Intelligent tracking analysis system for business

Abstract

The invention discloses an intelligent tracking analysis system for business, and particularly relates to the field of data intelligent analysis and business process modeling, the system accesses an enterprise internal multi-source heterogeneous system, collects structured and unstructured data related to the process, and constructs a behavior map model; and compressing and extracting key nodes and behavior fragments in the multi-role cooperation path. Simulating a future process trend by using a shadow business body, predicting a possible path and an accompanying risk thereof, and introducing a multi-agent strategy game mechanism to realize dynamic game and consensus formation of each role in path selection; and correcting the deviation behavior in the process by combining real-time path monitoring and an abnormal intervention mechanism. According to the method, dynamic modeling, prediction and intervention of a complex process can be realized, and the problems of approval lag, path conflict and the like are effectively avoided. Compared with a traditional control mode, the method has higher adaptability, flexibility and evolutionary ability, and is suitable for various high-collaboration scenes such as finance, government affairs and engineering.

Description

Technical Field

[0001] The present invention relates to the fields of data intelligent analysis and business process modeling, and in particular to a business intelligent tracking and analysis system. Background Art

[0002] As the level of digital office work in enterprises continues to improve, process collaboration systems have become an important part of the daily management of enterprises and institutions, and are widely used in multiple business areas such as administrative approval, project management, contract review, and human resources management. Existing process collaboration systems mostly rely on workflow engines, combined with unified form modules and permission control mechanisms to achieve scenario-based approval, cross-departmental flow, and task-driven execution. Typical systems, such as enterprise workflow systems based on BPMN (Business Process Modeling Language), or process engines integrated into OA platforms, can support basic functions such as node configuration, process redirection, and approval notifications.

[0003] Current process collaboration systems are mostly based on static process configurations, with coarse node behavior records and a lack of fine-grained behavioral data support and intelligent path prediction capabilities. In practical applications, these systems struggle to identify potential bottlenecks, path deviations, or delays within the process. Furthermore, in multi-role collaboration scenarios, inconsistent role preferences often lead to process interruptions and inefficiencies. Furthermore, existing technologies lack dynamic feedback mechanisms based on execution results, making it impossible to achieve self-optimization and evolution of process strategies. Summary of the Invention

[0004] The main purpose of the present invention is to provide an intelligent tracking and analysis system for business use, which can realize dynamic simulation, risk pre-rehearsal, deviation intervention and collaborative path optimization of business processes.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A business intelligent tracking and analysis system, comprising:

[0007] A multi-source data acquisition module collects data from multiple business systems, including structured and unstructured data, and extracts micro-action behavior features related to business entities, including click-back, field modification, task delay, page dwell time, input character deletion, and secondary submission behavior;

[0008] A behavior trajectory modeling and compression module is used to construct a behavior graph structure model based on the collected data. The nodes of the model represent specific business behaviors, and the edges represent the transfer logic between behaviors. The weights of the nodes and edges are calculated based on the behavior frequency and context relevance. The behavior graph is segmented, compressed, and temporally embedded using a graph neural network.

[0009] A shadow business entity generation engine is used to generate at least one shadow business entity for each business entity instance based on its historical behavior trajectory and contextual state. The shadow business entity is a virtual business structure used to simulate possible future behavior paths, including multiple behavior path sequences and risk deviation indicators corresponding to each sequence;

[0010] A multi-agent behavior consensus mechanism is used to propose behavior paths based on multiple shadow business entities or multiple business roles. Each of these is considered as a multiple agent for strategic game play, and a converged behavior consensus path is generated through multiple rounds of evolutionary strategy compression and federated optimization calculations.

[0011] The real-time path fitting and intervention control module is used to dynamically match and analyze the current behavior path with the consensus path or shadow body prediction path during the actual business execution process, and trigger system intervention operations when the deviation exceeds the set threshold. The intervention operations include policy adjustment, process jump, permission reconstruction or prompt information push.

