Work order processing method and device, electronic equipment, medium and program product

Abstract

The invention provides a work order processing method and device, electronic equipment, a medium and a program product, relates to application of a large model in the field of financial science and technology, and can be applied to the technical field of big data and the technical field of artificial intelligence. The method comprises the following steps: acquiring target work order information; based on the target work order information and a pre-constructed multi-source knowledge base, obtaining a target assignment post corresponding to the target work order information by using a work order agent; executing work order analysis by using the work order intelligent agent, and generating task processing decision information; and sending the task processing decision information to the target assignment post, and obtaining a work order processing result returned by the target assignment post.

Description

Technical Field

[0001] This application relates to the application of large models in the field of financial technology, which can be applied to the fields of big data technology and artificial intelligence technology, and more specifically to a work order processing method, device, equipment, medium and program product. Background Technology

[0002] Currently, the banking industry's technology systems have long construction cycles, complex business scenarios, and diverse architectures. Due to historical legacy issues and rapid business iterations, they generally suffer from fragmented technology stacks, outdated system architectures, and high code coupling, making it easy for new features to trigger a chain of failures. Banking system problems are primarily discovered through two methods: proactive monitoring by information monitoring systems and passive feedback from business channels. The former deploys monitoring modules at the infrastructure, application, and business layers to collect operational metrics and automatically generate work orders when, for example, server central processing unit (CPU) usage is consistently too high or transaction failure rates are abnormal. The latter relies on customers or internal personnel reporting anomalies, which are then verified to generate business-related work orders.

[0003] The current work order processing generally adopts a manual-led "three-stage" process, including reporting, assignment, and processing. Although some banks have introduced automatic assignment based on keyword matching or automatic processing based on rule matching to improve efficiency, there are still problems such as insufficient matching accuracy and difficulty in automatically identifying and handling complex faults. The overall level of intelligence and automation is limited. Summary of the Invention

[0004] In view of the above problems, this application provides work order processing methods, apparatus, equipment, media and program products.

[0005] According to a first aspect of this application, a work order processing method is provided, the method comprising: acquiring target work order information; obtaining a target assignment position corresponding to the target work order information using a work order intelligent agent based on the target work order information and a pre-constructed multi-source knowledge base, wherein the work order intelligent agent includes a work order processing model, the work order processing model being obtained by adjusting the target parameters of a pre-trained language model under the constraint of a task loss function using a parameter fine-tuning algorithm, the task loss function being constructed based on a semantic association graph in the multi-source knowledge base; performing work order analysis using the work order intelligent agent to generate task processing decision information; and sending the task processing decision information to the target assignment position to obtain the work order processing result returned by the target assignment position.

[0006] According to an embodiment of this application, a multi-source knowledge base is pre-constructed, including: acquiring multi-source knowledge data; performing semantic parsing and relation extraction on the multi-source knowledge data to identify semantic elements and corresponding relationships, so as to construct the semantic association graph; and fusing the multi-source knowledge data based on the semantic association graph to form the multi-source knowledge base.

[0007] According to an embodiment of this application, performing semantic parsing and relation extraction on the multi-source knowledge data includes: performing semantic parsing and relation extraction on knowledge data from different sources using corresponding processing algorithms, wherein the multi-source knowledge data includes at least historical work order data, unstructured document data, expert experience data, and external knowledge data.

[0008] According to embodiments of this application, the semantic parsing and relation extraction using corresponding processing algorithms for knowledge data from different sources includes: for the historical work order data, using sequence labeling algorithms to obtain work order semantic elements; for the unstructured document data, using document semantic segmentation and entity relation acquisition algorithms to extract functional semantic elements, which are used to characterize the semantic information of system structure, function, and operating scenario; for the expert experience data, using semantic clustering analysis to identify similar problem patterns and obtain supplementary knowledge information; and for the external knowledge data, using text summarization and semantic alignment algorithms to obtain structured knowledge fragments.

[0009] According to an embodiment of this application, obtaining the target assignment position corresponding to the target work order information using a work order agent includes: inputting the target work order information into the work order agent; guiding the work order agent to call the work order processing model and access the multi-source knowledge base based on an assignment task template pre-constructed through prompt word engineering to perform semantic matching on the target work order information; and obtaining the target assignment position and the corresponding assignment reason based on the semantic matching result.

[0010] According to an embodiment of this application, adjusting the target parameters of the pre-trained language model under the constraint of the task loss function using a parameter fine-tuning algorithm includes: updating the target trainable parameters while keeping the backbone parameters of the pre-trained language model frozen; and dynamically adjusting the parameter learning rate based on the gradient change magnitude determined by the semantic node weights in the task loss function during each training round.

[0011] According to an embodiment of this application, constructing the task loss function based on the semantic association graph in the multi-source knowledge base includes: extracting a target association path from the semantic association graph, wherein the target association path includes at least problem information, cause information, handling information, and job information; establishing a constraint relationship based on the node similarity and edge weight in the target association path; and associating the constraint relationship with the model output result to construct the task loss function.

[0012] According to an embodiment of this application, the step of using the work order agent to perform work order analysis and generate task processing decision information includes: retrieving similar cases in the multi-source knowledge base based on the target work order information to determine candidate problem categories; and performing semantic matching on the candidate problem categories to generate task processing decision information containing at least one handling strategy or verification scheme.

[0013] According to an embodiment of this application, the method further includes: comparing the work order processing result and the task processing decision information; in response to the existence of a difference, marking the difference content and the corresponding work order information as a knowledge gap; generating a question-and-answer task for the knowledge gap; initiating a question-and-answer task request to a domain expert based on the question-and-answer task to obtain supplementary missing knowledge information; and converting the supplementary missing knowledge information into structured knowledge entries and updating them to the multi-source knowledge base.

[0014] A second aspect of this application provides a work order processing apparatus, comprising: a data acquisition module for acquiring target work order information; a work order dispatch module for obtaining a target dispatch position corresponding to the target work order information based on the target work order information and a pre-built multi-source knowledge base using a work order intelligent agent, wherein the work order intelligent agent includes a work order processing model, which is obtained by adjusting the target parameters of a pre-trained language model under the constraint of a task loss function using a parameter fine-tuning algorithm, and the task loss function is constructed based on a semantic association graph in the multi-source knowledge base; a work order analysis module for performing work order analysis using the work order intelligent agent to generate task processing decision information; and a work order processing module for sending the task processing decision information to the target dispatch position and obtaining the work order processing result returned by the target dispatch position.

[0015] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0016] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0017] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0018] According to embodiments of this application, the pre-trained language model is fine-tuned by introducing a task loss function constrained by a semantic association graph. This enables the model to learn inherent relationships in the semantic space, thereby improving its understanding and decision-making capabilities regarding complex work order semantics. Furthermore, the pre-trained language model is constrained by a task loss function constructed using the semantic association graph, focusing the model's learning on the work order semantic space, reducing redundant computation, and improving inference convergence speed. In addition, by combining knowledge-driven semantic reasoning with a multi-role task template mechanism, automatic identification of work order issues, intelligent assignment of responsible positions, and proactive recommendation of handling solutions are achieved. This reduces manual intervention and cross-departmental communication time, allowing users to complete issue confirmation and processing feedback on a single interface, thus improving user satisfaction with the system's intelligent decision-making. Attached Figure Description

[0019] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustrations depict application scenarios of the work order processing method, apparatus, equipment, medium, and program products according to embodiments of this application.

