A cross-platform operation task collaborative allocation method and system based on feature fusion

Abstract

The present application relates to enterprise management and data processing technical field, specifically to a kind of cross-platform operation task collaborative allocation method and system based on feature fusion, comprising: the uniform operation situation view is formed by integrating cross-platform desensitization element feature, and business entity relationship diagram is constructed based on this;The association of node and attribute in the graph is analyzed using entity association analysis model to discover collaborative workflow requirements;And according to the desensitization element feature combination of demand candidate, adapt workflow template to determine task process, while evaluating and recommending the corresponding candidate;Finally, task scheduling instructions are assigned to related platforms and performance meta-feature feedback is monitored to achieve closed-loop management.The present application realizes intelligent discovery, automation planning and closed-loop optimization of cross-platform collaborative tasks through feature fusion technology, improving the accuracy of enterprise operation efficiency and resource allocation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of enterprise management and data processing, in particular to a cross-platform operation task collaborative allocation method and system based on feature fusion. BACKGROUND

[0002] In modern enterprise operation, multiple heterogeneous business platforms such as customer relationship management and enterprise resource planning operate in parallel, forming an "information island". When tasks need to be processed collaboratively across platforms, for example, to determine the delivery priority by integrating customer value, order information and inventory status, enterprises often rely on inefficient and error-prone manual coordination, which directly damages service quality and operational efficiency.

[0003] To solve this problem, existing technologies have made some explorations. One approach is to aggregate multi-source data through data warehouses or data lakes, but such solutions not only require huge investment, but also have data privacy and security risks, and their analysis capabilities are mostly limited to static rules, making it difficult to mine deep dynamic associations between cross-platform business entities (such as customers, orders, products, etc.), thus unable to actively identify complex collaboration needs.

[0004] On the other hand, although advanced AI technologies such as deep reinforcement learning are applied in the field of robots to achieve dynamic task collaboration, the models are designed for the state of the physical world (such as location, power, etc.), and cannot be directly applied to the highly abstract business data and workflows in enterprise operation. Therefore, after identifying the collaboration needs, how to automatically generate task processes and match the optimal human resources is still a major gap in existing technologies.

[0005] More importantly, current technologies generally lack an effective closed-loop feedback mechanism. Performance data (such as processing time, customer satisfaction) after task execution is often passively archived and cannot be systematically used to benefit and optimize the front-end decision-making model, business rules and workflow templates, resulting in the entire collaboration system being unable to learn from experience and evolve itself.

[0006] Therefore, a cross-platform operation task collaborative allocation method and system based on feature fusion is proposed. SUMMARY

[0007] The present application aims to provide a cross-platform operation task collaborative allocation method and system based on feature fusion. The present application integrates cross-platform de-sensitized meta-features to form a unified operational situation view, and builds a business entity relationship graph based on it. The entity association analysis model is used to analyze the association of nodes and attributes in the graph to discover collaborative workflow requirements. According to the candidate de-sensitized feature combinations, the workflow templates are adapted to determine the task process, and the corresponding candidates are recommended for evaluation. Finally, task scheduling instructions are assigned to related platforms and performance meta-features feedback is monitored to realize closed-loop management.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A cross-platform collaborative task allocation method based on feature fusion includes:

[0010] Anonymized meta-features representing the enterprise's business status are generated locally on multiple heterogeneous operating platforms; the collaborative decision engine integrates the anonymized meta-features to form a feature view reflecting the enterprise's cross-platform aggregated operation status.

[0011] The collaborative decision engine constructs the feature view into a business entity relationship graph, uses the entity association analysis model to analyze the association between entity nodes and their attributes in the graph, identifies combinations of entity nodes, and marks them as potential collaborative workflow requirements.

[0012] The collaborative decision engine adapts workflow templates from the workflow template library based on the combination of de-identified meta-features corresponding to the collaborative workflow requirements to determine the task flow; at the same time, the resource recommendation engine evaluates and recommends candidates for performing the task flow based on the availability and skill profile meta-features of enterprise personnel.

[0013] Based on the task flow and candidates, the collaborative decision engine assigns task scheduling instructions to relevant heterogeneous operation platforms and continuously monitors the meta-feature feedback reflecting task execution performance to achieve closed-loop management of the collaborative workflow.

[0014] Preferably, the step of generating de-identified meta-features representing the enterprise's business status locally on multiple heterogeneous operating platforms includes: deploying feature agents on each heterogeneous operating platform side to perform calculations based on multi-level business rules; the multi-level business rules include basic layer rules, combined layer rules, and trend layer rules; the basic layer rules and combined layer rules are calculated based on business events, and the trend layer rules are calculated periodically based on a preset time window to analyze time series data; the feature agents report the de-identified entity identifiers, meta-feature labels, and their confidence scores associated with the business entities to the collaborative decision engine.

