Task execution method and device based on hierarchical intelligent collaboration architecture

Abstract

The application relates to the technical field of enterprise information processing, in particular to a task execution method and device based on a hierarchical intelligent collaboration architecture, which comprises the following steps: constructing a hierarchical intelligent collaboration architecture comprising an enterprise kernel layer, an industry intermediate layer and a public network large model; privately deploying an enterprise exclusive knowledge database in the enterprise kernel layer; connecting public resource data on the Internet to the public network large model; and depositing and structuring an industry solution database in the industry intermediate layer; receiving a task instruction, decomposing the task instruction into executable subtasks and planning an optimal execution sequence; calling knowledge provided by any one or more layers of the enterprise kernel layer, the industry intermediate layer and the public network large model to execute the task, and recording task execution effect data; taking the task execution effect data as a core feedback signal, optimizing a task execution strategy based on a reinforcement learning mechanism, and updating the databases of the enterprise kernel layer and the industry intermediate layer. The scheme can improve the task execution efficiency and quality of AI digital employees of an enterprise.

Description

Technical Field

[0001] This application relates to the field of enterprise information processing technology, and more specifically, to a task execution method and apparatus based on a hierarchical intelligent collaboration architecture. Background Technology

[0002] With the acceleration of digital transformation, enterprise business models are becoming increasingly diversified, and the complexity of business processes is growing exponentially. At the same time, Large Language Models (LLM) have demonstrated outstanding capabilities in natural language processing and knowledge generation, providing enterprises with new possibilities for creating AI digital employees.

[0003] AI digital employees can simulate human employees in making business decisions and delivering results, helping companies reduce costs, increase efficiency, and undergo digital transformation. However, a single model often struggles to fully address the diverse task requirements within an enterprise, especially when faced with complex business processes and decisions. It may fail to fully utilize proprietary information such as historical experience, business processes, and customer data, resulting in AI performing less accurately in specific business scenarios.

[0004] Therefore, enterprises need an intelligent digital employee system that can combine multi-model capabilities, internal knowledge resources, and external public resources, while also possessing task decomposition, scheduling, and self-optimization capabilities, to meet the needs of efficient execution and accurate decision-making in complex business environments. Summary of the Invention

[0005] The main purpose of this application is to provide a task execution method, device, computing equipment and storage medium based on a hierarchical intelligent collaboration architecture. It addresses the pain points commonly encountered in the process of enterprise application of large models, such as the risk of core data leakage, the lack of understanding of enterprise-specific business by public network large models, insufficient industry knowledge accumulation and rapid knowledge obsolescence. It constructs an AI digital employee empowerment architecture with an intelligent execution center and a hierarchical knowledge system, which not only ensures the security of enterprise private information, but also achieves a dual improvement in operational efficiency and service quality.

[0006] To achieve the above objectives, the first aspect of this application proposes a task execution method based on a hierarchical intelligent collaborative architecture, comprising: Construct a layered intelligent collaboration architecture that includes an enterprise kernel layer, an industry middleware layer, and a public network big model. The enterprise kernel layer is privately deployed with an enterprise-specific knowledge database, the public network big model connects to public internet resource data, and the industry middleware layer accumulates and structures an industry solution database. Receive task instructions, identify the logical dependencies of tasks through semantic analysis, decompose the task instructions into executable subtasks and plan the optimal execution order; It can invoke knowledge provided by any one or more layers of the enterprise kernel layer, industry middleware layer, and public network big model to execute tasks and record task execution effect data. Using task execution performance data as the core feedback signal, the system optimizes task execution strategies based on reinforcement learning mechanisms and updates the databases of the enterprise kernel layer and the industry middleware layer.

[0007] In some embodiments of this disclosure, constructing a layered intelligent collaboration architecture comprising an enterprise kernel layer, an industry middleware layer, and a public network big data model includes: Collect internal business data, process specifications, and historical experience data to establish a proprietary enterprise knowledge database that includes customer preferences and service optimization solutions. Clean, standardize, and de-identify the collected data, extract entities and relationships, and construct an enterprise knowledge graph. The industry solution data, including information on equipment failure solutions, customer service complaint solutions, and emergency response solutions, will be structured and processed to build an industry knowledge graph. We selected a large public network model and used internal enterprise data to fine-tune the pre-trained model. Based on the large public network model, we obtained public resources including authoritative external data sources, industry standard databases, and academic paper repositories.

[0008] This solution targets enterprise-level vertical scenarios and constructs a three-layer architecture: an enterprise kernel layer, an industry middleware layer, and a public network big model. It clearly defines the boundaries of rights and responsibilities for each layer: the enterprise kernel layer privately deploys core business data to ensure data sovereignty; the industry middleware layer accumulates cross-enterprise anonymized high-frequency industry solutions to avoid redundant construction; and the public network big model layer supplements public authoritative resources.