[0012] Preferably, the multi-source data acquisition module includes a sub-module for identifying non-explicit behavioral features, which can identify the user's pause duration in the interface, the number of mouse track retractions, input content deletion operations and their time characteristics, and is used to enrich the behavioral modeling feature space.

[0013] Preferably, the behavior trajectory modeling and compression module uses a graph neural network with a temporal attention mechanism to model the behavior graph. The temporal attention mechanism is used to emphasize the importance of the temporal sequence between nodes to behavior inference, and to perform local path compression on the graph structure through a graph partitioning method.

[0014] Preferably, the shadow business body generation engine includes:

[0015] Behavior simulation module, used to generate multiple candidate behavior paths based on the historical behavior trajectory of business entities, environmental context and system rules;

[0016] A behavior path generation module generates multiple different behavior sequences using a model structure based on a combination of a generative adversarial network and behavior sequence modeling. The model structure includes a long short-term memory network for behavior sequence encoding and a generative adversarial network module for path diversity generation.

[0017] The risk deviation assessment module is used to calculate the risk deviation index between each candidate behavior path and the current system state. The deviation index includes path deviation, possible result failure probability and potential process conflict index, and to rank the risks of multiple paths.

[0018] Preferably, the multi-agent behavior consensus mechanism adopts the following process:

[0019] Treat each shadow business entity or business participant role as an independent intelligent entity and generate independent strategy proposals based on its path prediction results;

[0020] The system uses the evolutionary strategy compression method to screen candidate paths based on the strategy fitness scores between agents in each round, and adopts the federated optimization method to aggregate the paths to form a global behavior consensus path.

[0021] The strategy fitness score includes a path rationality score, a context fit score, and a historical effectiveness score.

[0022] Preferably, the real-time path fitting and intervention control module includes:

[0023] The behavior matching submodule is used to measure the difference between the current business execution path and the shadow path based on the path vector similarity;

[0024] Deviation calculation module, used to calculate the current execution deviation value based on the path edit distance, behavior node score difference and time delay comparison;

[0025] The intervention trigger module is used to automatically trigger the intervention strategy when the deviation value exceeds the set threshold. The intervention strategy includes: behavior redirection, process jump, task authority upgrade or user operation prompt information generation.

[0026] Preferably, it also includes a feedback learning module, which is used to update the shadow business body modeling rules and consensus strategy optimization algorithm parameters according to the actual deviation value between the behavioral results obtained after execution and the system goals, and record strategy penalties for inefficient paths to support system iterative evolution.

[0027] The present invention also discloses a method for intelligent tracking and analysis of business, the specific steps of which are as follows:

[0028] Step 1: Collect structured and unstructured data and extract explicit behavior and implicit micro-action behavior features related to business entities;

[0029] Step 2: Construct the behavior data into a graph structure model, where nodes represent business behavior units and edges represent transfer relationships between behaviors. Compress and embed the graph model using a time-dependent graph neural network.

[0030] Step 3: Generate a shadow business volume based on the behavior graph and context information. The shadow business volume contains multiple potential path sequences. Each path is generated by a generative adversarial network and a sequence prediction network, and is accompanied by a risk deviation indicator.

[0031] Step 4: Paths from multiple shadow entities or business roles are used as multiple policy proposals. A multi-agent game mechanism and federated optimization algorithm are used to select a global consensus path as the recommended behavior sequence.

[0032] Step 5: During the actual execution of the business, the deviation value of the current path and the consensus path is compared in real time. When the deviation value exceeds the preset threshold, an intervention action is triggered to control the direction of behavior evolution.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] This invention breaks through the reliance of traditional business process analysis on static rules and historical templates by introducing the "shadow business body" simulation prediction and "behavioral consensus fitting" mechanism. It can complete dynamic rehearsals before business behaviors occur, timely identify potential risks and provide targeted intervention strategies; at the same time, it realizes the coordination of behavioral paths and conflict resolution among multiple roles, significantly improving the decision-making consistency and execution efficiency in complex process environments. It has good adaptability, self-evolution ability and cross-system collaborative value, and is particularly suitable for business scenarios with high complexity, high collaboration requirements and low risk tolerance. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flowchart of the working module of the present invention;

[0036] Figure 2 Schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0037] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0038] like Figure 1 As shown, the present invention discloses a business intelligent tracking and analysis system, which specifically includes modules and the specific working process of each module is as follows:

[0039] 1. Multi-source data acquisition module.