[0021] Figure 2 A flowchart illustrating a work order processing method according to an embodiment of this application is shown schematically.

[0022] Figure 3 The flowchart illustrates a method for obtaining target assignment positions corresponding to target work order information according to some exemplary embodiments of this application.

[0023] Figure 4 The illustration shows a schematic diagram of a method for adjusting target parameters of a pre-trained language model according to some exemplary embodiments of the present application;

[0024] Figure 5 A schematic diagram illustrating the structure of a work order processing apparatus according to an embodiment of this application is shown; and

[0025] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a work order processing method according to an embodiment of this application. Detailed Implementation

[0026] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0030] Currently, banking technology systems are generally characterized by long construction cycles, complex and ever-changing business scenarios, and diverse system architecture layers. Influenced by historical construction legacies and rapid business iterations, many banks' technology systems suffer from fragmented technology stacks and outdated system architectures. For example, some core business systems still rely on traditional monolithic architectures, lack compatibility with emerging distributed technologies, have high code coupling, complex logic, high maintenance costs, and are prone to triggering cascading failures when new features are added.

[0031] Currently, problem detection in banking technology systems mainly relies on two paths:

[0032] First, there's the proactive monitoring by the information monitoring system. This system deploys monitoring modules at the infrastructure layer (servers, storage, network devices), application layer (core business logic, middleware, databases), and business layer (transaction success rate, response time, order processing volume, etc.), achieving real-time monitoring by collecting operational metrics. When a metric exceeds a preset threshold (e.g., server CPU utilization exceeding 90% for 10 consecutive minutes, transaction failure rate exceeding 0.5%), the system automatically triggers an alarm and generates a monitoring alarm work order. The work order typically includes basic information such as alarm time, device or application identifier, abnormal metric value, and alarm level.

[0033] Second, there is passive feedback through business channels. This path primarily relies on customer or internal business personnel reporting issues. Customers report usage anomalies (such as failed transfer page loading or no results found for wealth management product inquiries) through the bank's application, online banking, or customer service hotline. Internal personnel (such as tellers and account managers) also report system anomalies (such as incorrect form submissions or business process delays) through the internal system when handling transactions. After initial verification by customer service or business management departments, a business-related work order is generated and entered into the system.

[0034] For the two types of work orders mentioned above, banks generally adopt a three-stage, manually-led processing flow, including: Work order reporting: receiving and verifying the completeness of monitoring alarm or business-related work orders, and completing the work order creation; Work order dispatch: manually dispatching the work order to the corresponding processing team (such as infrastructure maintenance, application maintenance, business support) based on information such as the fault type, involved systems, and business areas in the work order, combined with the internal division of responsibilities system; Work order processing: responsible for problem diagnosis, fault location and repair, and recording the processing process and results, and providing feedback on the work order closed loop.

[0035] To improve work order processing efficiency, some banks have implemented partial automation optimizations on top of manual processes. For example, a keyword-based automatic dispatch scheme extracts keywords (such as "server crash," "transfer failure," and "high CPU") from work order titles and problem descriptions using natural language processing technology, and matches them against a pre-defined "keyword-responsible team" rule base to achieve automatic dispatch. Another example is a rule-based automatic handling scheme: for high-frequency, simple faults (such as server crashes and abnormal network traffic), a rule-triggered and script execution mechanism is built to achieve partial automatic repair.

[0036] However, existing technologies still have significant shortcomings: On the one hand, keyword-based dispatching schemes rely heavily on work order text. When the work order description is ambiguous (e.g., "system cannot be used") or contains multiple semantics (e.g., "high CPU caused transfer timeout"), the matching accuracy is insufficient, which can easily lead to mis-dispatch or missed dispatch, requiring manual intervention. On the other hand, rule-based automatic handling schemes can only cover a small number of high-frequency, structured faults. They cannot effectively identify and handle complex scenarios (e.g., code logic errors, cross-system interface call anomalies, transaction data anomalies, etc.), still requiring manual investigation, and the overall level of automation is limited.

[0037] Based on this, embodiments of this application provide a work order processing method, the method comprising: acquiring target work order information; based on the target work order information and a pre-constructed multi-source knowledge base, using a work order intelligent agent to obtain the target assignment position corresponding to the target work order information, wherein the work order intelligent agent includes a work order processing model, the work order processing model being obtained by adjusting the target parameters of a pre-trained language model under the constraint of a task loss function using a parameter fine-tuning algorithm, the task loss function being constructed based on a semantic association graph in the multi-source knowledge base; using the work order intelligent agent to perform work order analysis and generate task processing decision information; and sending the task processing decision information to the target assignment position and obtaining the work order processing result returned by the target assignment position. According to embodiments of this application, a task loss function constrained by a semantic association graph is introduced to fine-tune the parameters of a pre-trained language model, enabling the model to learn inherent relationships in the semantic space, thereby improving the model's understanding and decision-making ability regarding complex work order semantics. Furthermore, the task loss function constructed using the semantic association graph constrains the pre-trained language model, focusing its learning on the work order semantic space, reducing redundant computation and improving inference convergence speed. In addition, by combining knowledge-driven semantic reasoning with a multi-role task template mechanism, automatic identification of work order issues, intelligent assignment of responsible positions, and proactive recommendation of handling solutions are achieved, reducing manual intervention and cross-departmental communication time. Users can complete issue confirmation and processing feedback on a single interface, improving user satisfaction with the system's intelligent decision-making.

[0038] It should be noted that the work order processing method, apparatus, equipment, medium, and program products defined in this application relate to the application of large-scale models in the field of financial technology. They can be applied to the fields of big data technology and artificial intelligence technology, and can also be used in various fields other than big data technology, distributed technology technology, and financial technology. The application fields of the work order processing method, apparatus, equipment, medium, and program products provided in the embodiments of this application are not limited.

[0039] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0040] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all offer users corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process proceeds to expert decision-making. Here, "expert decision-making" refers to the decision-making activities of personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0041] Figure 1 The illustration shows an application scenario diagram of the work order processing method, apparatus, device, medium, and program product according to embodiments of this application.

[0042] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0043] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0044] In the embodiments of this application, the first terminal device 101 can be an example of the first device, and the second terminal device 102 and / or the third terminal device 103 can be an example of at least one second device. The first device and the second device can communicate collaboratively through an internal client mechanism to implement the data distribution and rendering logic described in the work order processing method.

[0045] In some embodiments, the first device and at least one second device may be different display modules, windows or screens on the same computing terminal (such as a host), or multiple physical devices that work together through a network, such as different client instances deployed on a desktop computer, tablet terminal or mobile device respectively.

[0046] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smart mobile terminals, tablet computers, laptop computers, and desktop computers.

[0047] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0048] It should be noted that the work order processing method provided in this application embodiment can generally be executed by server 105. Correspondingly, the work order processing device provided in this application embodiment can generally be located in server 105. The work order processing method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the work order processing device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0049] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0050] The following will be based on Figure 1 The described scene, through Figures 2-4 The work order processing method of the disclosed embodiments is described in detail.

[0051] Figure 2 A flowchart illustrating a work order processing method according to an embodiment of this application is shown schematically.

[0052] like Figure 2 As shown, the work order processing method 200 of this embodiment includes operations S210 to S240.

[0053] In operation S210, obtain the target work order information.