[0009] In some embodiments of this disclosure, internal business data, process specifications, and historical experience data are collected to establish a proprietary enterprise knowledge database containing customer preferences and service optimization solutions. The collected data undergoes data cleaning, standardization, and anonymization processing. Entities and relationships are extracted, and an enterprise knowledge graph is constructed, including: Collect structured, semi-structured, and experiential knowledge from within the enterprise, including basic customer information, internal business process documents, historical maintenance work orders, internal training materials, and enterprise-specific operating procedures. Based on the collected knowledge data, non-standardized descriptions are unified into business tags, and sensitive customer information is encrypted and subject to hierarchical access control. The cleaned knowledge is classified according to business scenarios, labeled with structured tags, and stored in the corresponding knowledge database. Based on the rules of the knowledge system defined by the actual business of the enterprise, all entities are extracted from the knowledge database, different identifiers of the same entity are merged, implicit relationships are mined from historical data, the knowledge network is completed, and the enterprise knowledge graph is obtained.

[0010] In some embodiments of this disclosure, industry solution data, including information on equipment failure solutions, customer service complaint solutions, and emergency response plans, is structured and processed to construct an industry knowledge graph, including: Based on the characteristics of the cultural and tourism industry, general knowledge for the cultural and tourism industry has been compiled, including a library of solutions to common problems reported by hotel customers, standard procedures for handling customer service complaints, a teaching resource library for network engineering courses, industry standards for the operation and maintenance of intelligent devices, and the latest strategies for the cultural and tourism industry. Extract key entities and their relationships from general knowledge in the cultural and tourism industry, construct entity-relationship pairs, and structure the general knowledge of the cultural and tourism industry into an industry knowledge graph.

[0011] This solution addresses actual business scenarios for enterprises by performing business tagging, entity extraction, relationship mining, anonymization, and permission-based classification on structured, semi-structured, and experience-based knowledge, thereby constructing an enterprise knowledge graph that aligns with the enterprise's business. Simultaneously, it develops structured solutions for high-frequency industry scenarios (such as hotel equipment malfunctions, customer service complaints, and emergency response in the example), constructing industry knowledge graphs that significantly improve the efficiency and accuracy of knowledge matching from a knowledge storage perspective.

[0012] In some embodiments of this disclosure, the task execution is performed by invoking knowledge provided by any one or more layers of the enterprise kernel layer, industry middleware layer, and public network big data model, and the task execution effect data is recorded, including: The system calls the enterprise kernel layer to retrieve the enterprise knowledge graph and match the corresponding solution. If no matching result is found, it calls the industry middleware layer to retrieve the industry knowledge graph and match the corresponding solution. If no matching result is found, it calls the public network big model to obtain task-related information from public internet resources, performs semantic analysis and filtering on the task-related information, generates a solution, and generates customer communication scripts or operation and maintenance guidelines based on the retrieved or generated solution.

[0013] This solution identifies the logical dependencies of tasks through semantic analysis, breaks down tasks into executable subtasks and plans the optimal execution order, and then calls knowledge according to the gradient logic of enterprise kernel layer → industry middle layer → public network big model. It only calls down the layer when there is no matching result in the current layer, which not only prioritizes the matching rate of enterprise-specific knowledge, but also avoids ineffective calls to public network resources.

[0014] In some embodiments of this disclosure, using task execution performance data as the core feedback signal, optimizing task execution strategies based on reinforcement learning mechanisms, and updating the databases of the enterprise kernel layer and the industry middleware layer include: Record the complete chain of each task execution, collect feedback information including customer satisfaction, problem resolution rate, task execution time, and whether compliance issues have occurred, and use the feedback information as a reward signal to adjust the execution logic; The solutions generated from the public network big data model are synchronized to the enterprise kernel layer and the industry middleware layer after being reviewed. The general solutions adapted to the enterprise are synchronized to the enterprise kernel layer after being reviewed by experts. The new solutions accumulated in the enterprise kernel layer are synchronized to the industry middleware layer after being de-identified.

[0015] This solution uses task execution performance data as the core feedback signal. Through a reinforcement learning mechanism, it synchronously optimizes task execution strategies and knowledge base content. Solutions with good execution performance are automatically accumulated in the corresponding knowledge layer, while strategies with poor performance are automatically adjusted. At the same time, it establishes cross-layer knowledge synchronization rules (solutions generated on the public network are synchronized to the enterprise / industry layer after review, general enterprise solutions are synchronized to the industry layer after review, and unique enterprise experience is synchronized to the industry layer after anonymization). This enables synchronous iteration of execution capabilities and knowledge reserves, solving the problems that static knowledge bases cannot adapt to dynamic business and execution strategies cannot be continuously optimized.

[0016] In some embodiments of this disclosure, the enterprise kernel layer is used to integrate internal business data, process specifications, and historical experience data to construct an enterprise knowledge graph. It also performs data cleaning, standardization, and anonymization on internal data and updates the enterprise-specific knowledge database based on new business knowledge received from task execution feedback. The industry middleware layer is used to accumulate standardized processing solutions for high-frequency industry scenarios, construct an industry knowledge graph, regularly synchronize the latest industry specifications and standards obtained from the public network, update the industry solution database, and anonymize the unique business experience accumulated by the enterprise kernel layer. The public network big model utilizes public resources including external authoritative data sources, industry standard databases, and academic paper repositories to send the acquired effective external information back to the enterprise kernel layer to update the enterprise-specific knowledge database and back to the industry middleware layer to update the industry solution database.

[0017] This solution uses a three-tiered architecture to ensure that core sensitive enterprise data does not need to leave the internal network. Only general solutions and public resources that need to be accessed have been de-identified and audited, which not only meets industry compliance requirements, but also breaks the closed nature of the enterprise's private knowledge base through the linkage of the three layers of knowledge.