[0015] Preferably, the step of the collaborative decision engine constructing the feature view into a business entity relationship graph includes: the collaborative decision engine instantiating business entities including customers, orders, products, and work orders into nodes with unique identifiers in the business entity relationship graph based on the received de-identified entity identifiers; using the de-identified meta-features as attributes of the nodes; mapping the associations between the business entities to edges between nodes in the business entity relationship graph; and updating the topology and node attributes of the business entity relationship graph with the newly aggregated de-identified meta-features.

[0016] Preferably, the step of analyzing the association between entity nodes and their desensitized meta-feature attributes in the graph using the entity association analysis model includes: employing a heterogeneous graph attention network model; the heterogeneous graph attention network model calculates the importance weights of different types of nodes and different association paths through node-level attention and meta-path-level attention mechanisms; the heterogeneous graph attention network model aggregates neighbor node information under the guidance of meta-paths to generate an embedded vector representation of each business entity node; the collaborative decision engine identifies entity node combinations and marks them as potential collaborative workflow requirements by performing cluster analysis and anomaly detection on the embedded vectors.

[0017] Preferably, the step of adapting workflow templates from the workflow template library to determine the task flow includes: each template in the workflow template library structurally defines the task decomposition structure, the execution platform of each subtask, the prerequisite dependencies, the quality acceptance criteria, and the processing time limit; the matching process adopts vector similarity calculation based on meta-feature combination; and the template variables are automatically filled according to the specific parameters of the current collaborative workflow requirements to generate a customized task flow.

[0018] Preferably, the step of evaluating and recommending candidates for performing the task process includes: the resource recommendation engine filtering out a pool of candidates with corresponding skill profile meta-features based on the task role requirements defined in the workflow template; the resource recommendation engine constructing a multi-objective optimization model, the objective function of which includes the candidate's current workload, historical task performance score, collaboration cost, and the priority of the collaborative workflow; and the resource recommendation engine using a heuristic algorithm to calculate the multi-objective optimization model and output candidate suggestions.

[0019] Preferably, the step of continuously monitoring the meta-feature feedback reflecting task execution performance includes: during the collaborative workflow execution process, the feature agents of each relevant platform continuously monitor the key nodes of the sub-tasks and generate new performance meta-features according to the preset calculation strategy; the performance meta-features include, but are not limited to, task processing time, first response time, customer satisfaction score, and one-time resolution rate; the collaborative decision engine aggregates the performance meta-features and compares them with the processing time limit set in the workflow template to calculate the deviation value.

[0020] Preferably, the closed-loop management steps include: the collaborative decision engine archives all performance meta-features generated after a collaborative workflow is completed and associates them with the initial de-identified meta-features of the corresponding collaborative workflow requirements; periodically performs statistical analysis on historical data to mine the correlation between the combination of de-identified meta-features and the final execution performance; the analysis results are used to update the parameters of the business rules, adjust the weights of the entity association analysis model, and provide data insights and adjustment suggestions for the optimization of the workflow template; the adjustment suggestions, after review, are used to optimize the preset adjustable parameters in the workflow template.

[0021] A cross-platform operational task collaborative allocation system based on feature fusion includes:

[0022] Business Status Awareness Module: Generates anonymized meta-features representing the enterprise's business status locally on multiple heterogeneous operating platforms; the collaborative decision engine integrates these anonymized meta-features to form a feature view reflecting the enterprise's cross-platform aggregated operation status; Collaborative Requirement Identification Module: The collaborative decision engine constructs the feature view into a business entity relationship graph, uses an entity association analysis model to analyze the associations between entity nodes and their attributes in the graph, identifies combinations of entity nodes, and marks them as potential collaborative workflow requirements; Collaborative Planning Module: Based on the anonymized meta-feature combinations corresponding to the collaborative workflow requirements, the collaborative decision engine adapts workflow templates from the workflow template library to determine the task flow; simultaneously, the resource recommendation engine evaluates and recommends candidates for executing the task flow based on the enterprise's personnel availability and skill profile meta-features; Scheduling and Tracking Module: Based on the task flow and candidates, the collaborative decision engine assigns task scheduling instructions to relevant heterogeneous operating platforms and continuously monitors meta-feature feedback reflecting task execution performance to achieve closed-loop management of the collaborative workflow.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] 1. This invention processes data locally on each heterogeneous platform, extracting and transmitting only anonymized metadata, thus avoiding the direct transmission and centralized storage of raw business data. This approach reduces data security risks while acquiring the necessary information for cross-platform business status, and also reduces the complexity and cost of comprehensive system integration.