[0040] This module is mainly used to collect business data from multiple systems and dimensions, focusing especially on "micro-actions" that are not easily perceived in user behavior but are crucial for subsequent modeling, to support the construction of more detailed behavior trajectories. The specific working process is as follows:

[0041] In actual operation, the system will first connect to multiple core business platforms within the enterprise, including customer relationship management systems (CRM), enterprise resource planning systems (ERP), approval platforms, and operation log services. Data collection can be obtained in real time or in batches, and the types of information covered include structured data (such as task nodes, form fields, time information, etc.) and unstructured content (such as operation log records, system call information, text notes, etc.);

[0042] At the same time, the system's built-in behavior monitoring components and operation interception mechanisms automatically identify detailed user behavior during the process. For example, whether the user repeatedly modifies form fields, pauses for extended periods without clicking, frequently jumps between pages or backtracks, or repeatedly deletes or re-enters information during the form. These subtle actions are often overlooked by traditional systems, but they are highly valuable for behavioral pattern analysis.

[0043] All collected data undergoes a unified formatting process. The system automatically timestamps each record, identifies the corresponding business entity number, and identifies its contextual state, such as the device used, login location, and sequence of events. This cleaned and labeled data is then passed as input to downstream behavioral modeling modules for further analysis.

[0044] 2. Behavior trajectory modeling and compression module.

[0045] This module is mainly responsible for converting a series of behaviors of business entities during operation into a structured graphical representation, and then compressing and extracting information based on this, providing a model basis for the subsequent "shadow body" generation. The specific working process is as follows:

[0046] First, in the trajectory modeling stage, the system will record the behavior process of each business entity one by one - such as the user submitting a form, approval, field modification, and other operations - and organize these behaviors into the form of a graph. Each node in the graph represents a specific action, and the connection between the nodes (that is, the edge) represents the order or dependency relationship between these actions. The system will also set weights for the edges based on the frequency and consistency of the behavior. At the same time, it will combine contextual information, such as department affiliation, operation time period, or the current load of the system, to assign certain weights to the nodes, so that the entire behavior graph is closer to the real business environment;

[0047] Next, the system uses a time-aware graph neural network, known as the GAT model with a temporal attention mechanism, to deeply model these graph structures. The core purpose of this step is to enable the model to not only understand the structural relationships between actions, but also to identify the relative importance of individual behavioral nodes at different time periods, particularly those critical nodes where process transitions or anomalies may occur.

[0048] After completing graph modeling, the system also compresses the trajectory. Using behavior clustering and path slicing algorithms, it decomposes the large behavior graph into several representative segments—called "trajectory segments." Each segment is embedded into a fixed-length vector to reduce data redundancy and facilitate subsequent prediction and deduction.

[0049] Furthermore, the system pays special attention to unusual behavioral paths. During trajectory analysis, it automatically marks nodes that occur infrequently, are prone to jumps, or have diverging paths. These are considered potential "ambiguities" and are subsequently focused on by the consensus module to identify critical turning points or potentially risky paths in the process.

[0050] 3. Shadow business body generation engine.

[0051] The core function of this module is to generate a set of "virtual mirrors"—i.e., shadow business entities—based on existing historical and current business context before actual processes are implemented. This is used to simulate various possible development paths in advance and to anticipate risks and deviations. The specific working process is as follows:

[0052] Specifically, the system first retrieves the past execution trajectory of the current business instance and extracts its behavioral graph embedding vector as the basis for modeling. It also incorporates current real-time status information, such as submission time, business type, and initiator identity, as input for behavioral simulation.