[0018] A second aspect of this disclosure provides a task execution device based on a hierarchical intelligent collaborative architecture, comprising: The layered construction module is used to build a layered intelligent collaboration architecture that includes an enterprise kernel layer, an industry middleware layer, and a public network big model. The enterprise kernel layer privately deploys the enterprise's exclusive knowledge database, the public network big model connects to public Internet resource data, and the industry middleware layer accumulates and structures an industry solution database. The task decomposition module is used to receive task instructions, identify the logical dependencies of tasks through semantic analysis, decompose the task instructions into executable subtasks, and plan the optimal execution order. The task execution module is used to call the knowledge provided by any one or more layers of the enterprise kernel layer, industry middleware layer and public network big model to execute tasks and record task execution effect data. The iterative optimization module is used to optimize task execution strategies based on reinforcement learning mechanisms, using task execution performance data as the core feedback signal, and to update the databases of the enterprise kernel layer and the industry middleware layer.

[0019] A third aspect of this disclosure provides a computer-readable storage medium storing computer instructions for causing a computer to perform the method provided in the first aspect.

[0020] A fourth aspect of this disclosure provides a computing device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the at least one processor to perform the method provided in the first aspect.

[0021] As can be seen from the above scheme, the task execution method and device based on a hierarchical intelligent collaborative architecture provided in this disclosure constructs a three-layer collaborative knowledge system by integrating enterprise knowledge graphs, industry knowledge graphs, and public network resources. It schedules knowledge at the corresponding level according to task requirements, achieving on-demand access and cross-domain fusion of the three layers of knowledge resources. During task execution, the performance is evaluated in real time. On the one hand, the evaluation results are fed back to the intelligent agent, dynamically optimizing the execution logic and knowledge allocation rules of subsequent tasks through reinforcement learning strategies. On the other hand, effective cases and problem solutions with good execution results are structured and used to update the enterprise knowledge graph and structured solution library, achieving automatic knowledge accumulation and iteration. Attached Figure Description

[0022] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application. In the drawings: Figure 1 A structural block diagram of the layered intelligent collaboration architecture provided in this application; Figure 2 A flowchart illustrating the task execution method based on a hierarchical intelligent collaboration architecture provided in this application; Figure 3 The structural block diagram of the task execution device based on the hierarchical intelligent collaborative architecture provided in this application; Figure 4 This is a structural block diagram of a computing device provided in an embodiment of the present disclosure. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] AI digital employees are virtual assistants developed based on artificial intelligence technology. They can simulate some or all of the functions of human employees and perform various tasks through technologies such as natural language processing, machine learning, and data analysis, thereby reducing the workload of human employees.

[0026] This disclosure provides a full-chain empowerment system for AI digital employees for enterprises. Through the architecture design of execution center + knowledge base, the AI ​​digital employee has both enterprise-specific business capabilities and industry-wide rules, and can also obtain cutting-edge industry information in real time. Ultimately, it can automate and standardize the execution of complex tasks, and replace or assist humans in completing repetitive, standardized and highly reusable work.

[0027] This disclosed embodiment targets scenarios such as cultural tourism, corporate training, and intelligent operation and maintenance. It addresses the core issues of poor business adaptability of general large-scale models, lack of enterprise data security, and easy obsolescence of knowledge systems. Through the design of a MOA intelligent agent execution center combined with a three-layer knowledge base collaborative architecture, it creates AI digital employees that can truly match enterprise business, ensure data security, and continuously evolve themselves, covering various needs such as improving internal operational efficiency, upgrading external services, and accumulating business knowledge.

[0028] To improve the task execution efficiency and quality of AI digital employees, this disclosure adopts an MOA (Multi-Level Architecture) agent as the execution hub, responsible for receiving tasks, breaking them down, allocating them, and optimizing strategies. A three-layer collaborative knowledge architecture, from the inside out, consists of an enterprise kernel layer, an industry middleware layer, and a public network big model. These three layers of knowledge are mutually complementary, providing comprehensive knowledge support for the MOA hub. Each layer undertakes a specific function and achieves knowledge complementarity and capability enhancement through collaborative work. While ensuring the security of core enterprise data, it deeply integrates the enterprise's proprietary experience, industry-wide wisdom, and globally recognized knowledge, comprehensively improving enterprise operational efficiency and service quality.

[0029] Reference Figure 1 As shown in this embodiment, the layered intelligent collaboration architecture includes an enterprise kernel layer, an industry middleware layer, and an outer large model. The enterprise kernel layer privately deploys an enterprise-specific knowledge base to integrate internal business data, process specifications, and historical experience data. It structures and learns fragmented knowledge, such as by deeply linking "a customer frequently complains about air conditioner noise" and "the failure rate of model A equipment increases in summer," to build an enterprise knowledge graph. It also cleans, standardizes, and de-identifies internal enterprise data and updates the enterprise-specific knowledge database based on new business knowledge feedback from task execution. This provides the intelligent agent with the most authoritative and relevant enterprise context, ensuring the accuracy and compliance of the responses.