[0025] 2. This invention constructs a business entity relationship graph from business entities and their relationships from different platforms, and applies an entity association analysis model for calculation. This method can quantitatively analyze entity nodes and their different association paths, thereby identifying specific entity combinations. These combinations indicate potential collaborative processing needs, providing technical support for discovering collaborative needs that are not easily detected by conventional methods.

[0026] 3. This invention, through a workflow template library and adaptation algorithm, can match and generate specific task execution processes based on identified requirement characteristics. Simultaneously, the resource recommendation engine calculates and recommends candidates to perform the task based on established personnel skills, availability, and a multi-objective optimization model, reducing subjectivity and delays in manual coordination.

[0027] 4. This invention continuously collects performance meta-features during task execution and archives them in association with the initial requirement features that triggered the task. Statistical analysis of historical data reveals the correlation between business models and execution results. The analysis results provide data support for adjusting parameters of business rules, analysis models, and workflow templates, thereby supporting iterative improvements to the entire collaborative allocation system. Attached Figure Description

[0028] Figure 1 This is an overall flowchart of a cross-platform collaborative task allocation method based on feature fusion according to the present invention;

[0029] Figure 2 This is an overall interaction sequence diagram of a cross-platform operational task collaborative allocation method and system based on feature fusion according to the present invention.

[0030] Figure 3 This is a schematic diagram of the structure of a cross-platform operation task collaborative allocation system based on feature fusion according to the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figures 1 to 3 This invention provides a cross-platform collaborative task allocation method and system based on feature fusion, the technical solution of which is as follows:

[0033] A cross-platform collaborative task allocation method based on feature fusion includes:

[0034] Anonymized meta-features representing the enterprise's business status are generated locally on multiple heterogeneous operating platforms; the collaborative decision engine integrates the anonymized meta-features to form a feature view reflecting the enterprise's cross-platform aggregated operation status.

[0035] The collaborative decision engine constructs the feature view into a business entity relationship graph, uses the entity association analysis model to analyze the association between entity nodes and their attributes in the graph, identifies combinations of entity nodes, and marks them as potential collaborative workflow requirements.

[0036] The collaborative decision engine adapts workflow templates from the workflow template library based on the combination of de-identified meta-features corresponding to the collaborative workflow requirements to determine the task flow; at the same time, the resource recommendation engine evaluates and recommends candidates for performing the task flow based on the availability and skill profile meta-features of enterprise personnel.

[0037] Based on the task flow and candidates, the collaborative decision engine assigns task scheduling instructions to relevant heterogeneous operation platforms and continuously monitors the meta-feature feedback reflecting task execution performance to achieve closed-loop management of the collaborative workflow. Example 1

[0038] This invention discloses a cross-platform operational task collaborative allocation method and system based on feature fusion. In a typical application scenario, this invention is applied to a large enterprise with a customer relationship management system, an enterprise resource planning system, and a customer service ticketing system to improve its overall operational efficiency. (Refer to...) Figure 1 , Figure 2 This embodiment performs detection according to the process of a cross-platform operation task collaborative allocation method based on feature fusion provided by the present invention, and displays the data flow.

[0039] Furthermore, de-identified meta-features representing the enterprise's business status are generated locally on multiple heterogeneous operating platforms; the collaborative decision engine integrates these de-identified meta-features to form a feature view reflecting the enterprise's cross-platform aggregated operational status. The specific process includes:

[0040] The entire collaborative allocation process begins with real-time, lightweight awareness of the enterprise's business status. When a business event occurs, for example, a customer "Zhang San," marked as a VIP in the Customer Relationship Management System, creates a work order in the Customer Service Work Order System regarding "TV unable to connect to Wi-Fi" after purchasing a high-end smart TV through the Enterprise Resource Planning System. At this time, the feature agents pre-deployed on various heterogeneous platforms are activated. Based on preset business rules and calculation strategies, on the Customer Service Work Order System side, the feature agents immediately generate and report anonymized meta-features according to the basic layer rules: {Entity ID: "Ticket_Hash_123", Meta-feature Label: "Issue_Type_Network", Confidence: 0.9}, and {Entity ID: "Customer_Hash_ZS", Meta-feature Label: "Has_Open_Ticket", Confidence: 1.0}. Here, Customer_Hash_ZS is a unique identifier obtained after anonymizing the customer's real ID. On the Customer Relationship Management (CRM) side, the feature agent receives the Customer_Hash_ZS event, triggers a composite layer rule, queries local data, and reports the anonymized meta-feature: {Entity ID: "Customer_Hash_ZS", Meta-feature Label: "Is_VIP_Customer", Confidence: 1.0}. On the Enterprise Resource Planning (ERP) side, the feature agent also calculates and reports the anonymized meta-feature based on this identifier and another composite layer rule: {Entity ID: "Customer_Hash_ZS", Meta-feature Label: "Recent_High_Value_Purchase", Confidence: 1.0}. Simultaneously, a trend layer rule based on a 30-day time window runs periodically; if it detects frequent recent complaints from the customer, it will additionally report an anonymized meta-feature for "high complaint frequency".

