Atomic operation and assembly method for big data law supervision model

Abstract

The invention discloses an atomic operation and assembly method for a big data legal supervision model, and the method comprises an atomic operation layer which is used for achieving the basic functions in a legal supervision process; the supervision model layer comprises a plurality of legal supervision models for different supervision targets; and the early warning pushing layer is used for sending the early warning information. According to the method, preconditions and postconditions of atomic operations are defined, a logical relationship between the atomic operations is established by utilizing predicate logic, and then a complete legal supervision model is constructed. According to the method based on predicate logic, the model has certain interpretability, and meanwhile, the flexibility and generalization of the model are enhanced by dynamically assembling atomic operation. According to the method, the formal description capability of the predicate logic is utilized, the logic association among the legal threshold, the supervised object behavior and the supervised target is established, the refined analysis and dynamic evaluation of the supervised object behavior are realized, and the visualization of the atomic operation path capable of reaching the supervised target is displayed.

Description

Technical Field

[0001] The present invention relates to the technical field of legal supervision models, and in particular to the application of big data in legal supervision models. Background Art

[0002] With the continuous development of information technology, big data technology has been widely used in various fields. In the field of legal supervision, the application of big data technology has also attracted widespread attention. Compared with traditional manual supervision, big data legal supervision has stronger coverage, real-time and accuracy, and can better discover abnormal behaviors and potential risks hidden in massive data.

[0003] The existing big data legal supervision models mainly include the following representative solutions:

[0004] 1. Big data legal supervision model based on machine learning

[0005] Such models usually use a large amount of historical regulatory data accumulated by regulatory agencies, combined with natural language processing, graph mining and other technologies to build regulatory early warning models. For example, some models will identify keywords or patterns that may indicate violations by analyzing unstructured data such as company announcements and news reports, and issue early warnings for these risks. Other models will use machine learning algorithms such as anomaly detection and clustering based on structured data such as corporate financial data and transaction records to detect potential violations. Such machine learning-based models have improved the coverage and timeliness of supervision to a certain extent, but they also face some challenges, such as the lack of interpretability of the model and susceptibility to data bias.

[0006] 2. Big data legal supervision model based on knowledge graph

[0007] This type of model builds a knowledge graph covering multiple fields of knowledge such as legal norms, regulatory policies, and corporate information, and uses graph reasoning, semantic analysis and other technologies to discover complex legal risk associations. Compared with a single machine learning model, this knowledge graph-based approach has stronger interpretability and reasoning capabilities. However, it also has problems such as high knowledge acquisition costs and complex graph construction, which limit its practical application.

[0008] 3. Big data legal supervision model based on blockchain

[0009] Blockchain technology is considered to be very suitable for the field of big data legal supervision due to its characteristics of being tamper-proof and traceable. Some exploratory studies have attempted to use blockchain to record various data in the supervision process to form an unalterable supervision traceability. At the same time, the smart contract function of blockchain can also be used to automatically execute regulatory rules and improve regulatory efficiency. However, the performance bottlenecks and privacy protection issues of blockchain technology itself still limit its application in large-scale legal supervision scenarios.

[0010] In general, the existing big data legal supervision model still faces some challenges in terms of standard unification, data sharing, model interpretability, etc. For example, the supervision standards and data formats used by different regulatory departments are different, making it difficult to achieve cross-departmental joint supervision; the ownership of regulatory data also restricts the sharing and utilization of data; the "black box" characteristics of machine learning models also reduce the interpretability and credibility of regulatory results. Summary of the invention

[0011] In order to overcome the shortcomings of the above-mentioned prior art, the main purpose of the present invention is to provide an atomic operation and assembly method of a big data legal supervision model to solve several key problems existing in the existing regulatory models. The first is the lack of modularity and flexibility. The existing regulatory models are usually end-to-end designs, which are difficult to customize and adjust according to the needs of different regulatory scenarios. This not only limits the applicability of the model in different regulatory fields, but also reduces its interpretability. Secondly, the model's adaptive ability is insufficient. With the continuous deepening of regulatory practice, regulatory data and needs are constantly changing. However, static models often find it difficult to absorb new regulatory experience in a timely manner, resulting in difficulty in continuously improving regulatory effects. Finally, interpretability needs to be enhanced. Most of the existing models are based on black-box machine learning algorithms, and it is difficult to explain their internal logic, which limits the regulatory authorities' understanding and trust in model decisions.

[0012] In order to achieve the above object, the technical solution adopted by the present invention is:

[0013] A legal supervision system based on an atomic operation big data model, comprising:

[0014] The atomic operation layer includes multiple independently executable atomic operation units, which are used to implement basic functions in the legal supervision process, including data reading, information extraction, anomaly detection, permission verification and error handling;

[0015] The supervision model layer includes several legal supervision models for different supervision objectives. The legal supervision model is constructed by defining the preconditions and postconditions of atomic operations based on the atomic operation assembly method of predicate logic. The supervision model layer constructs legal supervision models for specific scenarios based on different legal supervision objectives using the atomic operation assembly method based on predicate logic. These models can adapt to different regulatory needs and data characteristics.