[0030] The industry middleware layer accumulates and structures a database of industry solutions, such as standard operating procedures for troubleshooting hotel equipment malfunctions, customer complaint handling plans, and service script templates for the cultural tourism industry. It also constructs an industry knowledge graph, linking entities such as equipment, malfunctions, service processes, and customer emotions to enable intelligent retrieval and causal reasoning. Furthermore, the industry middleware layer regularly synchronizes the latest industry norms and standards obtained from the public internet, updates the industry solution database, and anonymizes and receives unique business experience accumulated by the enterprise's core layer.

[0031] The outer-layer large model provides a foundation for general language understanding, logical reasoning, and content generation capabilities, such as interpreting network engineering documents and providing hotel operation consulting. It connects to public internet resources to acquire and integrate massive amounts of internet information, academic papers, technical standards, and industry news, addressing knowledge gaps and timeliness issues in the enterprise kernel layer and industry middleware layer. The acquired effective external information is then fed back to the enterprise kernel layer to update the enterprise's proprietary knowledge database and back to the industry middleware layer to update the industry solution database.

[0032] The core idea of ​​the above three-layer collaborative knowledge system is "layered governance, on-demand access, and secure collaboration". The three-layer knowledge database does not exist in isolation, but provides dynamic, accurate and reliable knowledge supply to Moa (hybrid agent) through a unified knowledge index and intelligent routing mechanism.

[0033] Figure 2 A flowchart illustrating the task execution method based on a hierarchical intelligent collaboration architecture provided in this application. (Refer to...) Figure 2 As shown, in step S202, a layered intelligent collaboration architecture is constructed, which includes an enterprise kernel layer, an industry middleware layer, and a public network big model. The enterprise kernel layer privately deploys an enterprise-specific knowledge database, the public network big model connects to public resource data on the Internet, and the industry middleware layer accumulates and structures an industry solution database.

[0034] First, collect internal business data, process specifications, and historical experience data to establish a proprietary enterprise knowledge database containing important information such as customer preference data and service optimization plans. Then, clean and standardize the collected data, extract entities and relationships, and construct an enterprise knowledge graph.

[0035] Specifically, this includes: collecting structured, semi-structured, and experiential knowledge from within the enterprise, including basic customer information, internal business process documents, historical maintenance work orders, internal training materials, and enterprise-specific operating procedures; unifying non-standardized descriptions into business tags, encrypting and implementing hierarchical access control for sensitive customer information, classifying the cleaned knowledge according to business scenarios, adding structured tags, and storing it in the corresponding knowledge database; defining the rules of the knowledge system based on the enterprise's actual business, extracting all entities from the knowledge database, merging different identifiers of the same entity, mining implicit relationships from historical data, completing the knowledge network, and obtaining the enterprise knowledge graph.

[0036] In practical implementation, it can directly connect to the databases of enterprise ERP, CRM, work order systems, equipment management platforms, and HR systems to extract structured data such as customer information, equipment ledgers, historical fault records, service process standard operating procedures, and employee manuals. It can also parse semi-structured data such as internal business documents, emails, meeting minutes, and training materials, using document parsing tools to extract text, tables, and metadata. Furthermore, it can collect anonymized and authorized unstructured data such as customer service call recordings (converted to text), internal pages, instant messaging group chat records, and expert experience interview recordings / notes. Each piece of data is labeled with its source system, creation time, responsible person, data sensitivity level (public, internal, confidential), and business domain (e.g., room service, network operations).

[0037] Merge multiple records of the same customer / device across different systems, standardizing date format, units, and status enumeration (e.g., "Fault - Under Repair - Resolved"). Process customer information using masking, tokenization, or differential privacy techniques to ensure that core privacy is not leaked during subsequent knowledge graph construction and model training. Identify missing and outlier values ​​and repair or label them according to rules or simple models. Remove irrelevant symbols and stop words, and perform word segmentation and part-of-speech tagging.

[0038] Identify key entities in the text. For example, equipment entities: brand-model, asset number, installation location ("Air conditioner in hotel room 301"). Personnel entities: employee ID, job title, skill certification. Business entities: customer complaint type ("Network lag"), service process ID, SLA level. Identify semantic relationships between entities to form triples: (Equipment-A, belongs to, Equipment Category-B) (Fault symptom-C, commonly seen in device model-D) (Employee-E, Skilled at problem-solving, Problem type-F) (Process-G, the first step is, operation-H) It associates and merges attributes of the same entity from different systems, uses a native graph database to store the graph, and supports efficient complex association queries, such as: "find all Huawei devices located on the third floor that have experienced more than two network failures in the past six months, and associate them with the engineers in charge."

[0039] The system will structure industry solution data, including equipment failure solutions, customer service complaint handling procedures, and emergency response plans, to construct an industry knowledge graph. It can also accumulate general industry knowledge assets based on the characteristics of the cultural and tourism industry, including a database of solutions to common hotel customer fault reports, standard customer service complaint handling procedures, a network engineering course resource library, industry standards for intelligent equipment operation and maintenance, and the latest strategies for the cultural and tourism industry. Key entities and their relationships will be extracted from this general knowledge, and entity-relationship pairs will be constructed, thus structuring this general knowledge into an industry knowledge graph. Specific problems in the enterprise knowledge graph can be linked to generalized problems in the industry knowledge graph. In this way, when faced with a new failure, the system can first search for similar historical examples in the enterprise knowledge graph and then find a general solution by linking to the industry knowledge graph.