[0053] Based on this, the system initiates a path generation mechanism that uses a structure that combines an LSTM prediction model and a generative adversarial network (Shadow-GAN). The generator generates multiple possible behavior paths, while the discriminator evaluates whether these paths are reasonable and closely follow behavioral trends that have occurred in real business operations.

[0054] Each generated shadow business entity has more than one path, typically including a standard main path, evasive solutions, and potentially unusual divergent paths. The system also conducts a risk assessment for each path, scoring it based on three main dimensions: behavioral deviation, which measures the degree of difference between the current path and the standard process; potential conflict probability, which assesses whether the path is likely to interfere with other business processes or cause resource conflicts; and time expectation difference, which measures the difference between the expected time spent at each key node and the historical average.

[0055] Risk Deviation Assessment: Calculate three deviation indicators for each path: behavioral deviation (path distance from the standard flow chart); conflict risk score (probability of potential conflict with other business entities); and time expectation deviation (difference in time consumption prediction for key nodes).

[0056] Ultimately, the system will package and output the ID of each shadow body, its corresponding path sequence, and risk indicators to form a "future business evolution candidate set" for reference by subsequent consensus modules.

[0057] 4. Multi-agent behavior consensus mechanism.

[0058] The purpose of this module is to integrate multiple behavior paths generated by different shadow business entities or business roles into a global consensus path. The specific working process is as follows:

[0059] The system treats each business participant (whether a department or role) as an independent agent. Each agent proposes one or more preferred behavioral paths as strategic recommendations based on its own simulated shadow business entity.

[0060] These path strategies are then integrated into a strategic game process. During this process, the system evaluates the compatibility between different paths, including their consistency with the target task, execution efficiency, and past success rates. If any paths have obvious conflicts or low adaptability, they will be automatically eliminated to ensure the quality of the paths in subsequent negotiations.

[0061] The system aggregates the strategic models of each agent through a federated learning mechanism, integrating their preferences and weights without exposing their underlying business data. This not only protects data security but also achieves a consensus on the optimal path from a global perspective.

[0062] Ultimately, the system outputs a consensus path based on the input of all parties, weighing risks and resources. This path not only includes the execution steps but also identifies key behavioral nodes, triggering conditions, and expected costs or risk indicators, serving as a reference for subsequent implementation.

[0063] 5. Real-time path fitting and intervention control module.

[0064] The purpose of this module is to continuously track the actual business process operation trajectory and dynamically compare it with the shadow business path pre-generated by the system, so that timely intervention measures can be taken when obvious deviations are identified. The specific working process is as follows:

[0065] In actual operations, the system first tracks the execution process of the business in real time, recording the behavioral sequence formed at each operation node. These behavioral trajectories are constantly compared with the preview path. The system uses vector embedding methods or path edit distance algorithms to determine the degree of match between the current behavioral sequence and the consensus path.

[0066] Once a significant deviation in the execution path is detected, the system will further calculate the specific deviation. These deviations may manifest in various aspects, such as the misalignment of the execution order of certain key behavior nodes, a significant discrepancy between the actual execution time and the expected time, or an abnormal jump in the business process.

[0067] When deviations reach a set threshold, the system intelligently selects appropriate intervention methods based on the severity of the deviation. Minor deviations may require only interface prompts, color coding, or node highlighting. More severe deviations may prompt proactive suggestions for subsequent actions or process adjustments. In the event of significant deviations that impact the stability of business operations, the system directly intervenes, modifying process flow paths, increasing user permissions, or reassigning tasks to ensure the business remains on track.

[0068] 6. Feedback learning module.