[0016] The early warning push layer is used to send early warning information to improve the timeliness and accuracy of legal supervision.

[0017] In one embodiment, the atomic operation unit of the atomic operation layer includes:

[0018] The pre-operation unit acquires data related to legal events and perceives the environment in the preparation stage before the legal supervision process begins. It reads and perceives data related to legal events from databases, including personnel information, monitoring information, legal documents, judicial decisions, case registrations, administrative penalty records, etc. It acquires the required data from various sources, including government databases (such as courts, procuratorates, public security bureaus, etc.), public records (such as business registrations, social security information), social media data, news reports, and other information sources related to legal events, including basic case information (such as case information, party information), legal documents, penalty records, historical case data, and other data closely related to legal events.

[0019] The post-operation unit, in the analysis and feedback phase after the monitoring process, conducts content comparison, threshold setting, and in-depth analysis of results. It conducts in-depth analysis of detected abnormal results, such as statistical analysis and pattern recognition of abnormal transaction amounts, abnormal action trajectories, and frequent occurrence frequencies, to determine whether they reflect potential illegal behavior or risks;

[0020] The operating unit in operation performs information extraction during the operation phase of the supervision process. Based on the real-time extracted information such as name, vehicle, company, amount, time and location, and according to the abnormal threshold set based on prior knowledge, it performs real-time detection of possible problems and directly pushes some simple judgments to the early warning push layer, such as the use of medical insurance for deceased persons.

[0021] In one embodiment, the operation of the pre-operation operation unit includes:

[0022] Data reading operation, connecting to different databases according to the selected database type, reading the stored information, and storing it;

[0023] Environment-aware operations: obtaining the current terminal environment information to perform corresponding atomic operations according to different environments, such as adapting the reading of the operating system and database;

[0024] Permission-aware operation: Use static and dynamic permission-aware methods to determine whether the person currently performing the operation has the permission to perform the operation. The dynamic permission-aware method is a time factor.

[0025] Error perception operation, during the execution process, determines whether an error occurs due to operational reasons, thereby affecting the normal operation of the supervision model. When an error is found, the error location is recorded and saved.

[0026] In one embodiment, the operation of the post-operation unit includes:

[0027] Content comparison: compare the information extracted from the operation process related to business implementation, and extract various types of information including similarities, differences and frequencies;

[0028] Threshold setting: setting thresholds for each type of information. When the statistical data, such as information frequency, amount statistics, time interval, data distribution, etc., exceeds the set threshold, it will be marked as abnormal data and trigger an early warning.

[0029] In one embodiment, the legal supervision model is constructed by the following method:

[0030] Define the preconditions and postconditions of atomic operations; the preconditions refer to the conditions that must be met before a specific atomic operation is executed. These conditions ensure that the execution of the operation is reasonable and effective. Only when these conditions are met can the atomic operation proceed smoothly; the postconditions refer to the conditions that need to be met after the execution of a specific atomic operation; it is used to verify whether the execution result of the operation meets expectations and ensure the correctness and effectiveness of the operation. The specific definitions of the two are mainly based on demand analysis, condition identification and logical reasoning;

[0031] Use predicate logic to establish logical relationships between atomic operations and seek achievable atomic operation paths based on different supervision objectives;

[0032] By assembling atomic operation units that meet the prerequisites, a complete legal supervision model process is constructed to ensure the correctness and interpretability of the model.

[0033] In one embodiment, the predicate logic is used to establish the logical relationship between atomic operations, to seek achievable atomic operation paths according to different supervision objectives, to assemble atomic operation units that meet the preconditions, and to construct a complete legal supervision model process. The implementation method is as follows:

[0034] Use predicate logic to establish the logical relationship between atomic operations. First, clarify the input, output and function of each atomic operation, and use predicate logic to define the context between operations. Then, use logical reasoning to verify whether the logical relationship between operations is consistent, to ensure that there will be no contradictions or errors in the execution process.

[0035] According to the specific supervision objectives, analyze the required final results and intermediate steps, determine the required atomic operation types, and find ways to achieve the supervision objectives based on active and passive methods;

[0036] By assembling atomic operation units that meet specific data input or environmental conditions, and arranging the atomic operation units that meet logical relationships and preconditions in sequence to form a complete operation sequence, a complete legal supervision model is constructed that includes various operations before, during, and after operation and can achieve the supervision objectives.

[0037] In one embodiment, atomic operations are selected and assembled in combination with their multiple features with the supervision business logic, functions and domain knowledge as constraints. First, a comprehensive analysis of the supervision business logic, functions and domain knowledge is conducted to clarify the required functions and goals to guide the selection of atomic operations. Then, based on the business logic and domain knowledge, atomic operations that meet the current supervision goals are screened out. Finally, according to the bottom-up hierarchical structure, the selected atomic operations are assembled according to logical relationships to ensure the consistency of data flow and control flow at each level, thereby obtaining the legal supervision model. The legal supervision model has four levels from bottom to top: information source, preliminary information, intermediate judgment and result judgment. Among them, the information source includes open source data, government supervision data and specific data sources that are not open source data or government supervision data, which may contain industry internal data, social media data, etc. Preliminary information refers to information obtained by information extraction operations. For a specific supervision target, the required basic information is first obtained from the information source, and then content comparison and threshold setting logical judgment are performed. Finally, it is judged whether each information exceeds the threshold to reach an abnormal value, thereby achieving the final supervision target judgment.

