A data processing method, device and equipment based on a large language model

Abstract

This application relates to the field of data processing, and more particularly to a data processing method, apparatus, and device based on a large language model. The embodiments of this application skip the code generation and compilation processes, directly realizing the logical parsing of data processing instructions to obtain the processing flow. Furthermore, by displaying the processing flow, users can understand the entire processing logic. This not only significantly reduces computing power consumption, but more importantly, eliminates the barrier between ordinary users and complex IT systems, enabling users to directly drive data processing models through natural language, achieving low-threshold, high-efficiency business automation.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a data processing method, apparatus and device based on a large language model. Background Technology

[0002] With the widespread application of large language models in the field of natural language processing, their value in vertical domain question answering systems and retrieval question answering systems is becoming increasingly prominent. By pre-training on general corpora and fine-tuning with domain data, large language models can accomplish complex question answering and information retrieval tasks.

[0003] In related technologies, large language models often rely on Transformer matrix-vector operations for question answering and retrieval. This requires converting the user's input intent into traditional code such as Python, then running the generated code to produce and output the corresponding processing results. In these technologies, the entire data processing process is implemented internally by the large language model based on experience. This process is invisible and uncontrollable, which leads to the final result potentially not matching the user's intended outcome—that is, the final result does not match the intent. Summary of the Invention

[0004] This application provides a data processing method, apparatus, and device based on a large language model to solve the problem in related technologies where the data processing process is implemented internally by the large language model based on experience, which is invisible and uncontrollable, resulting in a mismatch between the final processing result and the intent.

[0005] In a first aspect, embodiments of this application provide a data processing method based on a large language model, the method comprising: Receive input data processing instructions; The data processing instruction is sent to the data processing model for parsing, and the processing flow output by the data processing model in response to the data processing instruction is obtained. The processing flow is demonstrated. If a confirmation instruction for the processing flow is received, the data processing model is controlled to respond to the data processing instruction based on the processing flow, and the data processing result output by the data processing model is obtained.

[0006] Furthermore, if the data processing instruction carries first constraint information, sending the data processing instruction to the data processing model for parsing includes: The first constraint information is sent to the data processing model for parsing; The first constraint information contains user requirements, data processing logic, and data information to be invoked.

[0007] Furthermore, if the data processing instruction does not carry the first constraint information, before sending the data processing instruction to the data processing model for parsing, the method further includes: Obtain the target intent carried in the data processing instructions; In a pre-configured logical database, search for the second constraint information corresponding to the target intent; The step of sending the data processing instructions to the data processing model for parsing includes: The second constraint information and the data processing instructions are sent to the data processing model for parsing.

[0008] Furthermore, after demonstrating the processing flow, the method further includes: If a modification instruction for the processing flow is received, the content of the sub-process in the processing flow is modified according to the information of the sub-process to be modified and the content to be modified carried in the modification instruction. The modified processing flow is shown.

[0009] Furthermore, after obtaining the data processing result output by the data processing model, the method further includes: The processing flow and the target intent carried in the data processing instructions are saved to the logical database.

[0010] Furthermore, before sending the data processing instructions to the data processing model for parsing, the method further includes: Determine whether the logical database contains candidate processing flows corresponding to the target intent; If the logical database does not store the candidate processing flow corresponding to the target intent, then the subsequent step of sending the data processing instruction to the data processing model for parsing is executed; If the logical database stores candidate processing flows corresponding to the target intent, then the candidate processing flow is determined as the processing flow, and the subsequent steps of displaying the processing flow are executed.

[0011] Furthermore, before saving the processing flow and the target intent carried in the data processing instruction to the logical database, the method further includes: Determine whether the logical database contains the initial processing flow corresponding to the target intent; If the logical database does not contain the initial processing flow corresponding to the target intent, then the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed. If the logical database contains an initial processing flow corresponding to the target intent, then the initial processing flow stored in the logical database is deleted, and the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed.

[0012] Furthermore, the data processing instructions are in API instruction format or natural language format; After obtaining the data processing result output by the data processing model, the method further includes: If the data processing instruction is in API instruction format, the data processing result is converted into API structured data, and the API structured data is sent to the sending device of the data processing instruction. If the data processing instruction is in natural language format, the data processing result will be displayed in the input window corresponding to the data processing instruction.