[0054] In some embodiments, target work order information can be obtained in various ways to adapt to the information system architecture of different banks, financial institutions, or enterprises. For example, target work order information may include: monitoring alarm work orders automatically generated by the information monitoring system, business work orders formed by feedback from business channels, problem reports automatically extracted by the system log or event tracking module, and user problem records entered by the customer service platform or manual work order system.

[0055] In some embodiments, target work order information can be obtained through interface integration with the bank's existing information monitoring system. The system can receive alarm event data generated by the monitoring platform in real time, such as abnormal server performance (e.g., CPU utilization consistently exceeding a threshold), database connection failures, transaction timeouts, and excessive network bandwidth usage. When the monitoring system detects that operational indicators exceed preset thresholds, it automatically generates a monitoring alarm work order and pushes it to this system.

[0056] In other embodiments, the target work order information may originate from feedback data. When a customer encounters a system anomaly while conducting business (such as failed transfer, page loading timeout, or inability to query financial products), the business front-end system can automatically generate a work order and report it to the platform through an event capture mechanism, or customer service personnel can enter the work order information into the problem handling system.

[0057] In another embodiment, target work order information can be automatically generated through a log aggregation and event extraction module. The system can periodically scan server operation logs, transaction logs, application error logs, etc., and use anomaly detection algorithms or log semantic recognition models to extract problem description information from unstructured log text, automatically generating work order records with problem semantic summaries. For example, if log parsing reveals that multiple consecutive transaction response times exceed the average level, the system can automatically generate a "transaction delay" work order and mark the potential business impact.

[0058] In another embodiment, the target work order information can also be obtained through a third-party collaborative system, such as interface data provided by outsourced maintenance units, cooperative service providers, or cloud platform operation and maintenance systems. By connecting with a secure communication channel with external systems, external fault report information can be periodically retrieved or subscribed to in real time, thereby achieving unified access and summary management of cross-system work orders.

[0059] In some embodiments, to improve data quality, the process of obtaining target work order information may also include data cleaning and preprocessing steps. The system can perform operations such as field alignment, deduplication, format standardization, timestamp unification, and redundant information elimination on work orders from different sources. For the text description portion, natural language preprocessing algorithms (such as word segmentation, stop word removal, named entity recognition, etc.) can be used to extract key semantic fields to ensure that the problem features can be accurately identified in the subsequent semantic matching stage.

[0060] In some embodiments, to further enhance the contextual understandability of work order data, the system can also synchronize relevant environmental information from associated systems, such as server operating status, network topology information, recent version change records, and configuration file modification logs. By associating the target work order information with this contextual data, "enhanced work order information" with rich semantic features can be formed, providing a more precise input feature space for the work order agent.

[0061] In operation S220, based on the target work order information and the pre-built multi-source knowledge base, the target assignment position corresponding to the target work order information is obtained by using the work order intelligent agent. The work order intelligent agent includes a work order processing model, which is obtained by adjusting the target parameters of the pre-trained language model under the constraint of the task loss function through a parameter fine-tuning algorithm. The task loss function is constructed based on the semantic association graph in the multi-source knowledge base.

[0062] In some embodiments, the work order agent can be used to perform semantic analysis and job decision-making on target work order information. Specifically, the acquired target work order information can be input into the work order agent, which then calls its work order processing model to perform semantic encoding processing on the work order text, converting the work order text into a high-dimensional semantic vector representation. Subsequently, based on the semantic understanding capabilities of the work order processing model and combined with a semantic association graph constructed in a multi-source knowledge base, the work order agent performs semantic recognition and association retrieval on the problem category, involved system modules, potential fault causes, etc., in the work order description, thereby determining the most matching responsible job or job combination for the work order.

[0063] In some embodiments, the work order agent can configure a preset task assignment template based on prompt words to guide the work order processing model to execute assignment tasks in different scenarios. For example, in a banking business scenario, the task assignment template can be defined as: "You are a work order assignment expert. Based on the work order problem description, the systems involved, and the relationship between the problem and the job in the knowledge base, determine which team should be responsible for handling the problem and provide a brief reason." After receiving the target work order information, the work order agent uses this task assignment template to prompt the work order processing model, guiding the model to call historical work order data, maintenance manuals, and job responsibility documents from multiple source knowledge bases for semantic association analysis, thereby outputting the target assignment job and its assignment reason.

[0064] In another embodiment, the work order agent can perform reasoning based on the similarity of semantic paths when making assignment decisions. For example, when keywords such as "transfer failure," "database connection error," and "response timeout" appear in the work order text, the work order agent uses the work order processing model to search for semantic paths with high relevance to these problem nodes in the semantic association graph, and finds the corresponding job node, such as "core transaction system development team" or "database maintenance position," thereby determining the target assignment position.

[0065] In another embodiment, the work order processing model is obtained by adjusting the target parameters of the pre-trained language model under the constraint of the task loss function using a parameter fine-tuning algorithm. Specifically, the system first constructs a training corpus using a large amount of work order sample data accumulated in banking or enterprise scenarios, and associates each sample with its correct job responsibility label. Subsequently, the parameter fine-tuning algorithm is used to train some target parameters of the pre-trained language model, making the output of the work order processing model more consistent with the specific semantic features of work order assignment tasks and improving its job responsibility determination ability in work order scenarios.

[0066] In another embodiment, the task loss function is constructed based on a semantic relationship graph in a multi-source knowledge base. The system generates structured semantic constraints by analyzing the semantic relationships between nodes in the knowledge base. During training, the model calculates the loss value based on these semantic paths, enabling the model to learn the potential relationships between different types of work orders and job positions.

[0067] In some embodiments, the system may also introduce a multi-stage reasoning mechanism. Specifically, after the initial assignment is completed, the work order AI will perform verification reasoning again based on the knowledge base. By judging the similarity of related work orders or by backtracking the feedback results of job positions, the system can confirm the rationality of the initial assignment results. If a situation is found where semantics are similar but there is a conflict in job position determination, the model can re-weight the assignment results based on semantic distance and output a final job position with higher confidence.

[0068] In some embodiments, to address complex work order scenarios involving cross-system and multi-team collaboration, the work order agent can output a composite assignment structure of primary and secondary roles. For example, for work orders such as "network latency caused payment transaction timeout," the model can identify that the problem involves two types of system modules: network infrastructure and transaction application, thereby generating an assignment result of "network maintenance group as primary responsible and transaction system group as secondary responsible."

[0069] In another embodiment, to achieve continuous optimization, the work order agent can periodically receive feedback data based on historical work order assignment results, which is used to correct the work order processing model weights and semantic association parameters. The system can automatically compare the consistency between the model assignment results and the manual confirmation of positions, and use the discrepancies as negative samples for reverse optimization training, so that the model's assignment capability gradually approaches the level of human experts.

[0070] In other embodiments, to meet the application needs of different types of enterprises or industries, the system can also deploy a lightweight work order agent. For example, in scenarios involving small and medium-sized financial institutions or manufacturing enterprises, only the task loss function constraint and knowledge base matching module can be retained, without performing full parameter fine-tuning, thereby achieving dispatch performance close to that of a large model under limited computing resources.

[0071] In operation S230, the work order agent is used to perform work order analysis and generate task processing decision information.

[0072] In some embodiments, the work order agent can perform multi-factor comprehensive analysis by combining contextual dynamic information. For example, when performing work order analysis, the system can simultaneously call real-time operating parameters, monitoring indicators (such as server load, transaction success rate, network latency, etc.) and log data associated with the target work order to form the work order's operational feature vector. After integrating this real-time data and knowledge base information, the model can automatically determine whether the cause of the fault is time-sensitive or systemic. For example, when the model identifies that "high CPU utilization" and "database query response latency" are positively correlated in the time series, it can infer that the problem belongs to the performance bottleneck category and provide a handling suggestion of "optimizing the index or expanding the computing node" in the task processing decision information.

