Electronic contract approval management system

Abstract

The invention, which relates to the technical field of electronic contract management, discloses an electronic contract approval management system comprising a process monitoring module, an editing tracking module, a risk symptom detection module, a data analysis module, a joint identification module and a response decision module. Dynamic behavior data such as an approval path, processing duration and transfer failure are obtained through process monitoring, and an editing tracking module collects multi-person editing behavior track data; the risk symptom detection module analyzes the data combination mode and outputs a risk trigger signal; the data analysis module constructs a multi-dimensional data set and calculates a process anomaly index and a content conflict index, and the joint recognition module performs fuzzy reasoning based on the two indexes and divides risk levels; the response decision module executes a grading intervention strategy according to the risk grade; according to the method, process abnormity can be identified and guided to be repaired, multi-person editing conflicts can be accurately warned, content consistency is guaranteed, the method has self-evaluation and optimization capability of strategy response effects, and the intelligent level of contract approval is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic contract management, and more specifically, to an electronic contract approval management system. Background Art

[0002] With the widespread application of electronic contracts in various business scenarios such as government affairs, finance, manufacturing, and the Internet, enterprise contract management systems are gradually developing towards online, automated, and intelligent. Contract approval is a key link in the life cycle of an electronic contract. Its process structure often involves multiple business roles, approval conditions, and branch paths, and the contract content itself usually needs to be edited, reviewed, and confirmed by multiple people. However, the existing contract approval system has obvious shortcomings when dealing with complex process configurations and high-concurrency collaborative editing scenarios.

[0003] On the one hand, traditional systems mostly use static process templates, lacking the ability to dynamically identify and provide feedback on abnormal behaviors in actual operation. When the process configuration is unreasonable or the execution status of the approval node is abnormal, it is difficult for the system to detect and repair it in time, resulting in frequent approval jams, missigning or omissions. On the other hand, in the contract content editing process, the common multi-person collaboration mode is prone to problems such as cross-editing, version overwriting, and content confusion. Especially when user editing behaviors are highly overlapping or editing conflicts are not controlled, it is more likely to cause major mistakes, which in turn affects contract performance or compliance audits.

[0004] Most risk response mechanisms in existing technologies are still at the fixed rule level, lacking dynamic evaluation and strategy adjustment capabilities based on data feedback, and are difficult to adapt to changing business processes and collaborative behaviors, resulting in delayed system response to risks, lack of precision in intervention, and insufficient overall management capabilities. Therefore, there is an urgent need for an electronic contract approval management system that can monitor approval processes and collaborative editing behaviors, and has intelligent identification, quantitative evaluation, and hierarchical response capabilities to improve the efficiency, accuracy, and security of contract circulation. Summary of the invention

[0005] To achieve the above object, the present invention provides the following technical solutions: An electronic contract approval management system, comprising: The process monitoring module is used to obtain dynamic behavior data in the electronic contract approval process, including approval path flow, processing duration, and transfer failure events; The editing tracking module is used to collect editing trajectory data of contract documents in multi-person editing scenarios, including editing time distribution, content coverage area, and behavioral interaction sequences between concurrent users; A risk sign detection module is used to analyze the above process behavior data and editing trajectory data, and output a risk calculation trigger signal after detecting an abnormal event combination pattern; The data analysis module responds to risk trigger signals, builds a multidimensional data set, and calculates two core indexes for joint risk identification, namely, the process anomaly index and the content conflict index, for downstream modules to call; The joint identification module performs fuzzy reasoning based on the process anomaly index and content conflict index to classify the joint risk level; The response decision module performs differentiated interventions based on risk level results, including process correction, authority constraints, notification push, and log archiving.

[0006] In a preferred embodiment, the process monitoring module further includes a node graph construction submodule and a time matrix analysis submodule. The node graph construction submodule abstracts the approval process into a node-edge data structure based on the graph neural structure, identifies disconnected segments and abnormal transfer paths, and the time matrix analysis submodule records the average residence time and fluctuation range of each node by constructing a time-node two-dimensional tensor structure. If the path jump probability exceeds the preset jump threshold, or the processing time standard deviation of a node exceeds the preset ratio threshold (for example, 20%) of the average processing time standard deviation of its adjacent nodes, it is determined to be an abnormal process behavior event.

[0007] In a preferred embodiment, the editing tracking module further includes an editing behavior encoding submodule and a collaborative interaction matrix generation submodule. The former serializes and encodes the operations of each user to generate an editing behavior vector; the latter forms a user interaction density matrix by establishing an editing cross-relationship table between users. If more than q users have high-frequency staggered editing behaviors in adjacent areas within a certain time window, it is identified as a conflict editing risk. All encoded data and matrix data enter the content conflict index construction model to provide input for subsequent conflict intensity quantification. High-frequency staggered editing behavior refers to the editing update frequency exceeding the preset editing threshold.

[0008] In a preferred embodiment, the risk symptom detection module adopts a combined rule judgment method to identify the continuous occurrence of a preset event pattern. When the corresponding event pattern appears, it is marked as 1, and when it does not appear, it is marked as 0, thereby obtaining an event combination data string composed of 1 and 0. If 1 appears in the event combination data string, a risk calculation trigger signal is output to trigger the subsequent data analysis module to enter the calculation state.

[0009] In a preferred embodiment, the calculation method of the process anomaly index in the data analysis module is: The approval process is modeled as a directed graph structure, where nodes represent approval operations and edges represent flow directions. The graph structure is vectorized through a graph embedding algorithm, and then the abnormal subpath vector and the standard process vector are extracted. The process abnormality index FI is defined as the distance between the two: ; Embedding vector i-th dimension component of the current approval process graph, is the i-th dimension component of the embedding vector of the standard approval flowchart, is the impact factor corresponding to the i-th dimension process structure, indicating the importance weight assigned to this dimension in the approval process diagram, which is pre-set by the system based on the training data. is a preset smoothing factor used to avoid the situation where the denominator is zero in the calculation. The value is a positive real number less than 1 (such as 0.01), and n is the dimension of the embedded vector.