[0013] Secondly, embodiments of this application also provide a data processing apparatus based on a large language model, the apparatus comprising: The receiving module is used to receive input data processing instructions; The logic parsing module is used to send the data processing instructions to the data processing model for parsing, and to obtain the processing flow output by the data processing model when responding to the data processing instructions. The processing module is used to display the processing flow. If a confirmation instruction for the processing flow is received, it controls the data processing model to respond to the data processing instruction based on the processing flow and obtains the data processing result output by the data processing model.

[0014] Thirdly, embodiments of this application also provide an electronic device, which includes at least a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the steps of the data processing method based on a large language model as described in any of the preceding claims.

[0015] This application's embodiments skip the code generation and compilation process, directly realizing the logical parsing of data processing instructions to obtain the processing flow, and by displaying the processing flow, users can understand the entire processing logic. This not only significantly reduces computing power consumption, but more importantly, eliminates the gap between ordinary users and complex IT systems, enabling users to directly drive data processing models through natural language, achieving truly low-threshold, high-efficiency business automation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This application provides a schematic diagram of a data processing process based on a large language model. Figure 2 A data processing flowchart provided for embodiments of this application; Figure 3 A schematic diagram of a data processing device based on a large language model is provided in an embodiment of this application; Figure 4 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0019] The terms "first" and "second" in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising" and any variations thereof are intended to cover non-exclusive protection. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. The term "multiple" in this application can mean at least two, for example, two, three, or more, and the embodiments of this application do not impose limitations.

[0020] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These embodiments should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that in the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solutions of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0021] With the widespread application of large language models in the field of natural language processing, their value in vertical domain question answering systems and retrieval question answering systems is becoming increasingly prominent. By pre-training on general corpora and fine-tuning with domain data, large language models can accomplish complex question answering and information retrieval tasks.

[0022] However, the data processing solutions based on large language models in related technologies have the following problems: 1. In related technologies, the logic execution schemes of large language models mostly rely on Transformer matrix and vector operations, which require converting natural language intent into traditional code such as Python, and then compiling and running it. This has a high threshold for use and a large computational cost.

[0023] 2. The related technologies require the deployment of complex programming environments and runtime dependencies, and have a single input / output mode, which cannot simultaneously meet the needs of manual operation and system API integration.

[0024] 3. The related technologies lack a unified logical storage and visualization construction carrier. After the logic is modified, it needs to be re-encoded and recompiled, resulting in high iteration costs, invisible execution process, and great debugging difficulty.

[0025] 4. When employees handle complex business processes involving multiple systems and steps within an enterprise, the relevant technologies still rely on IT development and cannot automatically automate the processes quickly using natural language.

[0026] 5. Related technologies cannot perform natural language processing, structured data collection, and reuse of complex business requirements. Each execution requires repeated description, resulting in low efficiency and poor consistency.

[0027] Therefore, there is an urgent need for a data processing method that meets the needs of ordinary users for logic execution and business automation with zero programming, low threshold, high efficiency, and instant iteration. This method should be able to respond to data processing instructions accurately and quickly, and achieve zero programming, zero compilation, and zero environment deployment for large language models. It should be able to directly parse and execute data processing instructions, improve reusability, maintainability, and execution consistency, and adapt to manual operation and system API integration.

[0028] Based on this, embodiments of this application propose a data processing method, apparatus, device, and medium based on a large language model. The method includes: receiving an input data processing instruction; sending the data processing instruction to a data processing model for parsing, obtaining a processing flow output by the data processing model in response to the data processing instruction; displaying the processing flow; and if a confirmation instruction for the processing flow is received, controlling the data processing model to respond to the data processing instruction based on the processing flow, and obtaining the data processing result output by the data processing model.

[0029] Example 1: Figure 1 A schematic diagram of a data processing process based on a large language model is provided for an embodiment of this application. The process includes: S101: Receive the input data processing instruction; send the data processing instruction to the data processing model for parsing, and obtain the processing flow output by the data processing model when responding to the data processing instruction.

[0030] This application provides a data processing method based on a large language model, which is applied to an electronic device, such as a PC, a server, or a Starry Sky large-model intelligent agent. The Starry Sky large-model intelligent agent is a large-model task execution and collaborative intelligent agent.

[0031] In this embodiment, the user can input data processing instructions in any way. These methods include, but are not limited to, inputting via an API interface or a client. When inputting data processing instructions via a client, the user can directly input them in a dialog box on the client, or trigger them by clicking buttons displayed on the client's interface. Furthermore, directly inputting in a dialog box on the client includes, but is not limited to, inputting text, voice, or documents. If voice is input, the electronic device can use a preset voice conversion algorithm to convert the voice into text.