[0069] The purpose of this module is to enable the entire system to continuously "learn from experience and adjust strategies" according to actual operating conditions, thereby achieving self-optimization and gradual evolution. The specific working process is as follows:

[0070] Specifically, after a business process is completed, the system collects key feedback from the entire process. For example, did this path ultimately achieve the intended goal? Did the system's intervention in the middle actually work? Did the user adopt the recommended course of action? All of this information will be recorded as a basis for judgment.

[0071] For those paths that fail to predict and result in large deviations in the results, the system will automatically lower their credibility, which is equivalent to "deducting points"; and for those paths that successfully guide the business to achieve its goals, the system will give "rewards" and increase their priority in future recommendations.

[0072] At the same time, the system adjusts key parameters in multiple strategy scoring models accordingly. For example, what path is most suitable for push in specific scenarios? Which behavioral characteristics have the greatest impact on the final outcome? With this real-time feedback, the entire shadow modeling logic and consensus generation mechanism are continuously optimized, making the system smarter through continuous practice and more adaptable to changes in the real business environment.

[0073] Based on the above modules, the present invention also discloses a method for intelligent tracking and analysis of business operations, the specific steps of which are as follows:

[0074] Step 1: Collect structured and unstructured data and extract explicit behavior and implicit micro-action behavior features related to business entities;

[0075] Step 2: Construct the behavior data into a graph structure model, where nodes represent business behavior units and edges represent transfer relationships between behaviors. Compress and embed the graph model using a time-dependent graph neural network.

[0076] Step 3: Generate a shadow business volume based on the behavior graph and context information. The shadow business volume contains multiple potential path sequences. Each path is generated by a generative adversarial network and a sequence prediction network, and is accompanied by a risk deviation indicator.

[0077] Step 4: Paths from multiple shadow entities or business roles are used as multiple policy proposals. A multi-agent game mechanism and federated optimization algorithm are used to select a global consensus path as the recommended behavior sequence.

[0078] Step 5: During the actual execution of the business, the deviation value of the current path and the consensus path is compared in real time. When the deviation value exceeds the preset threshold, an intervention action is triggered to control the direction of behavior evolution.

[0079] The present invention is further disclosed below with reference to specific examples:

[0080] The application context of this example is: intelligent tracking and analysis applied to the internal contract approval process of an enterprise: A large enterprise has a complex contract approval process involving multiple departments (such as the project team, legal department, finance department, and senior management approval). During the execution process, problems such as inconsistent approval paths, delayed tasks, and frequent process rollbacks often occur. In order to improve approval efficiency and risk control capabilities, the intelligent tracking and analysis system of the present invention is introduced and deployed. The implementation process is as follows:

[0081] Working process of multi-source data acquisition module:

[0082] The system is connected to the company's existing OA approval system, project management system, document management platform and email record system.

[0083] The collected data includes contract submission records, approval flow nodes, form filling content, operation logs and email interaction records.

[0084] Micro-action behaviors are captured by the system, including: multiple modifications after filling in contract fields, pauses on the approval page for a long time without taking any action, approval forms being opened multiple times without being submitted, and retry records after failed contract file uploads.

[0085] All data is timestamped, bound to the business contract ID, and uniformly cleaned and formatted before entering the behavioral trajectory modeling stage.

[0086] Behavior trajectory modeling and compression module working process:

[0087] Based on the approval history of each contract, the system constructs a graph-structured trajectory model based on "node behaviors." For example, the nodes are "Project Manager Submission" → "Legal Review" → "Financial Assessment" → "Executive Approval."

[0088] The system collects statistics on the frequency of node occurrence and transition probability across multiple instances. It also discovers that some processes experience a high-frequency cycle of "Legal Return to Project Team" → "Resubmit."

[0089] Graph neural networks are used to model these behavioral graphs and identify key behavioral paths and potential ambiguous nodes. For example, "inconsistent approval opinions" often lead to process jumps or interruptions.

[0090] The graph structure is segmented into multiple embedded segments to reduce modeling complexity and improve subsequent simulation efficiency.