[0038] In one embodiment, the early warning push layer uses visualization technology to provide visualization and reporting functions of the supervision situation.

[0039] The present invention also provides a method for atomic operation and assembly of a big data legal supervision model, which is used to form the legal supervision system based on the atomic operation big data model, and further used to realize legal supervision, including the following steps:

[0040] First, according to the legal supervision model, whether the authority is possessed is determined through dynamic authority perception;

[0041] Then, various information extraction operations are performed from the information source to extract basic information;

[0042] Subsequently, some of the information is processed for content comparison to generate secondary information such as frequency;

[0043] Ultimately, based on the information obtained, it is determined whether any data exceeds the set threshold and becomes abnormal data. When an anomaly is found, the path of the above atomic operations is recorded to form a model that meets the supervision objectives. By recording the path of atomic operations, the system can flexibly adjust and optimize the legal supervision model used in subsequent operations to adapt to specific regulatory needs.

[0044] In one embodiment, the present invention can find a path to achieve the supervision goal by one of the following methods:

[0045] Method 1: Active

[0046] In the entire process from data reading to warning push, all atomic operations that have been performed are combined to independently build a legal supervision model involving supervision targets;

[0047] Method 2: Passive

[0048] Based on the redundancy analysis of predicate logic, various atomic operation paths required to reach the abnormal data are obtained.

[0049] Compared with the prior art, the present invention divides the entire regulatory process into a series of atomic operation units, and uses predicate logic to describe the preconditions and postconditions between these atomic operations, thereby achieving refined modeling and dynamic assembly of the regulatory process. This greatly improves the modularity and flexibility of the model, and based on the interpretable atomic operation units and predicate logic description, significantly enhances the interpretability of the entire regulatory model, which is conducive to the regulatory authorities' understanding and trust in the model decision-making process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is an atomic operation segmentation and three-dimensional structure diagram.

[0051] Figure 2 It is a flowchart of the data reading operation.

[0052] Figure 3 It is a flow chart of the environment perception operation.

[0053] Figure 4 It is a flowchart of the information extraction operation.

[0054] Figure 5 It is a flowchart of post-run operations.

[0055] Figure 6 This is a flowchart of SMS warning push.

[0056] Figure 7It is an atomic operation input / output association analysis model diagram.

[0057] Figure 8 It is a flowchart for building atomic operation paths. DETAILED DESCRIPTION

[0058] The following is a detailed description of the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the embodiments of the present invention can be combined with each other in the absence of conflict.

[0059] In order to solve the aforementioned problems of the present invention, it is necessary to design a more flexible, interpretable and adaptable big data legal supervision model. On the one hand, the model should have good modularity and scalability, and be able to adapt to the business needs and data characteristics of different regulatory departments; on the other hand, it should also have strong interpretability, which can not only accurately identify potential violation risks, but also clearly explain the basis for identification, and enhance the credibility of regulatory results. In addition, the model should also have a certain degree of adaptive ability, and can be continuously optimized and updated as regulatory practices continue to enrich, and continuously improve the timeliness and accuracy of supervision.

[0060] To this end, the present invention proposes a legal supervision system and method based on an atomic operation big data model. The core idea is to subdivide the entire regulatory process into a series of atomic operation units, and use predicate logic to describe the preconditions and postconditions between these atomic operation units, so as to achieve refined modeling and dynamic assembly of the regulatory process. This not only improves the modularity and flexibility of the model, but also lays the foundation for enhancing its interpretability. Furthermore, an incremental learning mechanism can be incorporated to enable the model to continuously absorb new regulatory practice experience and improve the timeliness and accuracy of regulatory effects.

[0061] In the present invention, atomic operations refer to the basic units that constitute the regulatory process, such as data reading, information extraction, anomaly detection, permission verification, and error handling. These atomic operation units have a certain degree of independence and can be flexibly assembled according to the needs of different regulatory scenarios. The present invention divides the regulatory process into three layers, that is, the legal supervision system based on the atomic operation big data model mainly includes these three layers: atomic operation layer, supervision model layer, and early warning push layer. Among them:

[0062] The atomic operation layer can be further refined into multiple independently executable atomic operation units to implement basic functions in the legal supervision process, including data reading, information extraction, anomaly detection, permission verification, and error handling.

[0063] The supervision model layer includes several legal supervision models for different supervision targets. According to different legal supervision targets, the present invention uses an atomic operation assembly method based on predicate logic, defines the preconditions and postconditions of atomic operations, and uses predicate logic to establish the logical relationship between them, thereby constructing a complete legal supervision model for specific scenarios, that is, different supervision targets. These models can adapt to different regulatory needs and data characteristics. This method based on predicate logic not only ensures the effectiveness and interpretability of the model, but also realizes the flexibility and adaptability of the model by dynamically assembling atomic operations.