[0032] It should be noted that, in the embodiments of this application, regardless of the method by which the data processing instruction is input, the data processing instruction carries at least the intent for data processing. This intent may appear directly in the data processing instruction as an intent identifier, or it may appear in the data processing instruction as data describing the intent. Specifically, the intent identifier may be an intent name, sequence number, summary, etc., and the data describing the intent may be text, voice, etc.

[0033] It should be noted that, in the embodiments of this application, regardless of the channel through which the electronic device receives the data processing instruction, after receiving the instruction, the electronic device will convert the instruction into text in natural language format and perform subsequent processing based on the converted instruction. In any of the following embodiments, the data processing instruction mentioned is a converted data processing instruction, which is actually text in natural language format.

[0034] After receiving a data processing instruction, the electronic device forwards the instruction to a data processing model for parsing. Unlike related technologies, the data processing model does not convert the instruction into various code formats, but instead transforms it into a processing flow.

[0035] Specifically, the data processing model utilizes its own logical reasoning capabilities, combined with pre-loaded business context knowledge, to combine and map the semantic information contained in data processing instructions. The data processing model does not generate executable code files; instead, it directly constructs an abstract processing flow in memory. This processing flow indicates the operational procedures and content that the data processing model performs in response to the data processing instruction. Optionally, the processing flow can take the form of a flowchart, a knowledge graph, a textual description, or a logical topology diagram composed of nodes and connecting lines.

[0036] Based on the above embodiments, in the embodiments of this application, the data processing model can be a large language model, a multi-model large model, a neural network model, or other models with reasoning and learning capabilities.

[0037] For example, a user could be an employee in a department. When this user needs to handle the monthly work order statistics and distribution tasks, they can operate through a web-based or mobile application on an electronic device. The user can enter a natural language command in the input box, such as, "Query the completion status of all work orders in this department this month, filter overdue work orders, generate a completion rate statistical report, distribute overdue work orders to the pending tasks according to the responsible person, send the report to the department head, and synchronize it to the OA system for archiving." The electronic device receives this natural language command and recognizes it as a received data processing instruction.

[0038] Subsequently, the electronic device inputs the data processing instruction into the data processing model. Utilizing the logical reasoning capabilities of the large language model, the instruction is parsed, and a processing flow is generated and output. This processing flow specifically includes the following stages: Data Query: Identify the query interface that needs to be called from the work order system, and include the filter parameters "this month" and "this department"; Statistical calculation: Based on the filtering parameters of "this month" and "this department", retrieve all work orders for this department this month from the database; Conditional filtering: Construct a conditional judgment node to compare the "creation time" and "current time" of the work order and identify target work orders that exceed the specified time limit; Report generation: Generate reports based on the selected target work orders.

[0039] Task distribution: Send the generated reports to some responsible persons.

[0040] Multi-system synchronization: Identify the file upload interface that needs to be called in the OA system, and save the generated report to the OA system through the file upload interface.

[0041] In this embodiment, the data processing model can directly parse the requirements carried in the data processing instructions and generate corresponding processing flows, matching logical database rules. This process does not generate code, compile, or require a programming environment; it directly converts natural language into executable instructions, completely bypassing the code generation and compilation stages, resulting in lower computing power, faster speed, and a lower barrier to entry. The generated processing flow contains multiple layers, enabling lightweight computation, rule validation, and interface calls to be completed through layered logical processing, automatically connecting to work order systems, OA systems, reporting platforms, and to-do systems.

[0042] In this embodiment, the processing flow provides start, execution steps, condition judgment, and end standard nodes, supports natural language filling, automatically detects conflicts, infinite loops, and missing parameters, and forms a directly executable visual process. It achieves full visibility, easy understanding, easy modification, and easy debugging, greatly reducing the usage threshold and error rate.

[0043] S102: Display the processing flow. If a confirmation instruction for the processing flow is received, control the data processing model to respond to the data processing instruction based on the processing flow and obtain the data processing result output by the data processing model.

[0044] In this embodiment of the application, after the electronic device receives the processing flow sent by the data processing model, the electronic device will display the processing flow so that the user can confirm or modify the processing flow.

[0045] Specifically, on the electronic device's display interface, a visual workflow panel automatically pops up below the input data processing command to display the processing flow. This flow can be presented as a flowchart, composed of standard graphic symbols, such as rounded rectangles representing "start / end," rectangles representing "execution steps," and diamonds representing "conditional judgments." Different nodes are connected by arrows, clearly showing the data flow. For example, the user can see the flow from "querying a work order" to "determining if it has timed out," then branching into "yes" and "no" paths. The "yes" path flows to "generating a to-do task," while the "no" path ends directly, finally converging on the "generate a report" node.

