potential risks and improving the effectiveness and reliability of case supervision; and through human-computer interaction and visual display operations, it improves the ease of use and security of the system, thereby assisting decision-making and case handling, realizing comprehensive, intelligent, and real-time supervision of cases, and promoting the intelligent development of the judicial field.

[0157] The present invention uses a case analysis model to conduct cross-system data fusion and intelligent analysis to assess the risk level of evidence flaws. It can automatically discover hidden patterns and regularities from massive amounts of historical case data, providing data support for judicial decision-making and case handling process optimization, overcoming the shortcomings of existing technologies that rely on manual analysis, resulting in low efficiency and accuracy.

[0158] The present invention uses an anomaly monitoring model to perform multi-dimensional abnormal behavior identification and real-time warning, thereby identifying abnormal operations such as high-frequency retrieval of sensitive cases, unauthorized access, high-frequency access during non-working hours, and cross-departmental unauthorized retrieval. It can comprehensively and accurately detect abnormal behavior in file access, providing strong support for ensuring the security and compliance of judicial archives.

[0159] The present invention uses a risk warning model to provide real-time warning of potential case handling risks, realizes real-time monitoring and warning of the risks of evidence defects and file access risks in the case handling process, can timely discover potential case handling risks, provide decision-making support for judicial personnel, and improve the quality and efficiency of case handling. In addition, the system can continuously optimize model performance as judicial data continues to accumulate and judicial business continues to develop.

[0160] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0161] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0162] The above data processing is to remove the dimension and take its numerical calculation. The setting of the interval and threshold size is for the convenience of comparison. The size of the threshold depends on the amount of sample data and the cardinality set by technical personnel in this field for each group of sample data. As long as it does not affect the proportional relationship between the parameter and the quantized value, the preset parameters are set by technical personnel in this field according to actual conditions.

[0163] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A case supervision method based on artificial intelligence, characterized by: The following steps are involved: Step 1: Collect cross-system integrated data: Cross-system integrated data includes historical case information and archive access information; Step 2: Process historical case information and generate evidence defect risk signals: Analyze historical case information by establishing a case analysis model, identify case type distribution and high-incidence trends, and conduct in-depth analysis of the distribution of evidence defects and the probability of evidence validity, thereby assessing the degree of evidence defect risk and generating evidence defect risk signals; Step 3: Processing file access information and generating abnormal behavior risk signals: By establishing an abnormal monitoring model to analyze file access information, identify abnormal file access operations, and thus assess the risk level of abnormal behavior and generate abnormal behavior risk signals; Step 4: Establish a risk warning model, receive risk signal groups, and conduct warning processing: By combining the risk level of evidence flaws and the risk level of abnormal behavior, a comprehensive analysis of the potential case handling risk level is conducted to generate potential case handling risk signals. The evidence flaw risk signals, abnormal behavior risk signals, and potential case handling risk signals are integrated and marked as risk signal groups. By receiving the risk signal groups, corresponding real-time warnings are issued. Step 5: Store and display warning information: The warning information includes the initial value of the cross-system integrated data and the processed data analysis results.

2. The case supervision method based on artificial intelligence according to claim 1, characterized in that: The specific process of collecting cross-system integration data is as follows: Calling historical case information through the standardized interface of the judicial business management system, which includes the trial system, procuratorial system, agency case handling system, and archive management system; and collecting archive access information through the archive access log of the archive management system; Establish machine learning models to analyze and process cross-system integrated data. Machine learning models include case analysis models, anomaly monitoring models, and risk warning models. Historical case information includes case type, evidence type, and evidence defect type; Archive access information includes access behavior characteristics, user rank, and departmental authority level.

3. The case supervision method based on artificial intelligence according to claim 2, characterized in that: The case analysis model includes a case type classification sub-model, a case high incidence trend prediction sub-model, an evidence flaw identification sub-model and an evidence validity assessment sub-model; The case type classification sub-model is used to classify and label cases and obtain the distribution of case types; The case high incidence trend prediction sub-model is used to predict the high incidence trend of case types; The evidence flaw identification sub-model is used to identify the flaw patterns of case evidence and obtain the distribution of evidence flaws; The evidence validity assessment sub-model is used to evaluate the validity probability of case evidence; Then, through the validity probability of the case evidence, the potential risk of evidence flaws is analyzed to evaluate the degree of evidence flaw risk and generate an evidence flaw risk signal.

4. The case supervision method based on artificial intelligence according to claim 3, characterized in that: The anomaly monitoring model includes a sensitive case identification sub-model, an access pattern learning sub-model and an abnormal operation detection sub-model; Sensitive case identification sub-model, used to identify sensitive cases; Access pattern learning sub-model, used to learn normal access patterns; The abnormal operation detection sub-model is used to identify abnormal operations through normal access pattern detection. Abnormal operations include frequent access to sensitive cases, unauthorized access, frequent access during non-working hours, and cross-departmental unauthorized access. By identifying the results of abnormal operations, the risk level of abnormal behavior is assessed and abnormal behavior risk signals are generated.

5. The case supervision method based on artificial intelligence according to claim 4, characterized in that: The risk warning model includes an evidence flaw warning sub-model, an abnormal access warning sub-model and a comprehensive risk assessment sub-model; The evidence flaw warning sub-model is used to receive evidence flaw risk signals and provide real-time warnings on evidence flaw risks; The abnormal access warning sub-model is used to receive abnormal behavior risk signals and issue real-time warnings for abnormal access behaviors; The comprehensive risk assessment sub-model is used to combine the risk level of evidence flaws and the risk level of abnormal behavior, so as to comprehensively analyze the potential case handling risk level, generate potential case handling risk signals, and conduct comprehensive risk warnings.