[0064] The early warning push layer is mainly used to send early warning information to improve the timeliness and accuracy of legal supervision.

[0065] The specific implementation content of the present invention includes constructing atomic operation segmentation, designing the three-dimensional structure of the supervision model, the specific implementation of atomic operations, and the atomic operation assembly of predicate logic. Figure 1 This is a block diagram of the big data legal supervision model based on atomic operations of the present invention, refer to Figure 1 , which mainly includes the following contents:

[0066] 1. Construct a three-dimensional structure for atomic operation segmentation and design supervision model

[0067] According to the investigation of various existing supervision models, the main problem currently faced by the field of big data legal supervision is that it is difficult to share and integrate data between multiple administrative departments, and the reasonable division of atomic operations also faces challenges. Therefore, the goal of the atomic operation segmentation modeling design of the present invention is to achieve fine-grained decomposition of the legal supervision model, and decompose the supervision model into a combination of atomic functions according to the supervision function, and different atomic functions correspond to an atomic operation. Conversely, according to different atomic operations, they can be combined into a legal supervision model with specific functions according to needs, in order to achieve fine-grained decomposition and flexible combination of supervision models. According to the previous demand analysis, the atomic operation model of the legal supervision model has the following functions:

[0068] 1) It can improve the interpretability of the legal supervision model to a certain extent, because the legal supervision model is constructed by a combination of atomic operation models, and the functions of atomic operations are clear. Therefore, according to the combination of atomic operations, the flow characteristics of data in the legal supervision model can be realized, thereby improving the interpretability of the legal supervision model. 2) It can improve the coordination between departments to a certain extent. Due to the clarity and interpretability of atomic operations, it can promote data contributions between different departments to a certain extent and improve the coordination between departments. According to the above characteristics, the atomic operation modeling of the legal supervision model needs to consider the rationality of atomic operation segmentation on the one hand, and the analyzability and predictability of atomic operations on the other hand.

[0069] The hierarchical structure of the atomic operations preset in the present invention is as follows: Figure 1 As shown in the figure. The structure is roughly divided into three layers. The bottom layer is various atomic operations, such as data reading, information extraction, anomaly detection, permission verification and error handling. The functions of atomic operations are single and clear, which is conducive to improving the interpretability of the supervision model. At the same time, these atomic operations can connect to the data source that provides information, perform customized information extraction and analysis calculation for different supervision issues, extract different information to solve different specific problems, and perform weighted operations and judgments on different indicators under different evaluation criteria. For example, in medical insurance events, people, time, events, locations, purchase frequency, purchase amount, etc. can be extracted, and abnormal information can be filtered out through content comparison and threshold judgment. The middle layer is the supervision model layer for specific supervision scenarios, which corresponds to specific warning case reasons, such as various illegal and irregular issues. The atomic operations of the bottom layer can be flexibly combined according to the requirements of different cases to form different supervision models for specific supervision scenarios, such as legal supervision model 1, legal supervision model 2 and legal supervision model 3 in the figure. The top layer is event warning, which performs the final abnormal information summary and alarm push operation, such as abnormal insurance fraud SMS alarm push operation, email push operation, etc. This layer can summarize the abnormal information extracted after the previous model runs, and then package it into early warning information and push it to relevant personnel. In the system of the present invention, an atomic operation is only responsible for completing a specific function, which helps to improve the interpretability of the operation. Through this layered design, the supervision model is established by combining different atomic operations, which can improve the flexibility and interpretability of the supervision model as a whole, and can also achieve refined control over different regulatory areas. The design method of the big data legal supervision model based on atomic operations is expected to break through many bottlenecks in the current supervision model.

[0070] 2. Specific implementation of atomic operations

[0071] According to the division of atomic operations in operation time, the atomic operation units of the atomic operation layer of the present invention include: a pre-operation operation unit, a post-operation operation unit and an in-operation operation unit.

[0072] The pre-operation unit acquires legal event-related data and perceives the environment in the preparation stage before the legal supervision process begins. The main function of the data acquisition operation is to connect to different databases according to the selected database type, read the various information stored in the database, and store the data to facilitate the subsequent targeted extraction and further processing of the data. The database content used includes personnel information, monitoring information, legal documents, judicial decisions, case registration, administrative penalty records and other data related to legal events. The required data is obtained from various sources including government databases (such as courts, procuratorates, public security bureaus, etc.), public records (such as business registration, social security information), social media data, news reports and other information sources related to legal events, including basic case information (such as case information, party information), legal documents, penalty records, historical case data and other data closely related to legal events.

[0073] Figure 2 This is a flowchart of the data reading operation. In this operation, since different problems and different companies may choose different databases as information sources, it is necessary to first determine the type of database to be connected. Currently, there are seven databases that can be connected, namely Kingbase, PostgreSQL, MySQL, GaussDB, Oracle, SQLite and MSSQL databases. In the specific implementation, redundant methods are chosen for implementation. Write corresponding connection functions for each database type to operate. Users need to enter the required database information as prompted. For example, for MySQL database, host name, port number, user name, password and database name are required. After a successful connection, the information in the database will be temporarily stored in the Python cursor for further extraction and processing.