[0046] In this embodiment of the application, the content of each node in the flowchart of the processing flow is filled in natural language. For example, in the "condition judgment" node, it is displayed that "work order processing time > specified time". This display method allows employees without any programming background to understand the execution logic of the system at a glance, which greatly reduces the understanding threshold.

[0047] Furthermore, after displaying the processing flow, the electronic device will pop up a confirmation dialog box on the display interface to prompt the user whether they need to modify the processing flow. For example, the content of this confirmation dialog box could be: "The system has generated the following processing flow for you. Please confirm if it is correct." At this point, the user has complete control. If the user believes the flow is correct, they can enter a confirmation command through the confirmation dialog box, such as clicking the "Confirm Execution" button; if they find an error in the logic, the user can intervene and modify it at this time, such as clicking the "Modify" button in the confirmation dialog box.

[0048] In this embodiment, the user carefully checks the flowchart and, after confirming its accuracy, inputs a confirmation command into the electronic device. Upon receiving the confirmation command, the electronic device controls the data processing model to respond to the data processing command. The data processing model does not perform any code compilation; instead, it directly responds according to the order of each node in the processing flow, based on the content written at each node.

[0049] For example, if the data processing instruction is: query the completion status of all work orders in this department this month, filter overdue work orders, generate a completion rate statistical report, distribute overdue work orders to the pending tasks by responsible person, send the report to the department head, and synchronize it to the OA system for archiving; and the confirmed processing flow is: data query, statistical calculation, conditional filtering, report generation, task distribution, and multi-system synchronization, then the data processing model's processing procedure is as follows: Through the standard API interface of the BSS / OSS system, all original work order data for this month in this department is retrieved. This process does not require users to manually log in to the system to export Excel, achieving automated data acquisition; after obtaining all the original data, the data processing model performs lightweight calculations in memory, traversing each work order record, calculating the difference between the current time and the work order creation time, marking records with a difference exceeding 24 hours as "overdue," and extracting key information such as work order ID, responsible person, and title; the data processing model then transmits the filtered list of overdue work orders through the RESTful API of the pending tasks system. The API interface allows for the batch creation of new pending tasks; the data processing model automatically sets the task recipient to the responsible person recorded in the work order and sets the task title to "Please process overdue work orders: [Work Order Title]"; the data processing model uses the built-in report engine to fill the statistical results into a preset report template and generate a standard PDF or Excel file; the data processing model calls the OA system's file upload interface to upload the file to the specified archive directory and triggers a notification message to send to the department head, informing them that "This month's work order statistical report has been generated, please check it".

[0050] In this embodiment, the data processing model also sends the data processing results to an electronic device. After receiving the data processing results, the electronic device can display them on a display interface. These data processing results include, but are not limited to, the final execution status and key result data. The execution status is either successful or failed, and the key result data is the final determined data. For example, the data processing results could be: Execution successful; 5 overdue work orders have been selected and distributed to the relevant responsible persons; the report has been generated and archived, and the department head has received notification.

[0051] Furthermore, in this embodiment, users can also adjust rules such as timeout thresholds, statistical definitions, and recipients by sending modification commands. These modification commands can target the displayed processing flow or modify content stored in the logical database, taking effect immediately without compilation or publication.

[0052] The core of this application's embodiments lies in zero programming and direct execution. Compared with related technologies, this application's embodiments skip the code generation and compilation process, directly realizing the processing flow from the logical parsing of data processing instructions. Furthermore, by displaying the processing flow, users can understand the entire processing logic. This not only significantly reduces computing power consumption, but more importantly, eliminates the gap between ordinary users and complex IT systems, enabling users to directly drive data processing models through natural language, achieving truly low-threshold, high-efficiency business automation.

[0053] Example 2: In order to guide large models to generate more accurate logic and avoid logical deviations caused by model illusions, based on the above embodiments, in this embodiment, if the data processing instruction carries first constraint information, sending the data processing instruction to the data processing model for parsing includes: The first constraint information is sent to the data processing model for parsing; The first constraint information contains user requirements, data processing logic, and data information to be invoked.

[0054] In this embodiment, when inputting data processing instructions, the user can also proactively provide first constraint information. This first constraint information can include the background of the requirement, details of the processing logic, and additional explanations of a specific data source. Data processing instructions carrying first constraint information differ from simple one-line instructions. These instructions contain a more structured context, which can guide large models to generate more accurate logic and avoid logical deviations caused by model illusions.