[0041] By deploying feature agents locally on various platforms and adopting multi-layered business rules, standardized and multi-dimensional perception of business status was achieved. The combination of basic, composite, and trend layer rules can capture rich information from instantaneous events to long-term trends, improving the quality and depth of source feature data and providing a solid foundation for subsequent accurate decision-making.

[0042] The collaborative decision engine integrates all these anonymized meta-features from different platforms to form a feature view that reflects the cross-platform aggregated operation status of an enterprise.

[0043] Furthermore, the collaborative decision engine constructs the feature view into a business entity relationship graph, uses an entity association analysis model to analyze the relationships between entity nodes and their attributes in the graph, identifies combinations of entity nodes, and marks them as potential collaborative workflow requirements. The specific process includes:

[0044] The collaborative decision engine dynamically constructs this feature view into a business entity relationship graph. In this graph, customer Zhang San, his new order, the purchased TV product, and the newly created work order are all instantiated as uniquely identified nodes, while the newly integrated anonymized meta-features serve as attributes of the corresponding nodes. Business relationships between entities, such as "customer created work order" and "order includes product," are mapped as edges connecting the nodes. The topology and node attributes of this business entity relationship graph are updated in real time with the newly integrated anonymized meta-features.

[0045] The collaborative decision engine has a built-in association rule inferencer. When the combination pattern of a new set of aggregated desensitized meta-features matches the triggering condition of the association rule preset in the inferencer, the association rule inferencer automatically creates a new edge between the business entity nodes specified by the association rule, representing the potential business associations discovered by data-driven discovery.

[0046] This method enables graph structures to not only reflect pre-defined business logic but also to capture and solidify implicit business relationships emerging in the data. This enriches the data foundation for graph analysis, allowing subsequent entity relationship analysis models to be calculated based on a more comprehensive view that is closer to actual operational conditions, thereby improving the ability to uncover non-explicit collaborative needs.

[0047] Constructing business entities and their relationships into an entity relationship graph provides an intuitive and structured cross-platform overview of business operations. This graph structure facilitates the subsequent application of complex graph analysis algorithms, enabling a clearer understanding of the deep connections between entities and effectively supporting the identification of collaborative needs.

[0048] Subsequently, an advanced entity association analysis model begins to analyze this real-time business entity relationship graph. In this embodiment, the entity association analysis model is specifically a heterogeneous graph attention network model. Through its unique node-level and meta-path-level attention mechanisms, this model can deeply understand the importance of different types of nodes and different association paths in the graph. By aggregating neighbor node information under the guidance of meta-paths, the model generates an embedded vector representation for the node "Customer Zhang San". The collaborative decision engine performs clustering analysis and anomaly detection on this vector to uncover a potential collaborative workflow requirement and marks its key trigger de-identified meta-feature combination as {Is_VIP_Customer, Recent_High_Value_Purchase, High_Complaint_Frequency}.

[0049] Employing a heterogeneous graph attention network model effectively handles different types of business entities and their complex relationships. By distinguishing the importance of nodes and paths, the model can learn the vector representations of entities more accurately. This allows for the objective and in-depth discovery of potential collaborative needs that are difficult to detect using conventional methods, through techniques such as clustering or anomaly detection, thus improving the accuracy of need identification.

[0050] Furthermore, the collaborative decision-making engine adapts workflow templates from the workflow template library based on the de-identified meta-feature combinations corresponding to the collaborative workflow requirements to determine the task flow; simultaneously, the resource recommendation engine evaluates and recommends candidates for performing the task flow based on the availability and skill profile meta-features of enterprise personnel. The specific process includes:

[0051] Once a collaborative need is identified, collaborative planning unfolds automatically. The collaborative decision engine calculates vector similarity in the workflow template library based on the de-identified meta-features of the identified need, thereby adapting a template named "T01-VIP Customer Emergency Technical Support." This template structurally defines the task decomposition structure, the execution platform for each subtask, and processing time limits. The system then automatically populates template variables based on Zhang San's specific situation, generating a customized task flow.