[0091] Shadow business body generation engine working process:

[0092] The system generates three shadow business entities for a contract instance awaiting approval, each simulating a different future behavior path.

[0093] The first path simulates the smooth passing of the standard approval process; the second path simulates the legal department rejecting the contract due to abnormal contract terms and entering revision; the third path simulates the direct intervention of senior management for approval after the system urges the approval due to delayed approval.

[0094] Each path is assigned a corresponding risk deviation indicator, with the second path having a high "path deviation" and the third path having a significant "time delay risk." These shadow business paths are retained for use in the subsequent strategy consensus mechanism.

[0095] Working process of the multi-agent behavior consensus mechanism:

[0096] The system models the "Project Manager," "Legal Specialist," "Finance Director," and "Approval System Automated Strategy" as four behavioral agents. Each role independently proposes a recommended path based on their department's past operational habits and current contract information: the legal agent prefers a path that includes detailed review nodes; the financial agent tends to skip the legal review stage; and the automated strategy agent selects the path with the least risk based on the weights of the shadow agent's paths.

[0097] The system uses a strategic game mechanism to compare the scores of each path in terms of historical effectiveness, resource efficiency and risk control, and eliminates paths with obvious conflicts.

[0098] The final consensus path is: "Submission by project manager" → "Legal review" → "Financial estimate" → "Parallel approval by senior executives", and is accompanied by implementation recommendations and expected risk maps.

[0099] Working process of real-time path fitting and intervention control module:

[0100] After the actual contract is submitted, the system monitors the approval process in real time. If it detects a delay in processing a financial approval node, and if it deviates from the time prediction range of the shadow path, it triggers an intervention strategy: the system sends a "delay warning" to the finance staff and simultaneously advises the project manager whether to switch the approval path to executive-first approval. If the project manager accepts the system's suggestion, the process is rescheduled to the executive approval channel, avoiding the risk of contract approval timeouts.

[0101] Feedback learning module working process:

[0102] The process was ultimately completed successfully, with the adopted path largely consistent with the system consensus path, and the system concluded the prediction was successful. The system fine-tuned the modeling parameters for this round of shadow business entities, increasing the weighting of "approval path divergence" and "time delay." The system also recorded the reasons for the exclusion of the legal strategy path and appropriately lowered the scoring priority of similar strategies in subsequent training. This resulted in continuous optimization of the overall system strategy library and the accuracy of path recommendations.

[0103] This embodiment combines shadow simulation, behavior prediction, and a multi-role consensus mechanism to predict, control, and adaptively optimize complex approval processes. It effectively avoids process interruptions, role conflicts, and execution deviations, and establishes intelligent collaboration across multiple approval links, significantly improving approval efficiency and risk controllability.

[0104] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A business intelligent tracking and analysis system, characterized in that: include: A multi-source data acquisition module collects data from multiple business systems, including structured and unstructured data, and extracts micro-action behavior features related to business entities, including click-back, field modification, task delay, page dwell time, input character deletion, and secondary submission behavior; A behavior trajectory modeling and compression module is used to construct a behavior graph structure model based on the collected data. The nodes of the model represent specific business behaviors, and the edges represent the transfer logic between behaviors. The weights of the nodes and edges are calculated based on the behavior frequency and context relevance. The behavior graph is segmented, compressed, and temporally embedded using a graph neural network. A shadow business entity generation engine is used to generate at least one shadow business entity for each business entity instance based on its historical behavior trajectory and contextual state. The shadow business entity is a virtual business structure used to simulate possible future behavior paths, including multiple behavior path sequences and risk deviation indicators corresponding to each sequence; A multi-agent behavior consensus mechanism is used to propose behavior paths based on multiple shadow business entities or multiple business roles. Each of these is considered as a multiple agent for strategic game play, and a converged behavior consensus path is generated through multiple rounds of evolutionary strategy compression and federated optimization calculations. The real-time path fitting and intervention control module is used to dynamically match and analyze the current behavior path with the consensus path or shadow body prediction path during the actual business execution process, and trigger system intervention operations when the deviation exceeds the set threshold. The intervention operations include policy adjustment, process jump, permission reconstruction or prompt information push.