[0040] We selected a public network large-scale model and used internal enterprise data to fine-tune the pre-trained model. Based on this model, we acquired public resources including authoritative external data sources, industry standard databases, and academic paper repositories. The public network large-scale model used was a mature open-source model (such as GPT-4 or similar technologies) as its foundation, fully leveraging its capabilities in text understanding, knowledge reasoning, and semantic generation. To ensure the model aligned with the enterprise's specific context and business logic, we conducted secondary fine-tuning for specific business scenarios. All fine-tuning data came from the enterprise's internal business materials, such as customer consultation scripts for hotel operations, operation manuals for smart device maintenance, teaching materials for network engineering courses, and historical customer fault reports. After fine-tuning, the large-scale model perfectly matched the enterprise's specific context and business logic. For example, the output scripts in the hotel scenario were more in line with service standards, and the technical solutions in the maintenance scenario were more in line with the enterprise's internal operating standards.

[0041] In step S204, a task instruction is received, the logical dependencies of the task are identified through semantic analysis, the task instruction is decomposed into executable subtasks, and the optimal execution order is planned.

[0042] The MOA hybrid intelligent agent, acting as the execution hub of AI digital employees, is the core link connecting task input and knowledge scheduling. Upon receiving instructions from frontline business operations, it breaks down complex tasks into a sequence of executable and schedulable sub-tasks. For example, upon receiving the task instruction to "develop a summer room upgrade service plan for VIP customers," it decomposes the following sub-tasks: ① retrieve VIP customer historical preferences; ② query industry high-end service trends and competitor cases; ③ match available room types and equipment status; ④ generate a personalized draft plan; ⑤ review it using hotel service standards.

[0043] In step S206, the knowledge provided by any one or more layers of the enterprise kernel layer, industry middleware layer, and public network big model is invoked to execute the task, and the task execution effect data is recorded.

[0044] In one embodiment of this disclosure, firstly, the enterprise kernel layer is invoked to retrieve the enterprise knowledge graph and match the corresponding solution. If no matching result is found, the industry middleware layer is invoked to retrieve the industry knowledge graph and match the corresponding solution. If no matching result is found, the public network big data model is invoked to obtain task-related information from public internet resources, perform semantic analysis and filtering on the task-related information, generate a solution, and generate customer communication scripts or operation and maintenance instructions based on the retrieved or generated solution.

[0045] The three-tiered knowledge system coordinates according to the principle of prioritizing internal enterprise knowledge, followed by industry knowledge, and finally external knowledge. This ensures both data security and execution efficiency: all tasks first check the enterprise's core layer's dedicated knowledge database. If no match is found, the system then checks the industry's middle layer's general solutions. If still no match is found, the system then calls external information from the public network's large-scale model, forming a complete knowledge chain with internal knowledge as the priority, industry knowledge as a safety net, and external knowledge to fill in the gaps.

[0046] In one embodiment of this disclosure, past task data can be integrated to form a historical database, which includes successful and failed cases. By comparing task performance under different scenarios, key factors affecting task success are identified, and the success rate, average processing time, and resource consumption of various tasks are calculated. Based on the analysis results, a series of best practices are extracted as standard operating procedures. These identified best practices are integrated into a knowledge base, and when facing new tasks, decisions are made based on historical data and the current execution environment. Through data-driven decision support, the intelligent agent can not only effectively improve the quality and efficiency of task execution but also accumulate rich knowledge resources through continuous learning and optimization.

[0047] For example, in hotel management applications, historical customer preferences can be obtained during a guest's stay, and the latest industry service standards can be referenced to provide a personalized service experience. In network maintenance applications, when network failures occur, past solutions can be quickly extracted from the enterprise knowledge graph and combined with new technology standards obtained from public resources to guide maintenance personnel in conducting rapid and effective troubleshooting and repair.

[0048] In step S208, the task execution effect data is used as the core feedback signal to optimize the task execution strategy based on the reinforcement learning mechanism, and the databases of the enterprise kernel layer and the industry middleware layer are updated.

[0049] It can record the complete execution chain of each task, such as which knowledge was invoked, what results were generated, and the final user satisfaction / task completion rate. It collects feedback information including customer satisfaction, problem resolution rate, task execution time, and whether compliance issues arose, using this feedback as a reward signal to adjust execution logic. Successful resolution of complex problems and positive user feedback receive positive rewards, continuously optimizing task planning strategies and knowledge invocation preferences.

[0050] Solutions generated from public network big data models are reviewed and then synchronized to the enterprise kernel layer and industry middleware layer. General solutions adapted to enterprises are reviewed by experts and then synchronized to the enterprise kernel layer. New solutions accumulated in the enterprise kernel layer are anonymized and then synchronized to the industry middleware layer. As the task execution volume increases, the MOA execution strategy continues to iterate, accelerating the growth curve of AI digital employees.

[0051] Typical application scenarios are shown below: Hotel Customer Complaint Handling Scenario 1 The front desk received a complaint at night that "the air conditioning in room 1108 is not working well and is very noisy".

[0052] The AI ​​digital employee task execution workflow is as follows: Receiving and Dismantling: The MOA receiving task is broken down into: ① confirming equipment information, ② providing temporary reassurance solutions, ③ arranging maintenance, and ④ recording and archiving.

[0053] Enterprise kernel layer call: Query room 1108 file (air conditioner model, installation date, historical maintenance records), query customer's historical preferences (whether they are sensitive to temperature).