[0010] In a preferred embodiment, the content conflict index is calculated in the data analysis module as follows: In the unit time window T, extract the set of content blocks edited by all users, record the editing status of each content block, and construct the editing density function: ; Same document location Edit frequency, For the The document location corresponding to the edit operation. is the document position index point currently being evaluated, m is the total number of all editing operations within the time window, The content distribution diffusion coefficient indicates the degree of discreteness of editing behavior on the document. The calculated document position index point The corresponding edit density value; ; ; The average of all edit positions; Introducing content crossover rate : ; Indicates the number of content segments where multi-user editing occurs. Indicates the total number of content segments; The maximum editing density value is recorded as , and the content crossover rate Combined to form the content conflict index: ; Represents the content conflict index.

[0011] In a preferred embodiment, the joint identification module performs fuzzy reasoning based on the combined result of the process anomaly index FI and the content conflict index CI, wherein FI and CI are respectively divided into three levels of "low", "medium" and "high" according to a preset fuzzy membership function, for example, interval mapping is achieved through a triangular membership function, and the reasoning rule is as follows: if FI belongs to "high" and CI belongs to "high", the reasoning result is "serious risk"; If FI is "medium" and CI is "high", the inference result is "heavier risk"; If FI is "high" and CI is "medium", the inference result is "medium risk"; If both are “low”, the inference result is “normal risk”; The inference results are quantified into four levels of output signals, with corresponding values ​​of 0, 1, 2, and 3, representing normal, moderate, severe, and serious risks, respectively.

[0012] The response decision module includes the following strategy selection mechanisms: If the risk level is severe, the approval process is suspended, all editing permissions are locked, and administrators are forced to intervene; If the risk level is high, restrict multiple people from editing and rebuilding the conflicting nodes in the approval path; If the risk level is medium, issue a warning and record a risk log; If the risk level is normal, no intervention is made but the behavior data is archived.

[0013] In a preferred embodiment, the response decision module further includes a supervision mechanism: After each risk response is triggered, the following three indicators are recorded: process repair success rate, conflict reduction rate, and approval acceleration. These three indicators are then subjected to polynomial regression modeling. Based on the output data of the polynomial regression modeling, namely the improvement value, the least squares method is used to fit a fitting straight line between the improvement value and the response time. The slope of the fitting straight line is compared with the preset standard slope. If the slope of the fitting straight line is greater than or equal to the preset standard slope, the current response is valid. If the slope of the fitting straight line is less than the preset standard slope, the current response is invalid, and the manual intervention window mode is entered to force the administrator to intervene.

[0014] Technical effects and advantages of the present invention: The present invention constructs an approval process graph structure, combines graph embedding with time tensor analysis technology, and realizes dynamic identification of path jumps, disconnected segments, and processing time fluctuations between process nodes. Unlike the existing static template comparison inspection method, the present invention can continuously monitor the process execution status during the actual operation. Once structural deviation behavior is found, such as abnormal node processing time standards and irregular jump paths, the system immediately marks the process abnormality and records the abnormal path vector. This mechanism can not only detect process freezes, misrouting and other problems in the first place, but also trigger subsequent response strategies to repair or reconstruct abnormal nodes, thereby effectively improving the stability and closed-loop rate of the process.

[0015] The present invention constructs a behavior coding model and an interaction density matrix in terms of editing behavior monitoring, which can accurately identify high-frequency editing behaviors of multiple people on adjacent or overlapping document areas within the same time window, and then evaluate the intensity of conflicts. The system calculates the content conflict index through the editing density function and the content intersection rate. The content conflict index can warn of potential high-risk operations such as version overwrite and wrong modification. Compared with the traditional version control mechanism, the present invention can not only detect risks, but also conduct differentiated interventions based on the density of conflict locations, such as restricting editing permissions, etc., thereby effectively ensuring the consistency, traceability and correctness of the contract content in the multi-person collaboration process.

[0016] The present invention introduces a supervision mechanism in the risk response module. The system not only automatically intervenes according to the risk level, but also records and models the effect after each response, extracts multi-dimensional feedback indicators such as process repair success rate, conflict reduction rate and approval acceleration, and forms the evolution trend of the "strategy improvement value" through polynomial regression and trend fitting. The system then compares the fitting slope with the standard slope to realize the continuous effectiveness judgment of the response strategy, and forces the system to enter the manual intervention mode when the strategy fails. This mechanism gives the system the ability to "learn to judge whether the strategy is sustainable", gets rid of the defects of rigid rules and broken feedback in the past, and realizes the technical transition of the response mechanism from fixed to self-evolving. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1 The schematic diagram is a schematic diagram of an electronic contract approval management system in the present invention.

[0018] Figure 2 It is a schematic diagram of the process monitoring module in the present invention.

[0019] Figure 3 This is a schematic diagram of the editing tracking module in the present invention.

[0020] Figure 4 This is a schematic diagram of the collaboration between the joint identification module and the response decision module in the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0022] Reference Figure 1-4 The following examples are obtained:

[0023] As the degree of digital management of enterprises continues to improve, electronic contracts are widely used in various business processes. However, in complex practical application scenarios, the approval process of electronic contracts often involves multiple roles, multiple levels, and multiple conditional branches. If the approval path design is unreasonable, the role configuration is mismatched, or the node processing efficiency fluctuates, it is very easy to cause process jams, interruptions, or incorrect approvals. At the same time, the collaborative editing of contract content also faces problems such as multiple people operating at the same time, frequent content conflicts, and untraceable editing behavior. Once a wrong signature, missed signature, or disordered editing order occurs, it will seriously affect the legal effect and performance safety of the contract.

[0024] The present invention aims to provide an electronic contract approval management system with intelligent identification and active intervention capabilities, which can achieve dual protection of approval path integrity and content collaboration consistency through process monitoring and editing behavior analysis. After detecting abnormal signs, the system automatically starts risk assessment, constructs process anomaly index and content conflict index, and combines fuzzy reasoning models to judge the joint risk level. According to the risk level, the system dynamically selects the corresponding response strategy, and evaluates the strategy effect through regression modeling and trend judgment mechanism to ensure the effectiveness of the response results and improve the stability, reliability and adaptability of the contract approval system.