2. The business intelligent tracking and analysis system according to claim 1, characterized in that: The multi-source data acquisition module includes a submodule for non-explicit behavioral feature recognition, which can identify the user's pause duration in the interface, the number of mouse track retractions, input content deletion operations and their time characteristics, and is used to enrich the behavioral modeling feature space.

3. The business intelligent tracking and analysis system according to claim 1, characterized in that: The behavior trajectory modeling and compression module uses a graph neural network with a temporal attention mechanism to model the behavior graph. The temporal attention mechanism is used to emphasize the importance of the temporal sequence between nodes for behavior inference, and performs local path compression on the graph structure through a graph partitioning method.

4. The business intelligent tracking and analysis system according to claim 1, characterized in that: The shadow business body generation engine includes: Behavior simulation module, used to generate multiple candidate behavior paths based on the historical behavior trajectory of business entities, environmental context and system rules; A behavior path generation module generates multiple different behavior sequences using a model structure based on a combination of a generative adversarial network and behavior sequence modeling. The model structure includes a long short-term memory network for behavior sequence encoding and a generative adversarial network module for path diversity generation. The risk deviation assessment module is used to calculate the risk deviation index between each candidate behavior path and the current system state. The deviation index includes path deviation, possible result failure probability and potential process conflict index, and to rank the risks of multiple paths.

5. The business intelligent tracking and analysis system according to claim 1, characterized in that: The multi-agent behavior consensus mechanism adopts the following process: Treat each shadow business entity or business participant role as an independent intelligent entity and generate independent strategy proposals based on its path prediction results; The system uses the evolutionary strategy compression method to screen candidate paths based on the strategy fitness scores between agents in each round, and adopts the federated optimization method to aggregate the paths to form a global behavior consensus path. The strategy fitness score includes a path rationality score, a context fit score, and a historical effectiveness score.

6. The business intelligent tracking and analysis system according to claim 1, characterized in that: The real-time path fitting and intervention control module includes: The behavior matching submodule is used to measure the difference between the current business execution path and the shadow path based on the path vector similarity; Deviation calculation module, used to calculate the current execution deviation value based on the path edit distance, behavior node score difference and time delay comparison; The intervention trigger module is used to automatically trigger the intervention strategy when the deviation value exceeds the set threshold. The intervention strategy includes: behavior redirection, process jump, task authority upgrade or user operation prompt information generation.

7. The intelligent tracking and analysis system for business use according to claim 1, characterized in that: It also includes a feedback learning module, which is used to update the shadow business body modeling rules and consensus strategy optimization algorithm parameters based on the actual deviation value between the behavioral results obtained after execution and the system goals, and record strategy penalties for inefficient paths to support system iterative evolution.

8. A method for intelligent tracking and analysis of business, applicable to the system according to any one of claims 1 to 7, characterized in that: The specific steps are as follows: Step 1: Collect structured and unstructured data and extract explicit behavior and implicit micro-action behavior features related to business entities; Step 2: Construct the behavior data into a graph structure model, where nodes represent business behavior units and edges represent transfer relationships between behaviors. Compress and embed the graph model using a time-dependent graph neural network. Step 3: Generate a shadow business volume based on the behavior graph and context information. The shadow business volume contains multiple potential path sequences. Each path is generated by a generative adversarial network and a sequence prediction network, and is accompanied by a risk deviation indicator. Step 4: Paths from multiple shadow entities or business roles are used as multiple policy proposals. A multi-agent game mechanism and federated optimization algorithm are used to select a global consensus path as the recommended behavior sequence. Step 5: During the actual execution of the business, the deviation value of the current path and the consensus path is compared in real time. When the deviation value exceeds the preset threshold, an intervention action is triggered to control the direction of behavior evolution.

Images