[0073] In some embodiments, the work order agent can generate multi-level decision information based on a hierarchical reasoning structure. The first layer of reasoning is used to identify the problem category, such as performance, connectivity, permission, or configuration; the second layer of reasoning is used to locate the specific problem module, such as "middleware connection pool," "core transaction module," or "database index structure"; the third layer of reasoning combines historical cases and rule knowledge to generate operation suggestions, such as "restart service," "modify parameters," or "clear cache." The model can include a confidence score when outputting decision information, enabling operations and maintenance personnel to choose between automatic execution or manual review based on the confidence level, thereby achieving a closed loop of human-machine collaboration.

[0074] In another embodiment, the task processing decision information may include three categories: problem diagnosis report, recommended handling strategies, and verification plan. The problem diagnosis report explains the cause and scope of the failure; the recommended handling strategies provide multiple feasible solutions; and the verification plan guides the evaluation of the effect after processing, such as confirming whether the problem has been completely resolved through performance metric regression or log comparison.

[0075] In some embodiments, to adapt to business scenarios involving cross-system or multi-team collaboration, the work order agent can automatically identify the system boundaries and collaborative relationships involved when generating task processing decision information. When the model determines that a work order issue requires collaboration among multiple departments, it can add a description of "primary responsible position - collaborating position - task boundary" to the decision information, for example: "Primary responsible position: Database maintenance group; Collaborating position: Application development group; Task boundary: Structured Query Language statement optimization and interface parameter checking".

[0076] In another embodiment, the work order agent can also employ differentiated analysis models based on different business domains or system types. For example, in a financial transaction system, the model can focus on analyzing transaction logs, order status, and settlement delays; in a network infrastructure scenario, the model can focus on bandwidth utilization, network topology changes, and device health. The system can automatically invoke the corresponding domain sub-model for analysis based on the system type to which the work order belongs, in order to generate more targeted decision-making information.

[0077] In some embodiments, the generated task processing decision information can also be displayed in a visual form, such as through structured reports, flowcharts, or cause-and-effect diagrams to show the root causes of problems, handling paths, and verification steps. The system can also provide one-click execution or automated task scheduling interfaces, enabling high-confidence task solutions to directly trigger automated script execution, thereby reducing manual intervention.

[0078] In operation S240, the task processing decision information is sent to the target assignment post, and the work order processing result returned by the target assignment post is obtained.

[0079] In some embodiments, operation S240 can connect the task processing decision information generated by the intelligent agent with the actual work order processing stage. Specifically, after the work order intelligent agent generates task processing decision information, the system sends this decision information to the work order processing terminal of the target assignment position through a unified task interaction interface. The target assignment position can be different operation and maintenance teams, application development departments, network maintenance groups, or customer support centers within a bank or enterprise. When transmitting task processing decision information, the system can attach auxiliary information such as task descriptions, recommended operation steps, verification indicators, and expected results to enable personnel in the target position to quickly understand the task content and perform the corresponding processing operations.

[0080] For example, task processing decision information can be automatically pushed to the target responsible team through the enterprise's internal work order management platform interface. For instance, when the system determines the target assignment position is the "Core Transaction System Maintenance Group," it can automatically push the task information to the group's work order processing dashboard via message middleware. After receiving the task information on their terminals, maintenance personnel can directly view the decision suggestions generated by the agent, such as "It is recommended to check the core transaction database connection pool parameters and perform a restart operation," and can fill in processing feedback based on the actual execution results. When the system detects that the work order status has been updated to "Processed," it will automatically collect the processing results and store them in the database for subsequent knowledge updates and model retraining.

[0081] In another embodiment, task processing decision information can be sent to the target position via a multi-channel communication mechanism. For example, for cross-departmental collaborative tasks, the system can synchronously distribute decision information between the responsible team and the collaborating teams, automatically assigning task roles and time nodes, thereby achieving real-time collaboration and feedback loops in work order processing.

[0082] In another embodiment, the system can determine the sending strategy based on the confidence score in the task processing decision information. When the confidence score of the decision information is higher than a preset threshold (e.g., 0.9), the system can directly convert the handling instruction into an automated script or command, and call the bank's or enterprise's automated operation and maintenance platform to perform automatic repair operations. For example, when the agent determines that "insufficient disk space" is a problem that can be handled automatically, it can automatically call a script to perform log cleanup and verify whether the space occupancy rate has returned to normal, without manual intervention. Conversely, when the confidence score of the decision information is low, the system will send the task to the target position in a "manual confirmation" mode, and the corresponding handling will be performed after the personnel review it.

[0083] In another embodiment, to ensure the authenticity and accuracy of the processing results, the system can automatically trigger a verification and confirmation mechanism after receiving the work order processing results returned by the target assigned position. The system can obtain real-time operating indicators by calling the monitoring system interface to verify whether the fault has been eliminated. For example, when the processing result includes the operation of "service restarted", the system can automatically detect the operating status and performance indicators of the relevant services, such as response time and transaction success rate. If the detection results have all returned to the normal range, the work order is automatically marked as "closed-loop completed". If the indicators do not meet expectations, the system can mark the work order as "requires review" and re-trigger the analysis process to form a continuous optimization closed loop.

[0084] In some embodiments, to further enhance the transparency and traceability of the task processing, the system can attach interpretable tags to the task decision information. These tags record the knowledge entries, historical cases, and reasoning paths referenced by the model when generating decisions. Target personnel can directly view the "recommendation reasons" when receiving task information, such as "This work order has a similarity of 0.87 to a similar event in [month / year], the problem is caused by network latency, and it is recommended to check the network load balancing configuration." This embodiment enables operations and maintenance personnel to quickly understand the model's decision-making basis, thereby executing tasks more efficiently and avoiding blind operations.

[0085] In another embodiment, the work order processing result returned by the target assignment post includes not only the operation execution status (such as "success" or "failure"), but also structured feedback information, such as processing steps, tools used, modified parameters, execution time, and verification results. After receiving this feedback, the system can perform information extraction and semantic matching, mapping the results back to the corresponding nodes in the multi-source knowledge base. If the result is consistent with the original model suggestion, the confidence of the knowledge entry is increased; if there is a discrepancy, it is marked as a "knowledge conflict" and entered into the manual review queue for knowledge completion or model correction.

[0086] In another embodiment, to support multi-organizational collaborative environments, the system can send task processing decision information to external partners or third-party maintenance organizations via a cross-domain access mechanism. Before transmission, the system can anonymize and encrypt the data, retaining only necessary task parameters and processing information. After completing the task, the external organization returns standardized result data, which the system automatically parses and incorporates into its local knowledge base, enabling knowledge sharing and continuous accumulation among different organizations.

[0087] According to embodiments of this application, by acquiring target work order information, intelligently assigning tasks based on a multi-source knowledge base, performing semantic analysis using a work order intelligent agent to generate task processing decision information, and then sending the decision information to the target position and obtaining the processing result, a fully automated and intelligent closed-loop work order processing workflow is achieved. The embodiments of this application, through the integration of semantic understanding and knowledge reasoning, improve the computational efficiency and decision accuracy of work order assignment and analysis, and reduce manual intervention and communication costs. Simultaneously, through result feedback and knowledge self-learning mechanisms, the system's stability and adaptability are enhanced, making the operation and maintenance process more efficient, interpretable, and continuously optimized, thereby improving the system's intelligent operation and maintenance level and user experience.