[0074] There are also a series of pre-operation perception operations in the pre-operation operation. The main function of the environment perception operation is to obtain various environmental information of the current terminal, so as to facilitate the execution of different operations in different environments later, so that the operation has better adaptability and can run well in different environments. Figure 3The flowchart of the environment perception operation is shown. Usually as the first step after the model is executed, this operation obtains the environment information of the current device and software to ensure that the subsequent program can run effectively. In this operation, by calling the platform library in Python, the current environment can be perceived, and basic environment information such as the operating system name, operating system version, current Python interpreter version number, machine hardware name, processor name, and current system network name can be extracted, so that the corresponding operations can be performed according to different environments using redundant methods, thereby improving the adaptability of the operation. In different environments, it can ensure that the system can run well in various environments and enhance robustness. At the same time, this information will also be stored in a txt file for subsequent query and verification.

[0075] The main function of the permission-aware operation is to determine whether the person currently performing the operation has the authority to perform the operation. This permission-aware operation not only uses static permission-aware methods such as account and password to judge the personnel's authority, but also introduces dynamic permission judgment, that is, the time factor. The same user (a set of user names and passwords) may have different permissions at different times, which means that permissions may be dynamically updated over time. Introducing time variables in permission-aware operations can more effectively judge the user's permissions at the current moment. Finally, the main function of the error-aware operation is to determine whether an error has occurred during the user's execution due to human error or other reasons, resulting in the failure of the supervision model to operate normally. If the above situation occurs, the model will save and record the error so that subsequent operators can analyze and change it.

[0076] In the operation phase of the supervision process, the operation unit extracts various types of relevant information closely related to the specific business from various information sources (i.e., different types of databases) in the previous step according to the supervision cases and supervision targets. For different business combinations, the supervision model may need to extract different types of information. In order to cope with the diversity of information types in different businesses, these operations are further subdivided into multiple small operations. For example, if the target case is a medical insurance case, it is necessary to extract information such as the insured person, the person buying the medicine, the name of the pharmacy, the purchase amount, and the type of medicine purchased in the database. Although there are many types of information for different businesses, the processes of these operations are basically the same. For example, name information extraction, company information extraction, hospital information extraction, amount information extraction, etc. The main function of these atomic operations is to read data from the database and extract information in a targeted manner. Depending on the extraction, this information may be directly used for early warning of abnormal situations, such as using the ghost medical insurance of deceased people in medical insurance issues, or only store preliminary information for further comparison or judgment later, such as exceeding the amount limit or multiple card swipes between family members.

[0077] Figure 4 This is a flowchart of the information extraction operation. In this operation, first, it is necessary to determine what kind of information extraction operation needs to be performed based on the supervision case and the supervision target. Some information that is not related to this case does not need to be extracted. Then, according to the data in the database saved by the data reading operation, the corresponding information is extracted. In some information extraction, such as the amount information extraction, there are several advantages to sorting the extracted information in descending order. First, for target information such as the amount, descending order can make it easier to find abnormal information, and it is also more convenient when making threshold judgments. Second, when the extraction target is an information group containing a lot of information, it can be sorted by time to make the result more readable and more convenient for the next operation. Third, when it is necessary to obtain information such as personnel frequency, the names can be sorted in descending order in this step, so that the information of the same person will be together, which is convenient for the subsequent content comparison operation to count the frequency information. After the above process is completed, the information extraction operation will store the extracted primary information in a txt file, so that the content comparison operation that may be performed next can obtain secondary information such as the same information, the different information, and the frequency information.

[0078] The analysis and feedback phase of the post-operation operation unit after the supervision process is roughly divided into two parts, namely content comparison operation and threshold setting operation. Content comparison is to compare the contents of various types of information related to the specific business implementation extracted from the operation process, extract various types of secondary information including similarity, difference and frequency, and store these secondary information, waiting for further judgment to confirm whether there is abnormal information. Threshold setting operation is a key operation to determine whether there is abnormal information. Its main function is to set thresholds for various types of information. When some data exceeds the set threshold after statistics, it will be listed as abnormal data and an early warning will be issued.

[0079] For some information, such as the amount information extracted, content comparison may not be required, and the threshold setting can be used to directly determine whether it exceeds the threshold. However, for other information, such as in medical insurance cases, if a person uses his or her medical insurance for the entire family, it may lead to a large number of medical insurance reimbursement records in a short period of time. In order to conduct an early warning, content comparison is required to analyze the frequency of medical insurance reimbursement of the person and compare it with the set threshold to determine whether it is abnormal data. Threshold setting is to set thresholds for various types of information. When the statistical data, such as information frequency, amount statistics, time interval, data distribution, etc., exceeds the set threshold, it will be marked as abnormal data and trigger an early warning.

[0080] Figure 5This is a flowchart of the entire process of the post-run content comparison operation and the threshold setting operation. In this operation, you first need to input the information about the supervision case extracted by the previous information extraction operation, and then compare and extract the secondary information such as the same information, different information, and frequency information according to the actual requirements. Here, you can use the jieba and collections libraries in Python to achieve this. Then, the obtained frequency and other information are sorted in descending order. Similar to the first point of the information extraction operation in descending order, arranging this information in descending order can make it easier to find abnormal information and make it easier to make threshold judgments. Then, according to the actual situation of the supervision required, set thresholds for each piece of information and compare the processed information with the set thresholds. If abnormal data exceeding the threshold is found, these data will be recorded and summarized in the early warning push.