[0052] The specific steps for adapting workflow templates from the workflow template library are as follows: A preliminary matching set containing multiple candidate templates is selected from the workflow template library through vector similarity calculation based on meta-feature combinations. All templates in the preliminary matching set meet a preset lower similarity threshold. For the candidate templates in the preliminary matching set, their defined task role requirements are extracted. Based on the current enterprise personnel availability and skill profile meta-features, an estimated resource satisfaction level is calculated for the candidate templates. The vector similarity of the candidate template is multiplied by its estimated resource satisfaction level to obtain a comprehensive matching score. The candidate template with the highest comprehensive matching score is selected as the final adapted workflow template.

[0053] This method incorporates human resource feasibility into the early stages of task flow determination, ensuring that the selected workflow template is not only highly relevant to collaboration needs but also currently feasible. This avoids allocation delays or failures due to subsequent resource mismatches, improving the practicality of task flow design and overall allocation efficiency.

[0054] Through a structured workflow template library and a vector similarity-based matching mechanism, rapid and accurate adaptation of collaborative task processes is achieved. The standardized definition of templates ensures the consistency and standardization of task execution, while the automatic parameter filling function provides customization capabilities for specific scenarios, balancing efficiency and flexibility.

[0055] Meanwhile, the resource recommendation engine begins evaluating and recommending the best candidates for the task process. For Level 2 technical expert tasks requiring "smart TV network module" skills, the engine first filters out a pool of candidates with the corresponding skill profile features. Subsequently, the engine constructs and solves a multi-objective optimization model whose objective function comprehensively considers the candidate's current workload, historical task performance scores, collaboration costs, and the highest priority of the current collaborative workflow. Through rapid computation using heuristic algorithms, the engine ultimately outputs candidate suggestions; for example, assigning the follow-up task to senior customer service representative Xiao Liu and recommending the technical diagnostic task to the top-performing engineer, Wang.

[0056] The specific calculation method for the historical task performance score is as follows: After a candidate completes a task, all performance meta-features associated with the task are obtained; based on a preset performance dimension weight set, the quantified values ​​of all performance meta-features are weighted and summed to generate a single task performance score that represents the overall performance of the task; using a preset smoothing coefficient α, the single task performance score is calculated with the candidate's historical task performance score before performing the task, and the updated historical task performance score is equal to: the product of the smoothing coefficient α and the single task performance score, plus (1-α) and the product of the historical task performance score before performing the task; for candidates with no historical task records, their initial historical task performance score is set to a predefined baseline value.

[0057] By employing an exponentially weighted moving average algorithm, historical task performance scores can dynamically reflect the changing trends in employee work performance. Compared to a simple historical mean, this method assigns higher weight to recent task performance, making the scores more representative of an employee's current actual abilities and status, thus improving the accuracy and relevance of the multi-objective optimization model in selecting human resources.

[0058] By constructing a multi-objective optimization model for human resource recommendation, this approach comprehensively considers multiple realistic constraints such as personnel skills, workload, cost, and task priority. This method goes beyond simple skill matching, seeking optimal solutions across multiple dimensions to achieve rational allocation of human resources and improve the overall efficiency of task execution.

[0059] Furthermore, based on the task flow and candidates, the collaborative decision engine assigns task scheduling instructions to relevant heterogeneous operation platforms and continuously monitors the meta-feature feedback reflecting task execution performance, thereby achieving closed-loop management of the collaborative workflow. The specific process includes:

[0060] When Engineer Wang begins processing a work order, the feature agent continuously monitors the key nodes of the subtask. Upon Engineer Wang's initial response and contact with the customer, the feature agent generates and reports the first new performance meta-feature based on a pre-defined calculation strategy: {Entity ID: "Ticket_Hash_123", Meta-feature label: "Initial Response Time", Value: "15min"}. When the task is finally marked as "Resolved", the feature agent calculates and reports another performance meta-feature: {Entity ID: "Ticket_Hash_123", Meta-feature label: "Task Processing Time", Value: "1.5h"}. Subsequently, after the customer completes a satisfaction survey via email or SMS, the feature agent also captures and reports the performance meta-features: {Entity ID: "Customer_Hash_ZS", Meta-feature label: "Customer Satisfaction Score", Value: "5.0"} and {Entity ID: "Ticket_Hash_123", Meta-feature label: "First-Time Resolution Rate", Value: "1.0"}.

[0061] After receiving these performance meta-features, the collaborative decision-making engine aggregates them and immediately performs key comparative analysis. It compares the aggregated "task processing time" of 1.5 hours with the preset processing time of 2 hours in the "T01-VIP Customer Emergency Technical Support" workflow template, calculating a deviation of -0.5 hours. This indicates that the task was completed ahead of schedule, a positive performance signal.