[0088] The work order processing method of this application will be specifically described below by way of preferred embodiments.

[0089] In the embodiments of this application, a pre-built multi-source knowledge base is used to provide core support for semantic understanding and reasoning for the work order intelligent agent. Its construction process may include the acquisition of multi-source knowledge data, semantic parsing and relation extraction, and fusion processing based on semantic association graphs.

[0090] Specifically, the system can first automatically collect multi-source knowledge data from bank or enterprise information systems. To ensure the breadth and consistency of knowledge sources, the system can perform formatting and cleaning operations on data from different sources during the data collection phase, transforming unstructured text, tabular data, and mixed text and image data into a parsable data structure. After data collection, semantic parsing and relation extraction algorithms can be used to perform deep processing on the multi-source knowledge data to identify semantic elements. After semantic element identification, the system can construct a semantic association graph based on the extracted semantic relationships. The semantic association graph uses semantic nodes such as problems, causes, actions, and positions as core elements, and relationships such as causality, dependency, inclusion, and invocation as edges, forming a semantic association graph that can represent the logical structure of knowledge. This semantic association graph not only preserves the inherent logical connections between different knowledge items, but also reflects the relative importance and frequency of various relationships through node weights and edge strengths. Graph structure storage technology can be used during the construction process, enabling subsequent large models to quickly locate relevant paths when performing semantic matching and knowledge retrieval, improving knowledge retrieval efficiency.

[0091] Furthermore, a semantic association graph can be used as the core framework for knowledge fusion, enabling semantic-level fusion and deduplication of multi-source knowledge data. During the fusion process, knowledge items from different sources that are semantically repetitive or similar in expression can be clustered and merged, uniformly representing them as standardized semantic nodes. Subsequently, through semantic consistency checks and confidence assessment mechanisms, high-quality knowledge relationships are selected, while conflicting or low-reliability semantic connections are eliminated. After fusion processing, a structured, hierarchical, and dynamically expandable multi-source knowledge base can be formed. This multi-source knowledge base not only retains the original semantic information of various types of knowledge data but also achieves unified representation and logical coherence of cross-domain knowledge through the semantic association graph, providing a high-precision and interpretable knowledge support environment for subsequent work order intelligent agents.

[0092] In the embodiments of this application, to improve the semantic parsing accuracy and structured quality of knowledge extraction from multi-source knowledge data, appropriate processing algorithms can be used to perform semantic parsing and relation extraction for different data types from different sources, so as to fully explore the implicit business logic and causal relationships in various knowledge sources. Specifically, the input multi-source knowledge data can first be identified by source and pre-classified by content. Based on the structural features and semantic density of the data, the corresponding algorithm model can be automatically matched to achieve differentiated processing.

[0093] For example, for historical work order data, a semantic annotation algorithm based on sequence modeling can be used to identify key semantic fragments in the work order description text through an attention mechanism, such as "problem phenomenon", "fault cause", "handling measures" and "involved positions", thereby establishing a time series and semantic correspondence between the work order title, problem description and handling record, extracting the complete "problem - cause - handling - position" semantic chain, and mapping it into traceable semantic entities and relation pairs to form standardized knowledge entries.

[0094] For example, unstructured document data, such as system maintenance manuals, configuration instructions, interface documents, or project reports, can be parsed using semantic segmentation and entity relationship extraction algorithms based on the Transformer architecture. This algorithm can automatically identify logically related content fragments in the document, such as system modules, functionalities, operating environments, and abnormal scenarios, and model their semantic relationships, such as "module contains function," "function triggers failure," and "environment affects performance," thereby transforming the document content, originally presented in natural language, into a structured semantic relationship network.

[0095] For example, semantic clustering and pattern recognition algorithms can be used to aggregate and summarize expert experience data. Expert experience is usually expressed as semi-structured text descriptions, such as "When the system response delay exceeds the threshold, the index failure problem should be checked first." The system merges experience knowledge with the same or similar expressions into a unified knowledge pattern through semantic similarity calculation and cluster analysis, thereby eliminating individual expression differences and improving the universality and reusability of knowledge.

[0096] For example, for external knowledge data, such as content obtained from publicly available industry information, technical forums, or professional databases, text summarization and semantic alignment algorithms can be used for semantic compression and structural mapping. This algorithm automatically summarizes and extracts core concepts and knowledge associations from external text, and aligns its semantic space with the concept system in the internal knowledge base, enabling external knowledge to be seamlessly integrated into the local knowledge network while maintaining its original semantic integrity.

[0097] Through the synergistic application of the aforementioned differentiated algorithms, the embodiments of this application achieve adaptive semantic parsing and accurate relation extraction for heterogeneous data sources. This enables the system to effectively identify problem causes, business logic, and technological dependencies in multi-type knowledge environments, forming semantically consistent, structurally clear, and computable knowledge representations. This source-specific processing mechanism not only improves the automation level and knowledge accuracy of knowledge base construction but also provides a higher-quality knowledge input foundation for the subsequent training and reasoning of work order agents.

[0098] Figure 3The flowchart illustrates a method for obtaining target assignment positions corresponding to target work order information according to some exemplary embodiments of this application.

[0099] like Figure 3 As shown, the method for obtaining the target work order information and the corresponding target assignment position includes operations S310 to S330.

[0100] In operation S310, the target work order information is input into the work order intelligent body.

[0101] In the process of processing, the work order intelligent agent can semantically encode the work order title, problem description and related context content, and convert them into quantifiable semantic vector representations for subsequent knowledge retrieval and job matching.

[0102] In operation S320, based on the task assignment template pre-built through prompt word engineering, the work order agent is guided to call the work order processing model and access the multi-source knowledge base to perform semantic matching on the target work order information.

[0103] In the embodiments of this application, to achieve automatic work order assignment and explainable decision-making, a work order intelligent agent can be used to perform semantic recognition and knowledge matching on target work order information to determine the corresponding target assignment position and assignment reason. Specifically, after the target work order information is input into the work order intelligent agent, the agent first prompts the work order processing model according to a preset assignment task template, guiding the model to understand and encode semantic elements such as system-related information, operating status information, and fault phenomena in the work order text. Subsequently, based on the semantic representation output by the work order processing model, the agent retrieves corresponding semantic nodes and historical case paths from a multi-source knowledge base and performs semantic matching and association reasoning. By comparing similar historical work orders and their associated job paths, the agent can identify the most likely responsible target assignment position and generate corresponding assignment suggestions and explanations to support decision-making in subsequent work order processing stages.

[0104] In operation S330, the target assigned position and the corresponding assignment reason are obtained based on the semantic matching result.

[0105] In this process, the multi-source knowledge base integrates semantic elements such as historical work orders, system documents, and job responsibilities. The work order agent can determine the target assigned position based on semantic matching strength, node similarity, and path credibility. Furthermore, in complex scenarios (e.g., work orders involving cross-position collaboration, such as "network latency causing transaction timeout" requiring cooperation between network and application roles), it can output a "primary responsibility - collaborative" job combination and division of responsibilities (e.g., "network role is responsible for investigating link latency, application role is responsible for optimizing transaction timeout retry mechanism"). For example, for the work order "transaction interface response timeout," the system may identify the "core system maintenance group" as the target assigned position and provide an explanation based on past cases: "the problem involves a core module performance bottleneck." This implementation not only improves the automation and accuracy of work order assignment but also enhances the interpretability and transparency of the model output, thereby improving overall processing efficiency and user trust.