[0025] Based on this, the present invention proposes an electronic contract approval management system, comprising: The process monitoring module is used to obtain dynamic behavior data in the electronic contract approval process, including approval path flow, processing duration, and transfer failure events; The editing tracking module is used to collect editing trajectory data of contract documents in multi-person editing scenarios, including editing time distribution, content coverage area, and behavioral interaction sequences between concurrent users; The risk sign detection module is used to analyze process behavior data and edit trajectory data, and output a risk calculation trigger signal after detecting an abnormal event combination pattern; The data analysis module responds to risk trigger signals, builds a multidimensional data set, and calculates two core indexes for joint risk identification, namely, the process anomaly index and the content conflict index, for downstream modules to call; The joint identification module performs fuzzy reasoning based on the process anomaly index and content conflict index to classify the joint risk level; The response decision module performs differentiated interventions based on risk level results, including process correction, authority constraints, notification push, and log archiving.

[0026] "Dynamic behavior data" refers to the process status change data collected and analyzed by the system during the execution of the electronic contract approval process, which mainly includes the following three core contents: Approval path flow information: refers to whether the approval operations between nodes are carried out in the set order, including flow time, whether nodes are skipped, and whether they are looped back; Processing duration: refers to the time taken for each node to go from pending to completed processing, which is used to measure approval efficiency and identify node anomalies; Transfer failure event: refers to abnormal termination, interruption, failed return, inability to execute logic, and other behaviors in the flow of approval process between nodes.

[0027] The process monitoring module automatically connects with the approval system and records the corresponding information every time the node status changes; transmission destination: the collected data will be sent to the risk sign detection module for abnormal event combination analysis to determine whether the trigger conditions are met; association with the process anomaly index: some data (such as node jumps, delays) will enter the process anomaly index calculation model to calculate the degree of deviation between the current process and the standard process.

[0028] "Editing trajectory data" refers to the system's continuous record of each user's operation behavior during the process of multi-person collaborative editing of contract documents. Its core includes: Editing time distribution: refers to the time point and time period of the user's document operation, which is used to analyze the editing density and collaboration opportunity; Content coverage area: refers to the document content range involved in the user's editing operation (such as paragraph number, character index interval, etc.); Concurrent user behavior interaction sequence: refers to the intersection and overlap relationship of multiple users' editing behaviors in the document space within the same time window, which is used to build a user collaboration conflict model. Data source: recorded by the editing tracking module when the user uses the contract editing system; Transmission destination: The editing behavior is vectorized and sent to the content conflict index calculation model; The concurrent cross-behavior matrix is ​​sent to the risk sign detection module to identify high-frequency staggered editing patterns; Impact module: This type of data mainly supports the calculation of the content conflict index.

[0029] The electronic contract approval management system proposed in the present invention first accesses the approval process and contract document collaboration platform within the enterprise through the front end, and continuously collects and records the full amount of dynamic data on the approval execution process and content editing behavior. In terms of process, the system tracks the execution status, processing time and path flow logic of each approval node, and establishes a structured flow chart; in terms of content, the system performs time and space encoding on the editing behavior of collaborative users, and identifies changes in editing hot spots, behavior frequency and collaboration density. Through the fusion analysis of these two types of data, the system can discover abnormal patterns hidden behind imbalances in the approval process and conflicts among multiple editors, forming risk signs with early warning value.

[0030] After the risk signs are activated, the system starts the internal data analysis engine, extracts the current process structure status and editing behavior sequence, and maps them into a multi-dimensional feature vector. The system constructs two core indicators: process anomaly index and content conflict index. The former measures the degree of deviation of the current approval path from the preset process template, and the latter measures the degree of overlap and conflict intensity in the collaborative editing process. After the index is generated, the system converts the numerical results into identifiable risk levels through a fuzzy reasoning mechanism, which are divided into four levels: normal, medium, severe and serious, to present the actual severity and scope of the risk in a refined manner, providing accurate input for subsequent response modules.

[0031] According to the different joint risk levels, the system has the ability to respond with multiple strategies in linkage. For risks of minor or warning levels, the system only records logs or sends notification reminders; when the risk level reaches a heavier or severe level, the system will automatically perform intervention actions such as process reconstruction and editing permission adjustment to prevent the risk from spreading further. At the same time, the system also has a supervision mechanism, which records and trends the repair results after each response behavior, and uses the fitting results to evaluate the effectiveness of the current strategy. When the system detects that the policy execution effect continues to be poor, it will automatically switch to manual intervention mode to ensure the stable operation and final closure of the entire contract approval process. The whole process reflects the high integration and closed-loop control capabilities of the present invention in the four stages of risk identification, analysis, response, and evaluation.

[0032] The process monitoring module further includes a node graph construction submodule and a time matrix analysis submodule. The node graph construction submodule abstracts the approval process into a node-edge data structure based on the graph neural structure, identifies disconnected segments and abnormal transfer paths, and the time matrix analysis submodule records the average residence time and fluctuation range of each node by constructing a time-node two-dimensional tensor structure. If the path jump probability exceeds the preset jump threshold, or the processing time standard deviation of a node exceeds the preset ratio threshold (for example, 20%) of the average processing time standard deviation of its adjacent nodes, it is determined to be an abnormal process behavior event.

[0033] In the present invention, the process monitoring module is not only used to record the basic behavior information in the approval process, but also further includes a node graph construction submodule and a time matrix analysis submodule, so as to realize a comprehensive evaluation of the structural stability and operation efficiency of the approval process.

[0034] The node graph construction submodule abstracts the approval process into a graph structure model consisting of nodes and paths by modeling the contract approval process. In this model, each approval action is modeled as a graph node, and the sequential flow relationship between two adjacent approval links constitutes a directed edge. The system encodes the graph through a graph neural structure and extracts the graph embedding vector of each node for subsequent structural feature learning. During the graph construction process, the system simultaneously detects the following two types of structural anomalies: (i) There are non-connected fragments in the graph that cannot be accessed from the starting node to certain intermediate or terminal nodes; (ii) There are non-linear jumps in the approval path, that is, jumping directly from a certain node to a non-directly adjacent node, which violates the preset approval order.

[0035] To identify the above jump behaviors, the system counts the jump behavior frequencies of all actual approval paths, obtains the probability value of a node jumping to a non-adjacent node in all process instances, and compares it with the jump judgment threshold set in the system. When the jump probability exceeds the threshold and the path behavior is not configured as a legal conditional branch path, the system determines that the path is an abnormal transfer path.