[0062] By continuously monitoring and generating quantifiable performance meta-features during task execution, the system enables real-time and objective measurement of the collaborative workflow status. By comparing actual performance with preset standards, execution deviations can be identified promptly, providing accurate data for dynamic adjustments and post-event evaluations, thereby enhancing task control capabilities.

[0063] Once the entire collaborative workflow is completed, the collaborative decision engine archives all performance meta-features generated by this task and associates them with the initial de-identified meta-feature combination. Through regular in-depth statistical analysis of accumulated historical data, it uncovers strong correlations between the de-identified meta-feature combination and the final execution performance. These data insights will be used for self-optimization; for example, the analysis results can be used to update parameters of business rules, adjust the weights of entity association analysis models, and provide specific suggestions for optimizing workflow templates.

[0064] By correlating execution performance with initial demand characteristics, a data-driven self-optimizing closed loop was constructed. This mechanism can uncover the deep relationship between business models and execution results, and its analysis results can be used to iteratively optimize business rules, analysis models, and workflow templates, enabling the decision-making capability of the entire collaborative allocation system to continuously improve over time.

[0065] By generating and fusing anonymized meta-features locally on heterogeneous platforms, this invention breaks down data barriers between platforms. Entity relationship graph analysis enables the discovery of hidden collaborative needs from a global perspective and automates the matching of task flows and human resources. Ultimately, continuous optimization through closed-loop management improves the efficiency of task allocation and the level of collaboration across platforms. Example 2

[0066] This embodiment aims to detail the specific workflow of the business status perception module and the collaborative demand identification module in this invention. In the scenario described in this embodiment, a company providing enterprise-level solutions aims to proactively uncover potential significant sales opportunities from the interaction data between its customer relationship management system, product website, and analytics system using the system of this invention.

[0067] See Figure 3 The system structure of the present invention is shown in the figure.

[0068] The working process of the business status awareness module is as follows:

[0069] First, feature agents were deployed across the company's various heterogeneous operating platforms. When a series of seemingly isolated business events occur, the business status awareness module kicks in. For example, on the product website and analytics system, the feature agent detected an IP address range from an existing customer, "Company A," where multiple different employees frequently browsed and downloaded a white paper about "AI-enabled data center solutions" over three days. Based on preset combination layer and trend layer rules, the platform's feature agent calculated and generated anonymized meta-features and reported this information.

[0070] Around the same time, the account manager of client "Company A" updated a message in the company's customer relationship management system, noting that "the company recently completed its Series B financing and plans to expand its business." The feature proxy on the customer relationship management system side, based on a fundamental business rule for monitoring specific keywords, immediately captured this information and generated anonymized meta-features.

[0071] The business status awareness module integrates anonymized meta-features from two heterogeneous platforms, the product website and the customer relationship management system, in real time to form a dynamically aggregated feature view of "Company A".

[0072] Next, the collaborative requirement identification module intervenes. The business status awareness module first constructs the above feature view into a business entity relationship graph. In this graph, "Company A" exists as an entity node, and its attributes are enriched by the newly integrated de-identified meta-features (such as Focus_On_Datacenter_Solution, Recent_Major_Funding, etc.). Subsequently, the entity association analysis model in the system begins to analyze the graph.

[0073] This model, through its pre-defined meta-path and attention mechanism, calculates a combination of two cross-platform meta-features: "high-frequency attention to specific solutions" and "recent large-scale financing," which carries extremely high correlation importance weights. Individually, either feature might not be sufficient to trigger action, but their combination suggests a strong purchasing intention and ample budget. By performing anomaly detection on the embedded vector of the "Company A" node, the model identifies that this node and its associated feature combination constitute a "high-potential large customer upgrade sales opportunity" that significantly differs from typical customer behavior patterns. Finally, the collaborative demand identification module outputs this demand candidate and passes it to the subsequent collaborative planning module for processing. Example 3

[0074] This embodiment follows the stage where the collaborative requirement has been identified, aiming to detail the specific workflow of the collaborative planning module and the scheduling tracking module in this invention. In the scenario of this embodiment, a collaborative workflow requirement of "high-priority order out-of-stock emergency transfer" has been identified, and its key anonymized metadata features are {Order_Priority: High, Local_Stock: Zero, Neighbor_Stock: Sufficient}. This task requires the collaboration of the Enterprise Resource Planning system and the Warehouse Management System.

[0075] See Figure 3 The system structure of the present invention is shown in the figure.