[0106] Figure 4 The illustration shows a schematic diagram of a method for adjusting target parameters of a pre-trained language model according to some exemplary embodiments of the present application.

[0107] like Figure 4 As shown, the method for adjusting the target parameters of the pre-trained language model may include operations S410 to S420.

[0108] In operation S410, while keeping the backbone parameters of the pre-trained language model frozen, the target trainable parameters are updated.

[0109] In the embodiments of this application, the pre-trained language model is used to provide basic semantic understanding and contextual reasoning capabilities for the work order processing model. Through self-supervised learning on a large-scale general corpus, it possesses rich language representation capabilities and cross-domain knowledge capture capabilities, enabling comprehensive modeling of natural language text at the lexical, syntactic, and semantic levels. This provides a general semantic foundation for subsequent task fine-tuning and knowledge alignment. The pre-trained language model can adopt an encoder-decoder structure, an encoder-only structure, or a decoder-only structure, etc. Internally, it captures the dependencies between different semantic units through a multi-layer attention mechanism, achieving context-dependent language representation.

[0110] In the embodiments of this application, the backbone parameters can correspond to the general language understanding layer of the work order processing model, used to maintain the model's general semantic representation capability; the target trainable parameters can include task-related layer parameters or specific adaptation module parameters, used to capture semantic features closely related to the work order processing task. By updating only the target parameters, the training computation and memory overhead can be reduced, while avoiding damage to the model's original semantic capabilities, thereby achieving fast and efficient domain adaptation under small sample conditions.

[0111] In operation S420, during each training round, the learning rate is dynamically adjusted based on the gradient change magnitude determined by the semantic node weights in the task loss function.

[0112] In the embodiments of this application, to improve the stability and convergence efficiency of the work order processing model training, the learning rate can be adaptively adjusted during training by incorporating the semantic structure information of the task loss function. Specifically, the task loss function can be constructed based on the semantic association graph in a multi-source knowledge base, and the weights of different semantic nodes are used to characterize the importance of corresponding problem information, cause information, handling information, and job information in the work order task. During each round of gradient update, the system can dynamically amplify or decay the learning rate of the target trainable parameters according to the semantic node weights and the corresponding gradient change magnitude, so that parameters related to high-weight semantic nodes receive larger update steps, in order to focus on learning semantic patterns that have a greater impact on work order assignment and work order analysis; parameters related to low-weight semantic nodes adopt a relatively gentle update strategy to maintain the stability of the overall training process. Through this dynamic learning rate adjustment mechanism based on the task semantic structure, the optimization of effective parameters can be accelerated while ensuring the smooth convergence of the model, reducing the risk of overfitting and gradient oscillation, thereby enabling a more substantial improvement in the performance of the work order processing model on work order processing tasks.

[0113] Specifically, to enable the work order processing model to better learn the semantic rules upon which the work order processing task depends during parameter optimization, a task loss function can be constructed based on the semantic association graph in a multi-source knowledge base, ensuring that the model's training objective aligns with the knowledge's semantic structure. This task loss function, while considering basic task objectives such as work order assignment or problem classification, introduces constraints derived from the semantic association graph. This allows the work order processing model to not only reduce the deviation between predicted results and labeled tags during training but also simultaneously optimize semantic consistency with the knowledge system.

[0114] In the specific implementation process, the target association path can be extracted from the semantic association graph first. The semantic association graph uses problem information, cause information, handling information, and job information in the knowledge base as core nodes, and the nodes form a complete knowledge semantic chain through causal, dependency, or pointing relationships. For example, "transaction failure - database connection error - service restart restores normal operation - database maintenance position" is a typical association path. By extracting such paths in a structured manner, the system obtains the semantic relationship patterns that should be the focus of task training.

[0115] After extracting the target association path, constraints can be constructed based on node similarity and edge weights within the path. Node similarity measures the proximity between the prediction result given by the work order processing model in the semantic space and the target node in the semantic association graph, while edge weights reflect the importance and association strength of different semantic node relationships. Based on this information, the system forms constraints that ensure the work order processing model, during training, not only minimizes the loss of the basic task but also considers the coherence and rationality of node relationships in the semantic association graph. For example, when the fault cause output by the work order processing model deviates significantly from the target cause node connected to the problem information in the semantic association graph in the semantic space, the corresponding constraints will enhance the loss of that sample, guiding the work order processing model to adjust relevant parameters and gradually bring the prediction result closer to the known and reasonable semantic relationships in the knowledge graph.

[0116] Building upon this, the constraint relationships constructed based on node similarity and edge weights can be combined with the basic task loss of the work order processing model to form a task loss function. This task loss function comprehensively considers the task accuracy of the work order processing model in predicting job positions, problem categories, or fault causes, as well as its consistency with the semantic association graph. This allows the work order processing model to learn the association patterns and task mapping relationships between problem information, cause information, handling information, and job information at the semantic level during training. Through this knowledge structure-oriented loss design, the work order processing model can more fully utilize the semantic association information accumulated in multi-source knowledge bases during convergence, exhibiting higher robustness and generalization ability in actual work order assignment and fault analysis scenarios.

[0117] Furthermore, to achieve automated analysis and intelligent decision-making for target work orders, a work order intelligent agent can be used to perform semantic understanding, case retrieval, and strategy reasoning on the target work order information, thereby generating task processing decision information. During this process, the work order intelligent agent can call its work order processing model to encode and semantically represent the work order text, and combine structured knowledge and semantic relationships from a multi-source knowledge base to complete the intelligent analysis process from problem identification to handling strategy generation without human intervention. Through the synergistic effect of semantic retrieval and knowledge matching mechanisms, the work order intelligent agent can output executable handling strategies and / or verification schemes under the comprehensive constraints of historical experience, knowledge logic, and current context features, providing direct decision-making basis for subsequent manual or automated handling.

[0118] Specifically, similar cases can be retrieved from a multi-source knowledge base based on the target work order information. This knowledge base stores a large number of closed-loop work order records along with their corresponding problem categories, fault causes, handling steps, and verification results. The work order agent can invoke the work order processing model to semantically encode the target work order text, obtaining its semantic vector representation. This representation is then compared with the semantic vectors of existing work orders in the multi-source knowledge base to calculate similarity, thereby selecting several historical cases that are semantically closest to the target work order. By comparing and analyzing the semantic features and contextual logic of the problem descriptions, impact scope, and handling results in these similar cases, the work order agent can automatically determine the candidate problem categories for the target work order, such as performance bottlenecks, database anomalies, interface timeouts, and configuration errors, providing directional references for subsequent, more refined fault localization and strategy recommendations.

[0119] After identifying candidate problem categories, semantic matching and reasoning can be performed on these categories to generate task processing decision information containing at least one handling strategy and / or verification scheme. During this process, the work order agent can leverage the semantic association graph structure in a multi-source knowledge base to retrieve the "problem-cause-handling-verification" association path corresponding to the candidate problem category. Based on indicators such as semantic similarity, association path weight, and the success rate of historical processing results, multiple candidate schemes are scored and ranked to select the preferred or alternative handling scheme. For example, when the problem category is identified as "database connection timeout," the work order agent can generate task processing decision information similar to "Handling strategy: Restart the database connection pool or adjust the maximum number of connections in the connection pool; Verification scheme: After handling, check whether the response time of key transactions has recovered to within a preset threshold," and return this information as a structured suggestion to the work order handler or automated execution module.