[0036] The time matrix analysis submodule is used to record the processing time of each approval node in multiple contract instances, and construct a two-dimensional tensor structure based on time and node number. Each row of the matrix represents an approval node, and the column vector is its processing time series. The system calculates the average processing time and the standard deviation of the processing time for each node in the tensor. In order to identify efficiency anomalies in the approval process, the system introduces a time fluctuation indicator associated with the node, that is, the ratio of the standard deviation of the processing time of a node to the standard deviation of the processing time of its adjacent nodes. If the ratio exceeds the set percentage threshold, it means that the node has a significant processing time fluctuation relative to other links, which is a potential bottleneck or abnormal approval behavior.

[0037] When any of the above structural anomalies or time fluctuation conditions are met, the system marks the process instance as "process behavior anomaly" and writes the identification value into the abnormal record list for subsequent process anomaly index modeling. In this way, the process monitoring module realizes the dual-dimensional monitoring of structural rationality and efficiency stability, providing a powerful process feature input for the subsequent risk assessment and intervention mechanism in the present invention.

[0038] The editing tracking module further includes an editing behavior encoding submodule and a collaborative interaction matrix generation submodule. The former serializes and encodes the operations of each user to generate an editing behavior vector; the latter forms a user interaction density matrix by establishing an editing cross-relationship table between users. If more than q users have high-frequency interlaced editing behaviors in adjacent areas within a certain time window, it is identified as a conflict editing risk. All encoded data and matrix data enter the content conflict index construction model to provide input for subsequent conflict intensity quantification. High-frequency interlaced editing behavior refers to the editing update frequency exceeding the preset editing threshold.

[0039] In the present invention, in order to achieve high-precision identification and risk modeling of multi-person collaborative editing behaviors, the editing tracking module further includes an editing behavior encoding submodule and a collaborative interaction matrix generation submodule, which are used to characterize the behavior patterns from three dimensions of time, location and user relationship of the editing operation, and are used for the input generation of the subsequent content conflict index model.

[0040] The editing behavior coding submodule is used to collect the operation data of each contract editing user and convert it into a structured behavior coding sequence. Specifically, each user operation will record the following fields: user ID, operation timestamp, start and end position of the editing content block (expressed as character offset or paragraph number), operation type (such as insert, delete, replace), and device terminal information. The system combines the above fields into a multidimensional vector sequence to represent the editing trajectory of the user within the unit time window, and arranges them in chronological order to form their personal editing behavior vector. This encoding not only retains the temporal characteristics of the behavior, but also retains the spatial position of the content, which is convenient for subsequent cross-behavior detection.

[0041] The collaborative interaction matrix generation submodule constructs a collaborative cross-relationship table between users based on the behavior vectors of multiple users. According to the preset time window, the system compares the editing operations of all users within the time window in pairs to determine whether there is spatial overlap in their editing content areas, that is, whether two users edit adjacent or overlapping paragraphs within the time window. If there is an intersection, the corresponding position in the user cross-relationship table is marked as one; otherwise, it is marked as zero. Based on this table, a user interaction density matrix is ​​further generated. Each row of the matrix represents a user, and each column represents the intensity of editing interaction between it and other users within the time window. The intensity value is the product of the total character length of the overlapping area between users and the time overlap ratio, reflecting the possibility of conflict in collaborative behavior.

[0042] When the system detects that there are more than a preset threshold number of users (denoted as a constant q) in a certain time window, their editing areas have a high degree of overlap in content structure, and the editing update frequency of these users exceeds the editing threshold set by the system, it is defined as "high-frequency interlaced editing behavior". The editing update frequency can be obtained by dividing the number of effective edits of each user per unit time by the length of the time window. If the frequency exceeds the preset editing threshold continuously, it means that the editing density of the area is significant and it is a high-risk area.

[0043] Once the above behavior pattern is detected, the system will mark it as a conflict editing risk event. The event mark will be input into the content conflict index calculation module together with the interaction density matrix to evaluate the conflict intensity and potential error risk of the current contract in the multi-person editing stage. In this way, the system realizes a full process closed loop from the underlying editing action collection, structured coding, relationship modeling to risk event identification, providing stable and reliable data support for subsequent risk response.

[0044] The risk sign detection module adopts a combined rule judgment method to identify the continuous occurrence of preset event patterns. When the corresponding event pattern appears, it is marked as 1, and when it does not appear, it is marked as 0, and an event combination data string composed of 1 and 0 is obtained. If 1 appears in the event combination data string, a risk calculation trigger signal is output to trigger the subsequent data analysis module to enter the calculation state.

[0045] In the present invention, the risk sign detection module is used to determine whether there are potential abnormalities in the current electronic contract approval process and multi-person editing process that require the initiation of risk assessment. In order to achieve low-latency and high-accuracy capture of risk signals, the module adopts an identification mechanism based on combined rule judgment, combined with structured data input from the process monitoring module and the editing tracking module, to monitor and identify a set of preset event patterns.

[0046] "Event pattern" is a set of risk signs predefined by the system during the modeling phase. Each event pattern corresponds to a local feature that may cause process abnormality or content conflict. For example, event patterns may include: "A certain approval node fails to transfer twice in a row", "The same contract version is edited and overwritten by more than three users within five minutes", "Processing time fluctuation exceeds the percentage threshold of the previous and next nodes", "The number of characters in the overlapping area of ​​editing behavior is greater than the set threshold", etc. The system will detect the above multiple event patterns one by one in each time period.

[0047] In actual operation, the system assigns a site position to each event pattern according to the time sequence and logical number, and Boolean codes it according to the detection results. If an event pattern appears in the current detection cycle, its corresponding position is marked as "1"; if it is not detected, it is marked as "0". The judgment results of multiple event patterns are arranged in bit sequence to form a fixed-length event combination data string. For example, a detection window containing six event patterns may obtain a data string of "010010".

[0048] The system performs rule matching on the data string to determine whether any bit is "1". If one or more event patterns are activated, that is, "1" appears in the data string, the system determines that there are signs of risk in the current approval process or editing process, and immediately outputs a risk calculation trigger signal. This signal serves as a control signal to send a task instruction to the data analysis module to start the calculation process of the process anomaly index and content conflict index.