6. The case supervision method based on artificial intelligence according to claim 5, characterized in that: The specific processing process of the case analysis model is as follows: S2-1, the steps of the case type classification sub-model are: input case features, which include case type, case description text, evidence type, evidence defect type, timestamp and region code; Set and mark any case type as a, and mark the number of case types as N0; Set a predefined vocabulary and build a word vector matrix. Convert the case description text features into numerical vectors through word embedding. Then, perform one-hot encoding on the case type features. By analyzing the mapping relationship between the case description text features and the case type features, output the classification probabilities of N0 case types. S2-2, the steps of the case high-incidence trend prediction sub-model are as follows: input time series data, which includes case type, timestamp and area code; Count case types by timestamp and generate a case quantity fluctuation curve chart. From the case quantity fluctuation curve chart, obtain the historical baseline of the case quantity. The LSTM time series prediction model is used to perform regression fitting on the case type quantity fluctuation curve to output the predicted quantity Ya for case type a. The deviation between the predicted quantity Ya for case type a and the historical quantity baseline is calculated to obtain the confidence level φa for the predicted quantity Ya for case type a. The predicted quantity Ya and confidence level φa for case type a are then output. S2-3, the steps of the evidence defect identification sub-model are as follows: input evidence features and case type features, where the evidence features include evidence type, evidence defect type, evidence description text, and evidence collection time; Set and mark any evidence flaw pattern as b, and establish the cluster center μb of the evidence flaw pattern b through the K-means clustering algorithm; Mark the feature vector of the evidence to be identified as xi, obtain and output the predicted probability Pb ​​of the evidence flaw pattern being determined to be b; S2-4, the steps of the evidence validity assessment sub-model are: input evidence characteristics, including defect type, evidence reliability score, evidence submission time delay, and case type; Establish a structured feature vector and obtain the probability of evidence validity Pc by combining the evidence reliability score C1 and the evidence submission time delay C2; Then, the potential risk of evidence flaws is analyzed through the probability of evidence validity Pc, thereby evaluating the degree of evidence flaw risk and generating an evidence flaw risk signal; Set the evaluation interval of the probability of evidence validity Pc, determine the risk of evidence flaws by comparing the intervals, thereby evaluating the degree of evidence flaw risk and generating an evidence flaw risk signal.

7. The case supervision method based on artificial intelligence according to claim 6, characterized in that: The specific processing process of the anomaly monitoring model is as follows: S3-1, the steps of the sensitive case identification sub-model are as follows: input sensitive case characteristics, including the case confidentiality level, periodic visit volume and social attention, and obtain the sensitivity score D; Set the threshold Df of the sensitivity score D: When the sensitivity score D is higher than the threshold Df, the case’s sensitivity label E is determined to be a sensitive case; On the contrary, when the sensitivity score D is lower than or equal to the threshold Df, the sensitive label E of the case is determined to be a non-sensitive case; S3-2, the steps of the access pattern learning sub-model are: input historical access requests, access behavior characteristics, and user profiles; access behavior characteristics include periodic access volume and the proportion of access during non-working hours; user profiles include user rank and departmental authority level; Set a normal mode baseline for access behavior characteristics and obtain a confidence interval using the 3σ principle to determine the rationality of access requests. S3-3, the steps of the abnormal operation detection sub-model are: By accumulating the number of sensitive case calls within a unit period of working time and setting a threshold, when it exceeds the threshold, it is determined that sensitive case calls are frequently made; By comparing the user profile with the permissions required for the case, if the user's rank and departmental permissions are lower than the permissions required for the case, it is determined to be an unauthorized access operation; By accumulating the number of case calls in a unit cycle during non-working hours and setting a threshold, when it exceeds the threshold, it is determined to be a high-frequency access operation during non-working hours; Compare the user's department with the case's department, and if the two do not match, it will be determined as an unauthorized cross-departmental access operation; By combining the number of sensitive case operations, unauthorized access operations, non-working hours access operations, and cross-departmental unauthorized access operations, we can obtain a weighted abnormal behavior risk coefficient Gq, thereby assessing the degree of abnormal behavior risk and generating an abnormal behavior risk signal. Set the evaluation interval of the abnormal behavior risk coefficient Gq, determine the abnormal behavior risk by comparing the intervals, thereby evaluating the degree of abnormal behavior risk and generating an abnormal behavior risk signal.

8. The case supervision method based on artificial intelligence according to claim 7, characterized in that: The specific processing process of the risk warning model is as follows: S4-1, the steps of the evidence defect warning sub-model are: receiving the evidence defect risk signal and performing corresponding evidence defect warning processing according to the level of the evidence defect risk signal; S4-2, the steps of the abnormal access warning sub-model are: receiving abnormal behavior risk signals and performing corresponding abnormal behavior warning processing according to the level of abnormal behavior risk signals; S4-3, the steps of the comprehensive risk assessment sub-model are: By combining the probability of evidence validity Pc and the abnormal behavior risk coefficient Gq, the case risk index Risk is obtained; Set the evaluation range of the case risk index Risk, and comprehensively determine the degree of potential case handling risk through interval comparison, thereby generating a potential case handling risk signal of the corresponding level.

9. An artificial intelligence-based case supervision system, characterized by: The system comprises a data acquisition module, a case analysis module, an anomaly monitoring module, a risk warning module and a human-computer interaction module. When used, the system executes an artificial intelligence-based case supervision method according to any one of claims 1 to 8 above; The data acquisition module is used to collect cross-system integrated data: The case analysis module is used to process historical case information to generate evidence flaw risk signals; The anomaly monitoring module is used to process archive access information to generate abnormal behavior risk signals; The risk warning module is used to group risk signals and perform warning processing; The human-computer interaction module is used to store and display warning information.

Images