[0081] Furthermore, in-depth analysis can be introduced, specifically, in-depth analysis of the detected abnormal results, such as statistical analysis and pattern recognition of abnormal transaction amounts, abnormal movement trajectories, frequent occurrence frequencies, etc., to determine whether they reflect potential illegal behavior or risks.

[0082] The top-level warning push layer of the supervision model is the result layer. Its main function is to summarize the abnormal information extracted after the model is run, and then package and push the warning information to relevant personnel through SMS, email and other communication methods. Furthermore, it can also use visualization technology to provide visual display and reporting functions of the supervision situation.

[0083] Figure 6 This is a flowchart of SMS warning push. In the warning push operation, SMS is mainly sent by connecting to an external communication platform to send warning information to the target mobile phone. First, it is necessary to read and record the abnormal warning information output by the post-operation operation in the previous step. Then it is necessary to obtain the SMS signature, request address, request parameters and other information of the communication platform, and write them into the function of sending information. Finally, call this function, enter the user name and password, and determine whether you have the permission to send information through the permission-aware operation. If you have the permission, you can execute the SMS sending operation. After the SMS is sent normally, the sent parameters and SMS information are returned.

[0084] 3. Atomic operation assembly based on predicate logic

[0085] Based on the information initially extracted from the database, this paper studies how to achieve the legal supervision goal by combining atomic operations and constructs the atomic operation path of the legal supervision model, which belongs to the global analysis of legal supervision. The present invention is based on the atomic operation input / output association analysis method of predicate logic, and proposes an atomic operation path that combines a series of atomic operations to reach the supervision goal starting from basic initial conditions such as names. The core idea of ​​this principle is to use the input / output relationship of atomic operations, that is, subsequent atomic operations are based on the previous atomic operations, such as content comparison operations, and after they occur, the categories of input information are increased in turn, including frequency information, sameness information, threshold information, etc. At the same time, the present invention also proposes an atomic operation input / output association analysis model, which uses the output information results brought by the previous atomic operation and the input information required for the next atomic operation to find the atomic operation path that can achieve the supervision goal in the legal supervision model.

[0086] In order to construct a legal supervision model, the present invention needs to define the preconditions and postconditions of atomic operations, wherein the predicate logic is then used to establish the logical relationship between the atomic operations, clarify the input, output and function of each atomic operation, and use the predicate logic to define the before and after relationship between the operations. Through logical reasoning, it is verified whether the logical relationship between the operations is consistent, ensuring that no contradictions or errors will occur during the execution process, thereby seeking achievable atomic operation paths according to different supervision objectives, that is, according to the specific supervision objectives, analyzing the required final results and intermediate steps, determining the required atomic operation types, and finding ways to achieve the supervision objectives; finally, by assembling atomic operation units that meet specific data inputs or environmental states, the atomic operation units that meet the logical relationships and preconditions are arranged in order to form a complete operation sequence, and a complete legal supervision model is constructed that includes various operations before, during and after operation and can complete the supervision objectives, ensuring the correctness and interpretability of the model.

[0087] Predicate logic is the expansion and development of propositional logic. It is based on the further decomposition of simple propositions, decomposing simple propositions into two parts: individuals and predicates, to solve some simple and common reasoning processes that propositional logic cannot handle. The so-called individual refers to an objective entity that exists independently of people's subjectivity. It may be a concrete object or an abstract concept. Predicates are usually used to describe the properties of an object or the relationship between objects. The legal supervision model designed by the present invention is constrained by the supervision business logic, functions and domain knowledge, selects atomic operations and assembles them in combination with their various characteristics. First, a comprehensive analysis of the supervision business logic, functions and domain knowledge is carried out to clarify the required functions and goals to guide the selection of atomic operations. Then, based on the business logic and domain knowledge, the atomic operations that meet the current supervision goals are screened out. Finally, according to the hierarchy from bottom to top, the selected atomic operations are assembled according to the logical relationship to ensure the coherence of the data flow and control flow at each level, and the legal supervision model is obtained to achieve the purpose of solving various supervision goals.

[0088] The legal supervision model of the present invention has four levels from bottom to top: information source, preliminary information, intermediate judgment and result judgment; the information source includes open source data, government regulatory data and specific data sources that are not open source data or government regulatory data, which may contain internal industry data, social media data, etc.; preliminary information refers to the name, time, place and other information obtained by the information extraction operation; for a specific supervision target, first obtain the required basic information from the information source, then perform content comparison and threshold setting logical judgment, and finally judge whether each information exceeds the threshold to reach an abnormal value, so as to achieve the final supervision target judgment.