[0049] The calculation method of the process anomaly index in the data analysis module is: The approval process is modeled as a directed graph structure, where nodes represent approval operations and edges represent flow directions. The graph structure is vectorized through a graph embedding algorithm, and then the abnormal subpath vector and the standard process vector are extracted. The process abnormality index FI is defined as the distance between the two: ; Embedding vector i-th dimension component of the current approval process graph, is the i-th dimension component of the embedding vector of the standard approval flowchart, is the impact factor corresponding to the i-th dimension process structure, indicating the importance weight assigned to this dimension in the approval process diagram, which is pre-set by the system based on the training data. It is a preset smoothing factor used to avoid the situation where the denominator is zero in the calculation. The value is a positive real number less than 1 (such as 0.01). It is often set to a constant to avoid division by zero errors. n is the dimension of the embedded vector.

[0050] In the present invention, the process anomaly index FI is used to measure the degree of structural deviation between the currently executed approval process and the preset standard process. The index compares the high-dimensional representation of the path after embedding in the graph structure, quantifies the differences between the two in terms of structural form, path logic or node layout, and thus reflects whether the process has deviated from the standard execution track. The larger the FI value, the more significant the difference between the current process and the standard process at the structural level, and the higher the potential risk of abnormality; when the FI value approaches zero, it means that the process operation status is highly consistent with the standard path and is within the normal range.

[0051] The calculation of FI is based on modeling the complete approval process as a graph structure, and then encoding the entire process into a multidimensional vector through a graph embedding algorithm. Each dimension in the vector, namely the i-th dimension, represents the representation information of the flowchart in a certain structural feature dimension. Each i can represent an interpretable process feature, such as the relative position of a certain type of node in the flowchart, branch depth, path closure, node aggregation, cross-layer jump probability, loop nesting degree, etc. These dimensions together constitute the embedding space of the process structure, representing the geometric characteristics of the process in the high-dimensional semantic space.

[0052] For example, in a certain dimension, if i corresponds to the concentration of conditional branch nodes in the flowchart, then this dimension can be used to measure whether there are too many nonlinear paths or redundant judgment logic in the flowchart; another dimension may correspond to the density of parallel nodes, which is used to reflect whether there are abnormal multi-role concurrent execution paths in the process; another dimension may reflect the average length of the jump path, which can be used to detect whether there is structural jump or excessive path compression in the process. When the values ​​of the current process in these dimensions are significantly different from the standard process, the distance value of the corresponding dimension will be amplified, which will eventually promote the improvement of the overall FI value, and then trigger the system to warn or intervene in the execution status of the process.

[0053] FI is not only an overall quantitative score of process behavior anomalies, but also has an interpretable internal structure. Each process structure feature represented by the i-dimension is mapped from the actual process diagram structure feature through graph embedding and data modeling, ensuring that FI is data-driven and consistent with business semantics. As the core component of the risk identification model in the present invention, this index provides a clear, quantitative, and traceable basis for subsequent risk level reasoning and response strategy decision-making.

[0054] The calculation method of the content conflict index in the data analysis module is: In the unit time window T, extract the set of content blocks edited by all users, record the editing status of each content block, and construct the editing density function: ; Same document location Edit frequency, For the The document location corresponding to the edit operation. is the document position index point currently being evaluated, m is the total number of all editing operations within the time window, is the content distribution diffusion coefficient, which indicates the degree of discreteness of editing behavior on the document. The calculated document position index point The corresponding edit density value; ; ; is the average of all edited positions; Introducing content crossover rate : ; Indicates the number of content segments where multi-user editing occurs. Indicates the total number of content segments; The maximum editing density value is recorded as , and the content crossover rate Combined to form the content conflict index: ; Represents the content conflict index.

[0055] In the present invention, the content conflict index CI is used to measure whether there is a high-risk phenomenon of the superposition of editing aggregation behavior of content areas and cross-editing behavior of multiple users in the process of collaborative editing of contract documents by multiple people. This index reflects the intensity of concentrated modification of the document within a specific time window, and the probability of conflict that may be caused by high-frequency editing of multiple users in the same area. The higher the value of CI, the greater the editing overlap in the current collaborative environment, the higher the cross-editing frequency, and the higher the risk of problems such as wrong signing, wrong modification, and loss of coverage.

[0056] The system constructs an editing density function to obtain the editing concentration of the contract content area within a unit time window, determine whether there are obvious hot spots, and calculate the content crossover rate by counting the proportion of content segments that actually undergo multi-user cross-editing. The core significance of CI lies in the fusion modeling of two types of risks: spatial density and collaborative crossover, so as to achieve a quantifiable assessment of the degree of collaborative editing chaos. In the exponential function, the square term of the content distribution diffusion coefficient is treated with a protective additive constant to avoid gradient explosion or numerical instability caused by an exponential function that is too large. CI, as a key indicator in the system to support content conflict identification and response, can achieve pre-judgment of risky behaviors and ensure the stability and controllability of the contract editing stage.

[0057] The maximum edit density value represents the document location with the most concentrated edit behaviors in the current time window, which is the "high-risk area" where version conflicts or edit conflicts are most likely to occur. By using the maximum value instead of the mean, the system can capture extreme collaborative editing risks more sensitively and timely, thereby effectively triggering the system's early warning and response mechanism. The maximum edit density value will form a content conflict index together with the content intersection rate, which is used to quantify the intensity of collaborative conflicts that may occur during multi-person editing. The content conflict index CI can not only be used to determine whether the current collaborative operation requires permission adjustment or behavior restriction, but also can be used as one of the important input variables for judging the joint risk level in fuzzy reasoning, thereby ensuring the content integrity, version clarity and signing accuracy during the contract editing stage, and improving the quality of contract collaboration and the overall stability of the contract management system.

[0058] The joint identification module performs fuzzy reasoning based on the combined results of the process anomaly index FI and the content conflict index CI, where FI and CI are divided into three levels of "low", "medium" and "high" according to the preset fuzzy membership function, for example, interval mapping is achieved through a triangular membership function, and the reasoning rules are as follows: if FI belongs to "high" and CI belongs to "high", the reasoning result is "serious risk"; If FI is "medium" and CI is "high", the inference result is "heavier risk"; If FI is "high" and CI is "medium", the inference result is "medium risk"; If both are “low”, the inference result is “normal risk”; The inference results are quantified into four levels of output signals, with corresponding values ​​of 0, 1, 2, and 3, representing normal, moderate, severe, and serious risks, respectively.