[0054] Industry middleware layer access: Search the "Hotel Air Conditioner Fault Knowledge Graph", match the common causes of "poor cooling + loud noise" (possible causes: clogged filter, insufficient refrigerant, worn fan bearings), and retrieve the corresponding standard repair procedures and customer reassurance messages.

[0055] Public network big data model call: Generate a warm and comforting text message; check if there are any recent batch failure announcements for the same model of air conditioner (from brand website / industry forum).

[0056] Integration and Execution: MOA integrates information and automatically generates instructions: ① Sends maintenance work orders with standard operating procedures (including possible causes) to engineers; ② Sends reassurance messages and upgrade compensation suggestions to the front desk (based on customer preferences and company strategy); ③ Automatically updates customer files and equipment health status.

[0057] Reinforcement learning: If the customer gives a "very satisfied" rating after the repair, MOA will increase the weight of that fault handling process.

[0058] Hotel customer complaint handling scenario two: Task Received: MOA receives the task "Handle guest complaint that there is no hot water in the guest room water heater, and the guest requests an upgrade to a different room and additional benefits"; Task breakdown: MOA automatically breaks down the task into 4 sub-tasks: ① retrieve guest's historical check-in records and preferences, ② match standardized customer complaint handling solutions, ③ query the hotel's available upgrade room types, and ④ generate complete handling scripts and operation procedures. Three-layer collaborative knowledge retrieval: From the enterprise kernel layer, we can retrieve the following information: This is the guest's first stay, with no special preferences. The guest had previously reported a problem with the aging water heater in the room during their last stay. The knowledge graph is linked to the equipment failure history of the room. According to information obtained from industry insiders, the standardized solution for handling customer complaints caused by equipment failure is to "prioritize room upgrades, provide corresponding benefits, and arrange repairs simultaneously." The information retrieved from the enterprise's core layer indicates that there are currently vacancies in the hotel's deluxe king rooms, and the corresponding upgrade benefits are complimentary afternoon tea for two + late check-out until 2 PM. Task Execution: Generate a complete solution: ① Call the guest to apologize, explain the cause of the malfunction, upgrade to a deluxe king room free of charge, offer a complimentary afternoon tea for two, and extend check-out until 2 PM; ② Simultaneously dispatch the order to the operations and maintenance staff to repair the water heater in the room; ③ Follow up on the guest's stay experience. Strategy optimization: After the task is completed, the customer satisfaction is 5 stars, the repair is completed in one go, and a positive reward is obtained. The solution is stored in the enterprise's private brain and industry brain, and the solution can be directly called in the future for similar customer complaints, further improving the execution efficiency.

[0059] Figure 3 This is a structural block diagram of the task execution device based on a hierarchical intelligent collaborative architecture provided in this application. (Refer to...) Figure 3 As shown, the device 300 includes a hierarchical construction module 310, a task decomposition module 320, a task execution module 330, and an iterative optimization module 340.

[0060] Among them, the layered construction module 310 is used to build a layered intelligent collaboration architecture that includes an enterprise kernel layer, an industry middleware layer, and a public network big model. The enterprise kernel layer privately deploys an enterprise-specific knowledge database, the public network big model connects to public Internet resource data, and the industry middleware layer accumulates and structures an industry solution database.

[0061] A layered intelligent collaboration architecture enables a highly efficient, secure, and scalable AI digital employee system. Each layer undertakes a specific function and achieves knowledge complementarity and capability enhancement through collaborative work. Specifically, the enterprise kernel layer privately deploys core enterprise data to ensure the security of sensitive data and integrates internal enterprise knowledge, including business data, process specifications, and historical experience, to build a dedicated enterprise knowledge base. Utilizing knowledge graph technology, fragmented knowledge is structured, stored, and linked for learning, providing the AI ​​digital employee with accurate knowledge retrieval and deep reasoning capabilities.

[0062] The industry middleware layer focuses on industry characteristics, developing a library of solutions for common problems in specific scenarios (such as hotel customer malfunction reports). It accumulates best practice data on industry practices, such as hotel equipment troubleshooting and customer service complaint handling, improving the efficiency and service quality of frontline customer service and operations personnel. Collaborating with the enterprise core layer, it combines internal experience with industry-standard solutions to provide more precise business support. The outer large-scale model, based on public internet data, connects to authoritative external data sources, industry knowledge bases, academic paper databases, and other public resources to expand the knowledge breadth of the AI ​​digital employee, enabling it to access the latest technical standards, industry trends, and cutting-edge application cases, keeping its knowledge fresh and innovative. Linking with the enterprise core layer and the industry middleware layer, it compensates for the limitations of knowledge in specific domains, enhancing the overall system's intelligence.

[0063] The task decomposition module 320 is used to receive task instructions, identify the logical dependencies of tasks through semantic analysis, decompose the task instructions into executable subtasks, and plan the optimal execution order.

[0064] The task execution module 330 is used to call the knowledge provided by any one or more layers of the enterprise kernel layer, industry middleware layer and public network big model to execute tasks and record task execution effect data.

[0065] The iterative optimization module 340 is used to optimize the task execution strategy based on the task execution effect data as the core feedback signal, and update the databases of the enterprise kernel layer and the industry middleware layer.