[0089] In the implementation, the model first determines whether it has permission through a dynamic permission-aware operator, and then performs various information extraction operations from the information source to extract basic information. Subsequently, it performs content comparison and other processing on some information to generate secondary information such as frequency. Finally, based on this information, it determines whether any data exceeds the set threshold and becomes abnormal data. Once an anomaly is found, the path of the above atomic operations is recorded to form a model that meets the supervision objectives.

[0090] In order to find an atomic operation input / output correlation analysis method that conforms to predicate logic, the present invention also proposes an atomic operation input / output correlation analysis model, which consists of four modules: open source data processing, government regulatory data processing, and early warning information generation and visualization display, such as Figure 7 shown.

[0091] The four modules in the figure are both relatively independent and interrelated. In the processing of open source data and government regulatory data, a redundant method is used to design a data reading function that can read multiple databases to ensure that the data in the database can be initially obtained. The temporary classification discussion is mainly due to the fact that there may be obvious differences in the pre-operation perception part, especially the perception part about permissions. In the open source data part, the main perception operation used is to determine whether the source data is safe and reliable, and there may not be too many operations in the permission perception. In the government regulatory data part, since the data may have a certain degree of confidentiality, in order to prevent data leakage, the permission perception operation part may be more complicated. The atomic operation layer operations mainly used in these two modules in the legal supervision model are roughly pre-operation operations and information extraction operations during operation, and the extracted information is finally obtained for further comparison or judgment by the supervision model. In the early warning module, the main operations are post-operation operations and early warning push operations. According to different supervision models, the supervision target, various thresholds and operator permissions are input to achieve the screening and push of abnormal information. The main function of the final visualization module is to display the current supervision model, record all atomic operations from data reading to warning push, and form the operation path of the model in sequence. The legal supervision path derived by the current supervision model can be used directly the next time the same supervision target is encountered, which greatly improves the efficiency of building the supervision model. The difference in atomic operation paths represents the difference in supervision models, which is not only the key to distinguishing different supervision models, but also greatly improves the practicality and flexibility of supervision models.

[0092] At the same time, different atomic operations can be combined to form new atomic operation paths and different legal supervision models. This process can be divided into two ways: active construction and passive construction. Active construction means that the operator combines all atomic operations performed in the entire process from reading data before operation to pushing warnings after operation, and independently constructs a legal supervision model related to the supervision target. The general construction process is as follows: Figure 8 As shown. This method is generally applicable to two situations. One is that the operator can roughly determine which parts of the data may be abnormal, which generally means that the operator has used similar supervision models to process data sets similar to the current supervision target. The other situation is that the preliminary screening of abnormal information has been completed using passive construction, and now it is necessary to focus on certain information. For example, if a specific individual is found in the abnormal information report, a supervision model related to the individual can be established to further investigate whether there is a more specific abnormal situation or its relationship with other abnormal situations.

[0093] When faced with complex data where it is unclear where anomalies may occur, a passive approach can be used to build a supervision model. Through redundant analysis based on predicate logic, the various atomic operation paths required to reach the abnormal data can be obtained.

[0094] The present invention is defined in a passive supervision model f: a five-tuple (Im, Bj, Ml, Tk, Pn), which represents the input and output of a type of supervision model, where Im represents the input of various types of information, such as information read from a database. Bj represents the current state of the supervision model, such as information source, preliminary information, intermediate judgment or result judgment. Ml, Tk, Pn are information such as time, which are parameters in the dynamic permission perception service. Because various permissions may change over time, it is necessary to introduce information such as time to judge the current permissions.

[0095] In addition, the atomic operation G: Gi=H(Ix) Gi is various atomic operations, H is the atomic operation function, and Ix etc. are various input variables, such as information.

[0096] Finally, in the process of predicate logic association using atomic operation input / output, the input information source D, timing signal M, T, P, account password IP, and various thresholds are used. The legal supervision model set C obtained by this method represents the atomic operation path from the bottom layer to the achievement of the supervision goal this time. The atomic operation required for each step is the Bj set, and the input for each step is the Im set, and the rest is the time signal. The legal supervision path obtained each time can be used directly when encountering the same supervision target next time, or other paths can be found.

[0097] In general, the atomic operation and assembly method of the big data legal supervision model of the present invention not only effectively solves some key problems existing in the existing model, but also significantly enhances the model's effectiveness, interpretability and adaptability, bringing new breakthroughs to the application of big data in the field of legal supervision. This method is not only innovative in theory, but also has broad prospects in practical applications.

Claims

1. A legal supervision system based on atomic operation big data model, characterized in that: include: The atomic operation layer includes multiple independently executable atomic operation units, which are used to implement basic functions in the legal supervision process, including data reading, information extraction, anomaly detection, permission verification and error handling; The supervision model layer includes several legal supervision models for different supervision objectives. The legal supervision models are constructed based on the atomic operation assembly method of predicate logic by defining the preconditions and postconditions of the atomic operations. The warning push layer is used to send warning information.