[0076] The collaborative planning module works as follows:

[0077] After receiving the above requirements and their meta-feature combinations, the collaborative decision engine begins collaborative planning. First, it constructs a query vector based on the meta-feature combinations, calculates vector similarity in the workflow template library, and finally adapts a pre-defined "T02-Inter-warehouse Transfer Expedited Fulfillment" workflow template. This template structurally defines the task flow, including: Task 1, picking and outbound from the source warehouse (execution platform: warehouse management system); Task 2, creating an expedited transfer order and assigning logistics (execution platform: enterprise resource planning system / logistics system); Task 3, receiving and repackaging at the target warehouse (execution platform: warehouse management system); Task 4, proactively notifying the customer and updating the estimated delivery time. The system automatically populates template variables based on the specific parameters of the current requirements, generating a customized task flow.

[0078] Meanwhile, for tasks requiring human intervention, the resource recommendation engine kicks in. This task, for example, requires a "customer care specialist." The engine first filters out a pool of candidates possessing the relevant skill profile features. Then, it constructs a multi-objective optimization model whose objective function includes the candidate's current workload, historical performance scores for handling customer communication tasks, and the priority of this urgent task. By employing a heuristic algorithm to solve the model, the engine ultimately evaluates and recommends assigning this communication task to Xiao Wang, a customer service specialist with the highest historical customer satisfaction and a relatively low current workload.

[0079] Next, the scheduling and tracking module is responsible for execution and closed-loop management.

[0080] Based on the aforementioned task flow and candidates, the collaborative decision-making engine assigns task scheduling instructions to relevant heterogeneous operation platforms. Through API calls, the engine generates a high-priority picking task for the source warehouse in the warehouse management system, creates an expedited transfer order in the enterprise resource planning system, and generates a customer communication task for specialist Xiao Wang in the customer service platform.

[0081] Throughout the collaborative workflow execution, the scheduling and tracking module continuously monitors the meta-feature feedback reflecting task execution performance. When goods are shipped from the source warehouse, the feature agent reports {Task 1 processing time: 2h}; when goods arrive at the target warehouse, it reports {transfer and transportation time: 8h}; when Xiao Wang completes customer communication and receives positive feedback from the customer, it reports {customer satisfaction score: 5.0}. The collaborative decision engine aggregates these performance meta-features and compares them with the processing time limits set in the template to calculate the deviation value.

[0082] Finally, after the collaborative workflow is completed, the scheduling and tracking module achieves closed-loop management. The module archives all performance metadata generated by this task and associates them with the initial anonymized metadata combination. The system periodically performs statistical analysis on historical data. For example, it might discover that the average actual time for a transfer from warehouse A to warehouse B is only 8 hours, while the preset time limit in the template is 12 hours. Based on this data insight, the system generates an adjustment suggestion for review by operations experts or system administrators. Once approved, the corresponding adjustable parameters in the workflow template will be automatically optimized, making the system's predictions and planning more accurate and efficient when handling similar tasks in the future.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cross-platform collaborative task allocation method based on feature fusion, characterized in that, include: Generate de-identified meta-features representing the enterprise's business status locally on multiple heterogeneous operating platforms; The collaborative decision-making engine integrates the de-identified meta-features to form a feature view that reflects the cross-platform aggregated operation status of the enterprise. The collaborative decision engine constructs the feature view into a business entity relationship graph, uses the entity association analysis model to analyze the association between entity nodes and their desensitized meta-feature attributes in the graph, identifies combinations of entity nodes, and marks them as potential collaborative workflow requirements. The steps for analyzing the association between entity nodes and their desensitized meta-feature attributes in the graph using the entity association analysis model include: employing a heterogeneous graph attention network model; the heterogeneous graph attention network model calculates the importance weights of different types of nodes and different association paths through node-level attention and meta-path-level attention mechanisms; the heterogeneous graph attention network model aggregates neighbor node information under the guidance of meta-paths to generate an embedded vector representation of each business entity node; the collaborative decision engine identifies entity node combinations and marks them as potential collaborative workflow requirements by performing cluster analysis and anomaly detection on the embedded vectors. The collaborative decision engine adapts workflow templates from the workflow template library based on the combination of de-identified meta-features corresponding to the collaborative workflow requirements to determine the task flow; at the same time, the resource recommendation engine evaluates and recommends candidates for performing the task flow based on the availability and skill profile meta-features of enterprise personnel. Based on the task flow and candidates, the collaborative decision engine assigns task scheduling instructions to relevant heterogeneous operation platforms and continuously monitors the meta-feature feedback reflecting task execution performance to achieve closed-loop management of the collaborative workflow.