[0066] The description of a task execution device based on a hierarchical intelligent collaborative architecture can be found in the description of a task execution method based on a hierarchical intelligent collaborative architecture, and will not be repeated here.

[0067] In summary, the task execution method and apparatus based on a hierarchical intelligent collaborative architecture disclosed herein construct a three-layer collaborative knowledge system by building an enterprise knowledge graph, an industry knowledge graph, and public network resources. It schedules knowledge at the corresponding layer according to task requirements, enabling on-demand access and cross-domain fusion of the three layers of knowledge resources. During task execution, the method evaluates the execution effect in real time. On the one hand, the evaluation results are fed back to the intelligent agent, dynamically optimizing the execution logic and knowledge allocation rules of subsequent tasks through reinforcement learning strategies. On the other hand, effective cases and problem solutions with good execution results are structured and used to update the enterprise knowledge graph and structured solution library, achieving automatic knowledge accumulation and iteration.

[0068] To effectively cultivate the business processing capabilities of AI digital employees and ensure their rapid adaptation to enterprise needs, AI digital employees can be trained. Through standardized tests or scenario simulations, the initial capabilities of AI digital employees in text understanding, problem-solving, and business application can be assessed. Based on the specific business needs of the enterprise, it can be analyzed which areas require focused training, training objectives can be formulated, and corresponding course modules and learning content can be matched.

[0069] Each AI digital employee is assigned a specific simulation role, and various simulation scenarios are built based on the company's real business processes and environment. For example, the network engineering scenario simulates the setup and troubleshooting of local area networks, helping AI digital employees understand the practical operations of network configuration, monitoring, and troubleshooting. The smart device scenario creates an environment for the installation and debugging of smart devices, enabling AI digital employees to master the functions of the devices and solutions to common problems. The hotel operations scenario simulates the processes of room management and customer service, allowing AI digital employees to learn how to handle customer needs and optimize service processes.

[0070] After each simulation scenario, performance data of the AI ​​digital employee in task execution is collected. In the later stages of training, the AI ​​digital employee is paired with real employees to form cross-functional teams to complete business tasks. Team members and mentors evaluate and guide the AI ​​digital employee's performance in the collaborative process; regular feedback meetings are set up to adjust the AI ​​digital employee's behavior patterns and decision-making strategies.

[0071] Through online tests, practical assessments, and case studies, we regularly test the knowledge and skills of AI digital employees to evaluate their progress and optimize training content and implementation methods based on feedback.

[0072] This disclosure also provides a computing device, such as... Figure 4 As shown, the computing device includes one or more processors 401 and a memory 402. Figure 4 Take a processor 401 as an example.

[0073] The computing device may also include an input device 403 and an output device 404.

[0074] The processor 401, memory 402, input device 403, and output device 404 can be connected via a bus or other means. Figure 4 Taking the example of a connection between China and Israel via a bus.

[0075] Processor 401 can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips. The general-purpose processor can be a microprocessor or any conventional processor.

[0076] The memory 402, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 401 executes various functional applications and data processing of the server by running the non-transitory software programs, instructions, and modules stored in the memory 402, thereby implementing the methods in the above-described method embodiments.

[0077] Memory 402 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the use of the processing device operated by the server. Furthermore, memory 402 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory 402 may optionally include memory remotely located relative to processor 401, and these remote memories can be connected to network connectivity devices via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0078] One or more modules are stored in memory 402 and, when executed by one or more processors 401, perform the methods shown in the above embodiments.

[0079] Those skilled in the art will understand that all or part of the processes in the above method embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory (FM), hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0080] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0081] Obviously, those skilled in the art should understand that the various units or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps into a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0082] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A task execution method based on a hierarchical intelligent collaborative architecture, characterized in that, include: Construct a layered intelligent collaboration architecture that includes an enterprise kernel layer, an industry middleware layer, and a public network big model. The enterprise kernel layer privately deploys an enterprise-specific knowledge database, the public network big model connects to public Internet resource data, and the industry middleware layer accumulates and structures an industry solution database. Receive task instructions, identify the logical dependencies of tasks through semantic analysis, decompose the task instructions into executable subtasks, and plan the optimal execution order; The task execution is performed by invoking the knowledge provided by any one or more layers of the enterprise kernel layer, the industry middleware layer, and the public network big model, and the task execution effect data is recorded. Using the task execution performance data as the core feedback signal, the task execution strategy is optimized based on the reinforcement learning mechanism, and the databases of the enterprise kernel layer and the industry middleware layer are updated.

2. The task execution method based on a hierarchical intelligent collaborative architecture according to claim 1, characterized in that, The construction of the layered intelligent collaboration architecture, comprising an enterprise kernel layer, an industry middleware layer, and a public network big data model, includes: Collect internal business data, process specifications, and historical experience data to establish a proprietary enterprise knowledge database that includes customer preference data and service optimization solutions. After data cleaning, standardization, and anonymization of the collected data, extract entities and relationships to construct an enterprise knowledge graph. The industry solution data, including information on equipment failure solutions, customer service complaint solutions, and emergency response solutions, will be structured and processed to build an industry knowledge graph. We select a public network large model, use internal enterprise data to fine-tune the pre-trained large model, and acquire public resource data, including external authoritative data sources, industry standard databases, and academic paper databases, based on the public network large model.