[0059] In this invention, in order to avoid unclear response strategies due to the fuzzy boundary of the risk index, the system introduces a joint risk level judgment mechanism based on fuzzy set theory. This mechanism no longer uses absolute values ​​for rigid classification, but uses fuzzy logic to map the two input risk indexes to three levels: low, medium, and high, and derives the overall joint risk level based on the combined relationship between the two.

[0060] The system uses the process anomaly index and content conflict index as two-dimensional input items, defines their fuzzy membership on the numerical axis, and maps the original values ​​to membership by establishing three groups of membership functions. Each membership degree represents the possible degree to which the index belongs to a certain risk level. Subsequently, the system matches the combination of the two inputs in the fuzzy space with the four joint risk levels according to the predefined rule table, and outputs the final risk level signal to guide the response module to select the corresponding intervention strategy.

[0061] To achieve numerical processing of fuzzy logic input, the system constructs three sets of fuzzy membership functions for each of the two risk indices, corresponding to the three levels of "low", "medium" and "high". The membership functions are modeled in the form of symmetrical trigonometric functions, with adjustable control points and easy-to-implement numerical solutions. Each function has the maximum membership at its center point, extending to the left and right to form a linear descending boundary.

[0062] Taking a certain input as an example, its "medium" level trigonometric function has a symmetrical peak interval, and the left and right boundaries are set to the average value between the low and high threshold points of the index. The construction method of the trigonometric function makes the boundary transition natural, effectively avoiding the misjudgment of the critical value. At the same time, the fuzzy processing of the process anomaly index and the content conflict index are independent of each other, which is convenient for modeling and maintenance. In the entire two-dimensional fuzzy space, the system completes the evaluation of nine groups of combined situations through a matrix combination method, corresponding to four output risk levels: normal, medium, heavy, and severe.

[0063] In the fuzzy rule base, the system presets several basic rules. For example, when the process anomaly index is high and the content conflict index is also high, the system will output a "serious risk" signal; when the process anomaly index is medium and the content conflict index is high, it will output a "heavy risk" signal; when the process anomaly index is high and the content conflict index is medium, it will output a "medium risk" signal; when both are low, it will output a "normal risk" signal.

[0064] The final risk level result is determined by the maximum membership value calculated by the fuzzy inference engine. The output risk level is mapped to a fourth-order discrete control signal corresponding to four strategy levels. The signal is input to the response module to determine whether to perform operations such as path reconstruction, authority restriction, logging or manual intervention. Through the fuzzy inference mechanism, the system realizes the automatic conversion from fuzzy data to clear response strategy, making risk judgment more stable and adaptive in an uncertain environment.

[0065] The response decision module includes the following strategy selection mechanisms: If the risk level is serious, the approval process is suspended, all editing permissions are locked, and the administrator is forced to intervene; when the fuzzy reasoning module determines that the current joint risk level is serious, it means that the approval process structure deviates significantly from the standard path, and the editing behaviors of multiple people are highly overlapping in the same area, and there is a serious risk of overstepping authority, skipping review, or wrong version signing. In this case, the response decision module immediately triggers the highest level of intervention mechanism of the system. The system first suspends all approval processes of the current contract, freezes the approval status, and prevents the process from continuing to advance; at the same time, it locks all editing permissions of the current document to prevent any user from making changes. Subsequently, the system pushes a high-priority intervention notification to the designated administrator or contract person in charge, forcing manual intervention. The notification may include the contract number, trigger time, two index values, and diagnostic suggestions. The administrator needs to repair the process structure or review and confirm the document content before unlocking the state and reactivating the process. Through this level of response, the system ensures that in high-risk scenarios, the risk spread can be controlled in the first time to ensure the legality of the contract and data security.

[0066] If the risk level is high, restrict multiple people from editing and rebuild the conflicting nodes in the approval path; when the system determines that the joint risk level is high, it usually means that there are certain abnormalities in the process configuration, such as abnormal frequent node jumps, increased possibility of path interruption, and although the editing behaviors of multiple people do not completely overlap, there is an obvious cross-trend in the adjacent content area. Although such risks have not reached an uncontrollable state, they are very likely to escalate into serious problems if not intervened in time. Therefore, the system will automatically perform moderate intervention operations. Specifically, it includes: dynamically restricting the collaborative editing permissions of documents, allowing only one user to modify in the key content area, and other users enter read-only or delayed editing status to avoid the expansion of conflicts; at the same time, the system identifies and reconstructs the local path of the problem nodes with frequent abnormal jumps or repeated node processing in the approval path, and updates the abnormal boundaries in the flowchart structure. The system simultaneously records the intervention behavior and sets the monitoring cycle. If a number of pre-invalidations or risks rise again, it will automatically upgrade to the serious level processing process. This strategy takes into account both automatic prevention and control and manual reservation to ensure that the risk growth trend is suppressed in a timely manner within a controllable range.

[0067] If the risk level is medium, an early warning will be issued and a risk log will be recorded; when the system identifies that the joint risk level is medium, it indicates that the process structure and content collaboration are still within the overall controllable range, with only slight fluctuations in some indicators. For example, the approval processing time is partially abnormal, and the editing behavior overlaps in some paragraphs but does not reach the level of conflict. In this case, the system adopts a low-intensity response mechanism with early warning as the main method. The response module will generate a complete risk early warning log, which includes the process structure deviation, user editing behavior trend chart, triggered event pattern type and expected impact range. The log will be stored in the risk file area of ​​the contract management platform, and a notification will be sent to the contract collaboration group or the responsible person's email address at the same time, prompting attention to the current risk behavior. If the user takes the initiative to adjust or the behavior naturally stabilizes, the system will automatically lift the early warning state; if the risk index rises, it will enter a heavier level processing process. This mechanism reflects the system's flexible judgment ability, avoids unnecessary interference in the normal collaboration process, and retains the audit record of the entire process.