2. The method for cross-platform collaborative allocation of operational tasks based on feature fusion according to claim 1, characterized in that, The step of generating de-identified meta-features representing the enterprise's business status locally on multiple heterogeneous operating platforms includes: deploying feature agents on each heterogeneous operating platform side to perform calculations based on multi-level business rules; the multi-level business rules include basic layer rules, combined layer rules, and trend layer rules; the basic layer rules and combined layer rules are calculated by being triggered by business events, and the trend layer rules are calculated periodically based on a preset time window to analyze time series data; the feature agents report the de-identified entity identifiers, meta-feature labels, and their confidence scores associated with the business entities to the collaborative decision engine.

3. The method for cross-platform collaborative allocation of operational tasks based on feature fusion according to claim 1, characterized in that, The steps of the collaborative decision engine in constructing the feature view into a business entity relationship graph include: the collaborative decision engine instantiates business entities, including customers, orders, products, and work orders, into nodes with unique identifiers in the business entity relationship graph based on the received de-identified entity identifiers; the de-identified meta-features are used as attributes of the nodes; the associations between the business entities are mapped as edges between nodes in the business entity relationship graph; and the topology and node attributes of the business entity relationship graph are updated with the newly aggregated de-identified meta-features.

4. The method for cross-platform collaborative allocation of operational tasks based on feature fusion according to claim 1, characterized in that, The steps of adapting workflow templates from the workflow template library to determine the task flow include: each template in the workflow template library structurally defines the task decomposition structure, the execution platform of each subtask, the prerequisite dependencies, the quality acceptance criteria, and the processing time limit; based on the specific parameters of the current collaborative workflow requirements, the template variables are automatically filled to generate a customized task flow.

5. The method for cross-platform collaborative allocation of operational tasks based on feature fusion according to claim 1, characterized in that, The step of evaluating and recommending candidates for performing the task process includes: the resource recommendation engine filtering out a pool of candidates with corresponding skill profile meta-features based on the task role requirements defined in the workflow template; the resource recommendation engine constructing a multi-objective optimization model, the objective function of which includes the candidate's current workload, historical task performance score, collaboration cost, and the priority of the collaborative workflow; and the resource recommendation engine using a heuristic algorithm to calculate the multi-objective optimization model and output candidate suggestions.

6. The method for cross-platform collaborative allocation of operational tasks based on feature fusion according to claim 1, characterized in that, The steps for continuously monitoring the meta-feature feedback reflecting task execution performance include: during the collaborative workflow execution process, feature agents of each relevant platform continuously monitor the key nodes of the sub-tasks and generate new performance meta-features according to a preset calculation strategy; the performance meta-features include, but are not limited to, task processing time, first response time, customer satisfaction score, and one-time resolution rate; the collaborative decision engine aggregates the performance meta-features and compares them with the processing time limit set in the workflow template to calculate the deviation value.

7. The method for cross-platform collaborative allocation of operational tasks based on feature fusion according to claim 2, characterized in that, The closed-loop management steps include: the collaborative decision engine archives all performance meta-features generated after a collaborative workflow is completed and associates them with the initial de-identified meta-features of the corresponding collaborative workflow requirements; periodically performs statistical analysis on historical data to mine the correlation between the combination of de-identified meta-features and the final execution performance; the analysis results are used to update the parameters of the business rules, adjust the weights of the entity association analysis model, and provide data insights and adjustment suggestions for the optimization of the workflow template; after review, the adjustment suggestions are used to optimize the preset adjustable parameters in the workflow template.

8. A cross-platform operational task collaborative allocation system based on feature fusion, characterized in that, Performing the method as described in claim 1 includes: Business Status Awareness Module: Generates anonymized meta-features representing the enterprise's business status locally on multiple heterogeneous operating platforms; the collaborative decision engine integrates these anonymized meta-features to form a feature view reflecting the enterprise's cross-platform aggregated operation status; Collaborative Requirement Identification Module: The collaborative decision engine constructs the feature view into a business entity relationship graph, uses an entity association analysis model to analyze the associations between entity nodes and their attributes in the graph, identifies combinations of entity nodes, and marks them as potential collaborative workflow requirements; Collaborative Planning Module: Based on the anonymized meta-feature combinations corresponding to the collaborative workflow requirements, the collaborative decision engine adapts workflow templates from the workflow template library to determine the task flow; simultaneously, the resource recommendation engine evaluates and recommends candidates for executing the task flow based on the enterprise's personnel availability and skill profile meta-features; Scheduling and Tracking Module: Based on the task flow and candidates, the collaborative decision engine assigns task scheduling instructions to relevant heterogeneous operating platforms and continuously monitors meta-feature feedback reflecting task execution performance to achieve closed-loop management of the collaborative workflow.

Images