[0120] In some embodiments, when the system finds multiple effective handling strategies for the same problem category in historical cases, the work order agent can output multiple optional suggestions and attach expected effects, applicable preconditions and potential risk warnings to each strategy, so that manual operation and maintenance personnel can make a choice under actual business constraints (such as change window, system importance, regulatory requirements), or the rule engine can automatically select the appropriate solution to be executed based on the current operating status.

[0121] Through the above process, the system can perform semantic-level diagnosis and decision generation on complex work orders without prior human intervention in analysis and judgment, achieving intelligent analysis throughout the entire process from problem identification and category determination to the output of handling strategies and verification solutions. This embodiment effectively improves the accuracy and response speed of work order processing, enabling the work order intelligent agent to not only have automatic dispatch capabilities but also fault reasoning and strategy recommendation capabilities, thereby promoting the evolution of work order processing towards intelligence and closed-loop, reducing the analytical burden on maintenance personnel, and reducing the human uncertainty in fault handling.

[0122] In the embodiments of this application, to achieve continuous evolution of the multi-source knowledge base and dynamic enhancement of model intelligence, a difference comparison and knowledge completion process can be introduced after the work order processing loop is closed. This process is used to automatically identify knowledge gaps and update knowledge. This process can achieve self-improvement of the knowledge base and long-term optimization of model capabilities by analyzing the differences between the work order processing results and task processing decision information, combined with an expert question-and-answer mechanism.

[0123] Specifically, the system can first semantically compare the work order processing results returned by the target assigned position with the task processing decision information generated by the work order agent. The system then uses text semantic matching and result consistency calculation to determine whether the model's predicted causes, solutions, or verification results match the actual execution results. When a discrepancy is detected—for example, the model's recommended "index optimization" solution fails to solve the problem, while the actual solution is "adjust connection pool parameters"—this discrepancy and the corresponding work order information can be marked as a knowledge gap and stored in the knowledge maintenance module for later completion.

[0124] Upon identifying a knowledge gap, a question-and-answer task can be generated to address that gap. Based on this task, a knowledge supplementation request can be sent to relevant domain experts. This task can include contextual information about the missing knowledge and semantic fragments that the model failed to correctly identify, such as "Please supplement other handling measures besides index optimization and their applicable conditions in database connection anomaly scenarios." Domain experts can directly fill in or select appropriate knowledge content in the interactive interface. After receiving feedback, the system performs semantic parsing and formatting on the expert responses to ensure the content conforms to the structured requirements of the knowledge base.

[0125] After receiving supplementary information from experts, the information undergoes further manual review. Once approved, the newly added knowledge content is transformed into structured knowledge entries, generating standard semantic units containing elements such as problem, cause, action, and verification. These units are then automatically updated to the corresponding knowledge graph nodes in the multi-source knowledge base. If a similar problem reappears in subsequent work orders, the model can directly invoke the newly supplemented knowledge path, thus avoiding repeated erroneous reasoning. Through this cyclical mechanism of difference comparison, question-and-answer supplementation, and dynamic updates, the system achieves self-learning and self-evolution of the knowledge base, enabling the work order agent to continuously improve its analytical capabilities and decision-making accuracy over long-term operation. This embodiment effectively solves the problem of static knowledge bases being prone to aging and lagging updates in complex business environments, ensuring that the system maintains the integrity, timeliness, and usability of knowledge in constantly changing operational scenarios.

[0126] According to embodiments of this application, a knowledge system encompassing elements such as problems, causes, handling procedures, and job positions is constructed through multi-source data fusion and semantic parsing. This enables the large model to perform semantic understanding and decision-making reasoning based on structured knowledge. Furthermore, by designing parameter fine-tuning algorithms and semantic constraint loss functions, the pre-trained language model possesses domain-specific semantic focusing capabilities and stable convergence performance, thereby improving computational efficiency and reasoning accuracy in work order assignment and problem analysis. Through prompt word engineering, the large model is guided to perform job matching and semantic analysis, achieving interpretable responsibility job assignment and task strategy generation. Combined with confidence control mechanisms and historical case retrieval, the automation level and accuracy of assignment are significantly improved. When performing work order analysis, the model can intelligently generate task processing decision information containing multiple solutions, providing decision-making basis for manual maintenance or automated execution systems. In the work order closure phase, the system introduces a result comparison and knowledge completion process, automatically identifying differences between model output and actual processing, and supplementing missing knowledge through an expert question-and-answer mechanism, achieving continuous self-learning and dynamic evolution of the knowledge base.

[0127] Overall, the embodiments of this application improve the automation level, semantic understanding depth and knowledge update efficiency of work order processing through the system design of multi-source knowledge fusion, semantic constraint optimization, intelligent decision generation and self-learning closed loop. It can maintain high accuracy and high stability in complex and high-concurrency production environments, thereby effectively reducing operation and maintenance manpower costs and response latency, and improving business continuity and system reliability.

[0128] Corresponding to the above-described work order processing method, embodiments of this application also provide a work order processing device.

[0129] Figure 5 A schematic block diagram of a work order processing apparatus according to an embodiment of this application is shown.

[0130] like Figure 5As shown, the work order processing device 500 of this embodiment includes a data acquisition module 510, a work order dispatch module 520, a work order analysis module 530, and a work order processing module 540.

[0131] The data acquisition module 510 can be used to acquire target work order information. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.

[0132] The work order dispatch module 520 can be used to obtain the target dispatch position corresponding to the target work order information based on the target work order information and a pre-built multi-source knowledge base, using a work order agent. The work order agent includes a work order processing model, which is obtained by adjusting the target parameters of a pre-trained language model under the constraint of a task loss function using a parameter fine-tuning algorithm. The task loss function is constructed based on the semantic association graph in the multi-source knowledge base. In one embodiment, the work order dispatch module 520 can be used to execute the operation S220 described above, which will not be repeated here.

[0133] The work order analysis module 530 can be used to perform work order analysis using the work order intelligent agent and generate task processing decision information. In one embodiment, the work order analysis module 530 can be used to perform the operation S230 described above, which will not be repeated here.

[0134] The work order processing module 540 can be used to send the task processing decision information to the target assignment post and obtain the work order processing result returned by the target assignment post. In one embodiment, the work order processing module 540 can be used to execute the operation S240 described above, which will not be repeated here.

[0135] According to an embodiment of this application, the work order processing device 500 may further include a knowledge base construction module. The knowledge base construction module can be used to acquire multi-source knowledge data; perform semantic parsing and relation extraction on the multi-source knowledge data to identify semantic elements and corresponding relationships, thereby constructing the semantic association graph; and fuse the multi-source knowledge data based on the semantic association graph to form the multi-source knowledge base.

[0136] According to embodiments of this application, the knowledge base construction module can also be used to perform semantic parsing and relation extraction on knowledge data from different sources using corresponding processing algorithms. The multi-source knowledge data includes at least historical work order data, unstructured document data, expert experience data, and external knowledge data.