2. According to the legal supervision system based on the atomic operation big data model of claim 1, it is characterized in that: The atomic operation unit of the atomic operation layer includes: The pre-operation unit acquires data related to legal events and senses the environment in the preparation stage before the legal supervision process begins; The post-operation operation unit, in the analysis and feedback phase after the monitoring process, conducts content comparison and threshold setting as well as in-depth analysis of results, and conducts in-depth analysis of detected abnormal results to determine whether they reflect potential illegal behavior or risks; The operating unit in operation performs information extraction during the operation phase of the supervision process. Based on the real-time extracted information and the abnormal thresholds set according to prior knowledge, it performs real-time detection of possible problems and directly pushes some simple judgments to the early warning push layer.

3. According to claim 2, the legal supervision system based on the atomic operation big data model is characterized in that: The operations of the pre-operation operation unit include: Data reading operation, connecting to different databases according to the selected database type, reading the stored information, and storing it; Environment-aware operations: obtaining the current terminal environment information to perform corresponding atomic operations according to different environments; Permission-aware operation: Use static and dynamic permission-aware methods to determine whether the person currently performing the operation has the permission to perform the operation. The dynamic permission-aware method is a time factor. Error perception operation, during the execution process, determines whether an error occurs due to operational reasons, thereby affecting the normal operation of the supervision model. When an error is found, the error location is recorded and saved.

4. According to claim 2, the legal supervision system based on the atomic operation big data model is characterized in that: The operation of the post-operation operation unit includes: Content comparison: compare the information extracted from the operation process related to business implementation, and extract various types of information including similarities, differences and frequencies; Threshold setting: setting thresholds for the various types of information. When the statistical data exceeds the set threshold, it is marked as abnormal data and triggers an early warning.

5. According to claim 1, the legal supervision system based on the atomic operation big data model is characterized in that: The legal supervision model is constructed by the following method: Define the preconditions and postconditions of atomic operations; the preconditions refer to the conditions that must be met before a specific atomic operation is executed, and the postconditions refer to the conditions that need to be met after a specific atomic operation is executed; Use predicate logic to establish logical relationships between atomic operations and seek achievable atomic operation paths based on different supervision objectives; By assembling atomic operation units that meet the prerequisites, a complete legal supervision model process is constructed to ensure the correctness and interpretability of the model.

6. The legal supervision system based on the atomic operation big data model according to claim 5 is characterized in that: The predicate logic is used to establish the logical relationship between atomic operations, seek achievable atomic operation paths according to different supervision objectives, assemble atomic operation units that meet the preconditions, and construct a complete legal supervision model process. The implementation method is as follows: Use predicate logic to establish the logical relationship between atomic operations. First, clarify the input, output and function of each atomic operation, and use predicate logic to define the context between operations. Then, use logical reasoning to verify whether the logical relationship between operations is consistent, to ensure that there will be no contradictions or errors in the execution process. According to the specific supervision objectives, analyze the required final results and intermediate steps, determine the required atomic operation types, and find ways to achieve the supervision objectives based on active and passive methods; By assembling atomic operation units that meet specific data input or environmental conditions, and arranging the atomic operation units that meet logical relationships and preconditions in sequence to form a complete operation sequence, a complete legal supervision model is constructed that includes various operations before, during, and after operation and can achieve the supervision objectives.

7. The legal supervision system based on the atomic operation big data model according to claim 1 is characterized in that: Taking the supervision business logic, function and domain knowledge as constraints, atomic operations are selected and assembled in combination with their multiple characteristics. First, a comprehensive analysis of the supervision business logic, function and domain knowledge is carried out to clarify the required functions and goals to guide the selection of atomic operations. Then, based on the business logic and domain knowledge, atomic operations that meet the current supervision goals are screened out. Finally, according to the bottom-up hierarchical structure, the selected atomic operations are assembled according to logical relationships to ensure the consistency of data flow and control flow at each level, so as to obtain the legal supervision model; the legal supervision model has four levels from bottom to top: information source, preliminary information, intermediate judgment and result judgment; the information source includes open source data, government supervision data and specific data sources that are not open source data or government supervision data; preliminary information refers to information obtained by information extraction operations; for a specific supervision target, the required basic information is first obtained from the information source, and then content comparison and threshold setting logical judgment are carried out, and finally it is judged whether each information exceeds the threshold to reach an abnormal value, so as to achieve the final supervision target judgment.

8. According to claim 1, the legal supervision system based on the atomic operation big data model is characterized in that: The early warning push layer uses visualization technology to provide visual display and reporting functions of the supervision situation.

9. A method for atomic operation and assembly of a big data legal supervision model, characterized in that: The steps include: First, determine whether you have the permission through dynamic permission perception; Then, various information extraction operations are performed from the information source to extract basic information; Subsequently, some of the information is processed for content comparison to generate secondary information; Finally, based on the information obtained, it is determined whether any data exceeds the set threshold and becomes abnormal data. When an anomaly is found, the path of the above atomic operations is recorded to form a model that meets the supervision objectives.

10. The atomic operation and assembly method of the big data legal supervision model according to claim 9 is characterized in that: Find a path to achieve your monitoring goals by doing one of the following: Method 1: Active In the entire process from data reading to warning push, all atomic operations that have been performed are combined to independently build a legal supervision model involving supervision targets; Method 2: Passive Based on the redundancy analysis of predicate logic, various atomic operation paths required to reach the abnormal data are obtained.

Images