[0068] If the risk level is normal, no intervention is made but the behavior data is archived. When the joint risk level is judged to be normal, it means that the current contract approval process structure is highly consistent with the standard template, and no abnormal path behavior is detected. At the same time, there is no high-frequency staggered or content aggregation risk in the collaborative editing process. In this scenario, the system does not actively perform any intervention operations, and contract approval and editing behaviors can proceed normally. However, in order to support subsequent behavior analysis and strategy optimization, the system will package and archive the current approval process and collaborative behavior data, and store them in the data training library of the risk identification module for updating risk models and indicator weights. This type of "normal sample" will be contrasted with subsequent risk events to enhance the system's risk identification capabilities and the adaptability of fuzzy rules, thereby continuously optimizing model performance. This mechanism achieves long-term evolution and self-optimization of system performance while ensuring business efficiency.

[0069] The response decision module also includes a supervision mechanism: After each risk response is triggered, the following three indicators are recorded: process repair success rate, conflict reduction rate, and approval acceleration. These three indicators are then subjected to polynomial regression modeling. Based on the output data of the polynomial regression modeling, namely the improvement value, the least squares method is used to fit a fitting straight line between the improvement value and the response time. The slope of the fitting straight line is compared with the preset standard slope. If the slope of the fitting straight line is greater than or equal to the preset standard slope, the current response is valid. If the slope of the fitting straight line is less than the preset standard slope, the current response is invalid, and the manual intervention window mode is entered to force the administrator to intervene.

[0070] The process repair success rate is used to measure whether the system has successfully restored the abnormal process to a complete closed-loop state after executing the risk response strategy. The system tracks the process operation status after each triggered response. If the contract process finally reaches the termination node smoothly (such as the contract has been signed and the approval process has been completed), the response is counted as "successful", otherwise it is "failed". During the statistical period, the number of successes is divided by the total number of responses to obtain the process repair success rate of the strategy. This indicator mainly reflects the strategy's ability to recover from structural abnormalities and is a key dimension to measure the effectiveness of process stability repair.

[0071] The conflict reduction rate reflects whether the response strategy has significantly reduced the number of conflict events in which multiple people edit content before and after execution. The system will collect the number of editing conflicts per unit time before and after the strategy is executed. For example, the behavior of paragraph content being repeatedly overwritten or cross-edited by different users in a short period of time is recorded as a conflict event. The conflict reduction rate is compared by comparing the data collected twice, specifically the reduction in the number of conflicts divided by the total number of conflicts in the previous period. This indicator is used to evaluate the control effect of the strategy in terms of content editing consistency and version security.

[0072] The approval acceleration is used to measure whether the response strategy has improved the efficiency of the overall approval process after execution. The system calculates the total time taken for contract approval from the start of the process to the end of the process, and averages multiple contract instances. The average approval time before and after the response is compared to obtain the approval acceleration. This indicator measures the contribution of the strategy to improving approval efficiency, and is particularly suitable for evaluating issues such as process congestion and node delays.

[0073] The core goal of the supervision mechanism is to determine the continuous improvement capability of a strategy in actual application. The system fits the response effect by constructing a polynomial regression model, and uses the least squares method to fit the slope of the trend curve of the improvement value evolving with the number of responses, and then determines whether the strategy has positive optimization potential. When the fitted trend slope is higher than or equal to the preset evaluation standard, it means that the strategy has gradually produced positive effects in multiple consecutive executions, and the system marks it as "effective"; if the slope is lower than the standard, it means that the strategy effect is unstable or ineffective for a long time, the system will suspend automatic response, and force the manual intervention window mode to be entered, and the administrator will re-evaluate the process and document status.

[0074] Implementation method 1: Cross-departmental contract collaborative approval scenario for large enterprises: In a large enterprise with multiple business units, electronic contracts often involve collaborative approvals by multiple business departments, with complex process levels and numerous roles, and there is a high risk of path configuration conflicts and approval interruptions. The implementation method of this system in this scenario is: establish a global node map through the process monitoring module, and use the graph neural structure to extract the logical jump relationship between the nodes of each department; combined with the time matrix analysis, the system can find that there are problems such as cross-department transfer interruptions or large fluctuations in approval response time in the approval node. During the operation of the system, once abnormal behaviors such as role duplication, path loops, or continuous timeouts of approval actions are identified in the approval path, the risk sign detection module triggers a risk calculation signal, and then quantifies the risk level through the process anomaly index model. For the severity level, the system will suspend the current process and execute the node reconstruction strategy, and notify the relevant business person in charge to confirm the path, so as to ensure that the contract approval can be completed in an online closed loop.

[0075] Implementation method 2: Multi-person contract editing scenario in a remote office environment: In the enterprise remote office mode, multiple employees may edit the same draft contract at different times and locations, which may easily cause version confusion and wrong signing problems. The implementation method of this system in this scenario is: enable the behavior coding submodule in the editing tracking module to record the editing action sequence of all online users, including editing time, location and paragraph content; at the same time, build a collaborative interaction matrix to analyze whether there are multiple users frequently editing adjacent paragraphs in a short period of time. When high-frequency interlaced editing behavior is identified in the unit time window, the system marks it as a content conflict risk and starts the content conflict index calculation mechanism to judge the degree of conflict by combining the diffusion coefficient and the cross rate. If it is judged to be a medium or higher risk level, the response decision module will automatically lock the key area being operated by multiple users, and only retain the priority editor's permissions. At the same time, it will send collaborative editing reminders to encourage users to avoid editing conflicts. Finally, the regression mechanism is used to evaluate whether the intervention effect has achieved collaborative stability.

[0076] Implementation method 3: Automatic strategy evolution scenario driven by contract approval efficiency optimization: In business scenarios that are sensitive to approval efficiency, such as the procurement contract signing process in the supply chain, enterprises are concerned not only about whether the approval is successful, but also about the approval speed and editing stability. The implementation method of this system in such scenarios is as follows: the system enables the risk identification and response mechanism by default, but emphasizes the effectiveness and self-optimization ability of the intervention strategy. After each response operation (such as flow-limited editing, node reconstruction, authority constraints, etc.) is executed, the system will record three feedback indicators: process repair success rate, conflict reduction rate, and approval acceleration, and use them as input to build a polynomial regression prediction model to fit the trend curve between the score output and the number of responses. The system then calculates the slope of the trend and compares it with the preset standard to determine whether the current strategy is effective. If the trend is positive and the growth rate is good, the system automatically records the strategy as a recommended path; if the trend slope continues to decline, the manual intervention module is triggered, and the administrator evaluates the applicability of the current strategy, supports strategy replacement or response plan adjustment, thereby forming a data-driven risk response self-optimization capability.