[0137] According to embodiments of this application, the knowledge base construction module can also be used to: obtain work order semantic elements using sequence labeling algorithms for the historical work order data; extract functional semantic elements using document semantic segmentation and entity relationship acquisition algorithms for the unstructured document data, wherein the functional semantic elements are used to characterize the semantic information of system structure, function, and operating scenario; identify similar problem patterns and obtain supplementary knowledge information using semantic clustering analysis for the expert experience data; and obtain structured knowledge fragments using text summarization and semantic alignment algorithms for the external knowledge data.

[0138] According to an embodiment of this application, the work order dispatch module 520 can also be used to input the target work order information into the work order agent; based on the dispatch task template pre-constructed through prompt word engineering, guide the work order agent to call the work order processing model and access the multi-source knowledge base to perform semantic matching on the target work order information; and obtain the target dispatch position and the corresponding dispatch reason based on the semantic matching result.

[0139] According to embodiments of this application, the work order processing device 500 may further include a training module. The training module can be used to update the target trainable parameters while keeping the backbone parameters of the pre-trained language model frozen; and to dynamically adjust the parameter learning rate based on the gradient change magnitude determined by the semantic node weights in the task loss function during each training round.

[0140] According to an embodiment of this application, the training module can also be used to extract target association paths from the semantic association graph, wherein the target association paths include at least problem information, cause information, handling information, and job information; establish constraint relationships based on node similarity and edge weights in the target association paths; and associate the constraint relationships with the model output results to construct the task loss function.

[0141] According to an embodiment of this application, the work order analysis module 530 can also be used to retrieve similar cases in the multi-source knowledge base based on the target work order information, determine candidate problem categories, and perform semantic matching on the candidate problem categories to generate task processing decision information containing at least one handling strategy or verification scheme.

[0142] According to an embodiment of this application, the anomaly identification device 500 may further include an update module. The update module can be used to compare the work order processing result with the task processing decision information; in response to the existence of a difference, mark the difference content and the corresponding work order information as a knowledge gap; generate a question-and-answer task for the knowledge gap; initiate a question-and-answer task request to a domain expert based on the question-and-answer task to obtain supplementary missing knowledge information; and convert the supplementary missing knowledge information into structured knowledge entries and update them to the multi-source knowledge base.

[0143] According to embodiments of this application, any multiple modules among the data acquisition module 510, work order dispatch module 520, work order analysis module 530, and work order processing module 540 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the data acquisition module 510, work order dispatch module 520, work order analysis module 530, and work order processing module 540 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any one of the three implementation methods, or in a suitable combination of any of them. Alternatively, at least one of the data acquisition module 510, work order dispatch module 520, work order analysis module 530, and work order processing module 540 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0144] Figure 6 A block diagram schematically illustrates an electronic device suitable for implementing a work order processing method according to an embodiment of this application.

[0145] like Figure 6 As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0146] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0147] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0148] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0149] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0150] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the work order processing method provided in the embodiments of this application.

[0151] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0152] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0153] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0154] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0155] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0156] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0157] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A work order processing method, characterized in that, The method includes: Obtain target work order information; Based on the target work order information and the pre-built multi-source knowledge base, the target assignment position corresponding to the target work order information is obtained by using a work order intelligent agent. The work order intelligent agent includes a work order processing model, which is obtained by adjusting the target parameters of a pre-trained language model under the constraint of a task loss function through a parameter fine-tuning algorithm. The task loss function is constructed based on the semantic association graph in the multi-source knowledge base. The work order agent is used to perform work order analysis and generate task processing decision information; and The task processing decision information is sent to the target assignment post, and the work order processing result returned by the target assignment post is obtained.

2. The method according to claim 1, characterized in that, Pre-build a multi-source knowledge base, including: Acquire multi-source knowledge data; Semantic parsing and relation extraction are performed on the multi-source knowledge data to identify semantic elements and corresponding relationships, thereby constructing the semantic relationship graph; and The multi-source knowledge data is fused based on the semantic association graph to form the multi-source knowledge base.

3. The method according to claim 2, characterized in that, The step of performing semantic parsing and relation extraction on the multi-source knowledge data includes: Semantic parsing and relation extraction are performed on knowledge data from different sources using corresponding processing algorithms. The multi-source knowledge data includes at least historical work order data, unstructured document data, expert experience data, and external knowledge data.

4. The method according to claim 3, characterized in that, The semantic parsing and relation extraction for knowledge data from different sources are performed using corresponding processing algorithms, including: For the historical work order data, the semantic elements of the work order are obtained using a sequence labeling algorithm; For the unstructured document data, functional semantic elements are extracted using document semantic segmentation and entity relationship acquisition algorithms. These functional semantic elements are used to characterize the semantic information of the system structure, function, and operating scenario. For the aforementioned expert experience data, semantic clustering analysis is used to identify similar problem patterns and obtain supplementary knowledge information; and For the external knowledge data, structured knowledge fragments are obtained using text summarization and semantic alignment algorithms.

5. The method according to claim 1, characterized in that, The step of obtaining the target assignment position corresponding to the target work order information using the work order intelligence agent includes: The target work order information is input into the work order intelligent agent; Based on the task assignment template pre-built through prompt word engineering, the work order agent is guided to call the work order processing model and access the multi-source knowledge base to perform semantic matching on the target work order information. as well as The target assigned position and the corresponding assignment reason are obtained based on the semantic matching results.

6. The method according to claim 1, characterized in that, The adjustment of the target parameters of the pre-trained language model under the constraint of the task loss function through parameter fine-tuning algorithm includes: While keeping the backbone parameters of the pre-trained language model frozen, update the target trainable parameters; and During each training round, the learning rate is dynamically adjusted based on the gradient change magnitude determined by the semantic node weights in the task loss function.

7. The method according to claim 1, characterized in that, The task loss function is constructed based on the semantic association graph in the multi-source knowledge base, including: Extract target association paths from the semantic association graph, wherein the target association paths include at least problem information, cause information, handling information, and job information; Based on the node similarity and edge weights in the target association path, a constraint relationship is established; as well as The constraints and model outputs are correlated to construct the task loss function.

8. The method according to claim 1, characterized in that, The step of using the work order intelligence agent to perform work order analysis and generate task processing decision information includes: Based on the target work order information, similar cases are retrieved from the multi-source knowledge base to determine candidate problem categories; and Semantic matching is performed on the candidate problem categories to generate task processing decision information containing at least one disposal strategy or verification scheme.

9. The method according to claim 1 or 8, characterized in that, The method further includes: The work order processing result and the task processing decision information are compared for differences. In response to the existence of differences, the differences and the corresponding work order information are marked as knowledge gaps. A question-and-answer task is generated to address the knowledge gap, and a question-and-answer task request is initiated to domain experts based on the question-and-answer task to obtain supplementary missing knowledge information; as well as The supplemented missing knowledge information is transformed into structured knowledge entries and updated in the multi-source knowledge base.

10. A work order processing device, characterized in that, The device includes: The data acquisition module is used to: acquire target work order information; The work order assignment module is used to: obtain the target assignment position corresponding to the target work order information using a work order agent based on the target work order information and a pre-built multi-source knowledge base. The work order agent includes a work order processing model, which is obtained by adjusting the target parameters of a pre-trained language model under the constraint of a task loss function through a parameter fine-tuning algorithm. The task loss function is constructed based on the semantic association graph in the multi-source knowledge base. The work order analysis module is used to: perform work order analysis using the work order intelligent agent to generate task processing decision information; and The work order processing module is used to: send the task processing decision information to the target assignment post and obtain the work order processing result returned by the target assignment post.

11. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 9.

13. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 9.

Images