3. The task execution method based on a hierarchical intelligent collaborative architecture according to claim 2, characterized in that, The process involves collecting internal business data, process specifications, and historical experience data to establish a proprietary enterprise knowledge database that includes customer preferences and service optimization solutions. After data cleaning, standardization, and anonymization, entities and relationships are extracted from the collected data to construct an enterprise knowledge graph, including: Collect structured, semi-structured, and experiential knowledge from within the enterprise, including basic customer information, internal business process documents, historical maintenance work orders, internal training materials, and enterprise-specific operating procedures. Based on the collected knowledge data, non-standardized descriptions are unified into business tags, and sensitive customer information is encrypted and subject to hierarchical access control. The cleaned knowledge is classified according to business scenarios, labeled with structured tags, and stored in the corresponding knowledge database. Based on the rules of the knowledge system defined by the actual business of the enterprise, all entities are extracted from the knowledge database, different identifiers of the same entity are merged, implicit relationships are mined from historical data, the knowledge network is completed, and the enterprise knowledge graph is obtained.

4. The task execution method based on a hierarchical intelligent collaborative architecture according to claim 2, characterized in that, The process involves structuring industry solution data, including information on equipment failure solutions, customer service complaint solutions, and emergency response plans, to construct an industry knowledge graph, including: Based on the characteristics of the cultural and tourism industry, general knowledge for the cultural and tourism industry has been compiled, including a library of solutions to common problems reported by hotel customers, standard procedures for handling customer service complaints, a teaching resource library for network engineering courses, industry standards for the operation and maintenance of intelligent devices, and the latest strategies for the cultural and tourism industry. Extract key entities and their relationships from the general knowledge of the cultural and tourism industry, construct entity-relationship pairs, and structure the general knowledge of the cultural and tourism industry into an industry knowledge graph.

5. The task execution method based on a hierarchical intelligent collaborative architecture according to claim 1, characterized in that, The process of invoking knowledge provided by any one or more layers of the architecture, including the enterprise kernel layer, the industry middleware layer, and the public network big data model, and recording task execution effect data, includes: The enterprise kernel layer is invoked to retrieve the enterprise knowledge graph and match the corresponding solution. If no matching result is found, the industry middleware layer is invoked to retrieve the industry knowledge graph and match the corresponding solution. If no matching result is found, the public network big model is invoked to obtain task-related information from public internet resources, perform semantic analysis and filtering on the task-related information, and generate a solution. Generate customer communication scripts or operation and maintenance guidelines based on the retrieved or generated solutions.

6. The task execution method based on a hierarchical intelligent collaborative architecture according to claim 1, characterized in that, The process of using the task execution performance data as the core feedback signal, optimizing the task execution strategy based on a reinforcement learning mechanism, and updating the databases of the enterprise kernel layer and the industry middleware layer includes: Record the complete chain of each task execution, collect feedback information including customer satisfaction, problem resolution rate, task execution time, and whether compliance issues have occurred, and use the feedback information as a reward signal to adjust the execution logic; After the solution generated from the public network big model is reviewed, it is synchronized to the enterprise kernel layer and the industry middleware layer. The general solution adapted to the enterprise is synchronized to the enterprise kernel layer after being reviewed by experts. The new solution accumulated in the enterprise kernel layer is synchronized to the industry middleware layer after being de-identified.

7. The task execution method based on a hierarchical intelligent collaborative architecture according to claim 1, characterized in that, The enterprise kernel layer integrates internal business data, process specifications, and historical experience data to construct an enterprise knowledge graph. It also performs data cleaning, standardization, and anonymization on internal data and updates the enterprise-specific knowledge database based on new business knowledge feedback from task execution. The industry middleware layer accumulates standardized processing solutions for high-frequency industry scenarios, constructs an industry knowledge graph, regularly synchronizes the latest industry specifications and standards obtained from the public internet, updates the industry solution database, and anonymizes the unique business experience accumulated by the enterprise kernel layer. The public internet big data model utilizes public resources including external authoritative data sources, industry standard databases, and academic paper repositories. It transmits the acquired effective external information back to the enterprise kernel layer to update the enterprise-specific knowledge database and back to the industry middleware layer to update the industry solution database.

8. A task execution device based on a hierarchical intelligent collaborative architecture, characterized in that, include: The layered construction module is used to build a layered intelligent collaboration architecture that includes an enterprise kernel layer, an industry middleware layer, and a public network big model. The enterprise kernel layer privately deploys an enterprise-specific knowledge database, the public network big model connects to public Internet resource data, and the industry middleware layer accumulates and structures an industry solution database. The task decomposition module is used to receive task instructions, identify the logical dependencies of tasks through semantic analysis, decompose the task instructions into executable subtasks, and plan the optimal execution order. The task execution module is used to call the knowledge provided by any one or more layers of the enterprise kernel layer, the industry middleware layer, and the public network big model to execute tasks and record task execution effect data. The iterative optimization module is used to optimize the task execution strategy based on the task execution effect data as the core feedback signal, and to update the databases of the enterprise kernel layer and the industry middleware layer.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the task execution method based on the hierarchical intelligent collaborative architecture as described in any one of claims 1 to 7.

10. A computing device, characterized in that, The computing device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the at least one processor to perform the task execution method based on the hierarchical intelligent collaborative architecture as described in any one of claims 1 to 7.

Images