[0077] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0078] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0079] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example 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. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0080] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0081] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An electronic contract approval management system, characterized in that: include: The process monitoring module is used to obtain dynamic behavior data in the electronic contract approval process, including approval path flow, processing duration, and transfer failure events; The editing tracking module is used to collect editing trajectory data of contract documents in multi-person editing scenarios, including editing time distribution, content coverage area, and behavioral interaction sequences between concurrent users; The risk sign detection module is used to analyze process behavior data and edit trajectory data, and output a risk calculation trigger signal after detecting an abnormal event combination pattern; The data analysis module responds to risk trigger signals, builds a multidimensional data set, and calculates two core indexes for joint risk identification, namely, the process anomaly index and the content conflict index, for downstream modules to call; The joint identification module performs fuzzy reasoning based on the process anomaly index and content conflict index to classify the joint risk level; The response decision module performs differentiated interventions based on risk level results, including process correction, authority constraints, notification push, and log archiving.

2. According to claim 1, an electronic contract approval management system is characterized in that: The process monitoring module further includes a node graph construction submodule and a time matrix analysis submodule. The node graph construction submodule abstracts the approval process into a node-edge data structure based on the graph neural structure, identifies disconnected segments and abnormal transfer paths, and the time matrix analysis submodule records the average residence time and fluctuation range of each node by constructing a time-node two-dimensional tensor structure. If the path jump probability exceeds the preset jump threshold, or the processing time standard deviation of a node exceeds the preset ratio threshold of the average processing time standard deviation of its adjacent nodes, it is judged as an abnormal process behavior event.

3. The electronic contract approval management system according to claim 2, characterized in that: The editing tracking module further includes an editing behavior encoding submodule and a collaborative interaction matrix generation submodule. The former serializes and encodes the operations of each user to generate an editing behavior vector; the latter forms a user interaction density matrix by establishing an editing cross-relationship table between users. If more than q users have high-frequency interlaced editing behaviors in adjacent areas within a certain time window, it is identified as a conflict editing risk. All encoded data and matrix data enter the content conflict index construction model to provide input for subsequent conflict intensity quantification. High-frequency interlaced editing behavior refers to the editing update frequency exceeding the preset editing threshold.

4. The electronic contract approval management system according to claim 3, characterized in that: The risk sign detection module adopts a combined rule judgment method to identify the continuous occurrence of preset event patterns. When the corresponding event pattern appears, it is marked as 1, and when it does not appear, it is marked as 0, and an event combination data string composed of 1 and 0 is obtained. If 1 appears in the event combination data string, a risk calculation trigger signal is output to trigger the subsequent data analysis module to enter the calculation state.

5. The electronic contract approval management system according to claim 4, characterized in that: The calculation method of the process anomaly index in the data analysis module is: The approval process is modeled as a directed graph structure, where nodes represent approval operations and edges represent flow directions. The graph structure is vectorized through a graph embedding algorithm, and then the abnormal subpath vector and the standard process vector are extracted. The process abnormality index FI is defined as the distance between the two: ; Embedding vector i-th dimension component of the current approval flow chart, is the i-th dimension component of the embedding vector of the standard approval flowchart, is the impact factor corresponding to the process structure of the i-th dimension, indicating the importance weight assigned to this dimension in the approval process diagram. is a preset smoothing factor used to avoid the situation where the denominator is zero in the calculation, and n is the dimension of the embedded vector.

6. The electronic contract approval management system according to claim 5, characterized in that: The calculation method of the content conflict index in the data analysis module is: In the unit time window T, extract the set of content blocks edited by all users, record the editing status of each content block, and construct the editing density function: ; Same document location Edit frequency, For the The document location corresponding to the edit operation. is the document position index point currently being evaluated, m is the total number of all editing operations within the time window, is the content distribution diffusion coefficient, which indicates the discrete degree of editing behavior on the document. The calculated document position index point The corresponding edit density value; ; ; is the average of all edited positions; Introducing content crossover rate : ; Indicates the number of content segments where multi-user editing occurs. Indicates the total number of content segments; The maximum editing density value is recorded as , and the content crossover rate Combined content conflict index: ; Represents the content conflict index.

7. An electronic contract approval management system according to claim 6, characterized in that: The joint identification module performs fuzzy reasoning based on the combined results of the process anomaly index FI and the content conflict index CI, where FI and CI are divided into three levels of "low", "medium" and "high" according to the preset fuzzy membership function. The reasoning rules are as follows: If FI is "high" and CI is "high", the inference result is "serious risk"; If FI is "medium" and CI is "high", the inference result is "heavier risk"; If FI is "high" and CI is "medium", the inference result is "medium risk"; If both are "low", the inference result is "normal risk"; The inference results are quantified into four levels of output signals, with corresponding values ​​of 0, 1, 2, and 3, representing normal, moderate, severe, and serious risks, respectively.

8. An electronic contract approval management system according to claim 7, characterized in that: The response decision module includes the following strategy selection mechanisms: If the risk level is severe, the approval process is suspended, all editing permissions are locked, and administrators are forced to intervene; If the risk level is high, restrict multiple people from editing and rebuilding the conflicting nodes in the approval path; If the risk level is medium, issue a warning and record a risk log; If the risk level is normal, no intervention is made but the behavior data is archived.

9. The electronic contract approval management system according to claim 8, characterized in that: The response decision module also includes a supervision mechanism: After each risk response is triggered, the following three indicators are recorded: process repair success rate, conflict reduction rate, and approval acceleration. These three indicators are then subjected to polynomial regression modeling. Based on the output data of the polynomial regression modeling, namely the improvement value, the least squares method is used to fit a fitting straight line between the improvement value and the response time. The slope of the fitting straight line is compared with the preset standard slope. If the slope of the fitting straight line is greater than or equal to the preset standard slope, the current response is valid. If the slope of the fitting straight line is less than the preset standard slope, the current response is invalid, and the manual intervention window mode is entered to force the administrator to intervene.

Images