[0055] In real-world applications, business logic is often very complex and rigorous, such as configuring tariff packages and settling cross-provincial transactions. In these scenarios, a vague instruction alone is insufficient to complete the task; specific constraints must be provided.

[0056] The scenario is as follows: A client manager in a department needs to handle a complex client requirement breakdown task. This requirement involves a combination of multiple product lines, specific discount rules, and the need to reference specific historical project data. The client manager can enter a data processing command containing detailed first constraint information through the input interface of an electronic device.

[0057] The first constraint information includes, but is not limited to, user requirements, data processing logic, and data information to be invoked. This first constraint information may also include the components, devices, and systems that need to be invoked.

[0058] After receiving a data instruction carrying the first constraint information, the electronic device performs structured parsing of the data processing instruction. Optionally, the electronic device can quickly identify the first constraint information contained in the data instruction using regular expressions or semantic recognition technology.

[0059] The electronic device can send the first constraint information to the data processing model, at which point the data processing model's parsing process has a clear focus. The data processing model can first parse the user requirements carried in the first constraint information, and then parse the data processing logic and the data information to be invoked carried in the first constraint information. Based on the user requirements, the data processing logic, and the data information to be invoked, the data processing model can generate a processing flow. This processing flow can be used to implement the user requirements, satisfy the data processing logic, and utilize the data information to be invoked.

[0060] The embodiments of this application greatly enhance the flexibility and accuracy of the generated processing logic. In the embodiments of this application, by introducing first constraint information, users can directly inject complex business rules, specific data source references, and strict restrictions into the data processing instructions. This makes the data processing model no longer generate logic out of thin air, but rather reason based on a clear business context. This not only avoids the illusion of a data processing model, but also enables the data processing model to handle enterprise-level core businesses that have extremely high requirements for data sources and logical rigor.

[0061] In order to guide large models to generate more accurate logic and avoid logical deviations caused by model illusions, based on the above embodiments, in this embodiment of the application, if the data processing instruction does not carry first constraint information, before sending the data processing instruction to the data processing model for parsing, the method further includes: Obtain the target intent carried in the data processing instructions; In a pre-configured logical database, search for the second constraint information corresponding to the target intent; The step of sending the data processing instructions to the data processing model for parsing includes: The second constraint information and the data processing instructions are sent to the data processing model for parsing.

[0062] In practical applications, users often cannot describe all business rules in detail for every operation. Therefore, in this embodiment, when the user-input data processing command does not carry explicit first constraint information, the electronic device can retrieve second constraint information by searching a pre-configured logical database, thereby assisting the data processing model in completing logical parsing.

[0063] Specifically, the electronic device can obtain the target intent carried in the data processing instruction and search for the second constraint information corresponding to the target intent in a pre-configured logical database. In addition, the logical database can also be configured with standard logical units such as OA process rules, data permissions, statistical definitions, approval conditions, interface configurations, and task distribution strategies.

[0064] Electronic devices can determine the target intent based on the following possible methods: Method 1: If the data processing instruction carries an intent identifier, the electronic device determines that the intent corresponding to the intent identifier is the target intent carried in the data processing instruction.

[0065] The intent identifier can be the name of the intent, or it can be a number, sequence number, or other characters that can uniquely represent an intent.

[0066] Method 2: If the data processing instruction does not carry an intent identifier, the electronic device can call the intent recognition model to perform intent recognition on the data processing instruction and determine the target intent carried in the data processing instruction.

[0067] For example, electronic devices can perform deep semantic analysis on data processing instructions, and then use the intent recognition capability of the data processing model or a trained intent recognition model to strip away the entity information carried in the data processing instructions, thereby extracting the target intent corresponding to the data processing instructions.

[0068] In this embodiment, the electronic device is pre-configured with a logical database. This logical database is not an ordinary storage repository, but a business rule repository pre-configured by IT experts or administrators. The logical database stores tens of thousands of standardized logical units, each corresponding to an intent and containing the constraint information corresponding to that intent.

[0069] Electronic devices can perform fuzzy matching and precise retrieval in a logical database based on the target intent carried in the data processing instructions to find the constraint information corresponding to the target intent. If the electronic device does not find the target intent in the logical database, it can calculate the similarity between the target intent and each intent stored in the logical database, and determine the other intent with the highest similarity to the target intent. The electronic device then determines the constraint information corresponding to the other intent stored in the logical database as the second constraint information corresponding to the target intent.

[0070] Subsequently, the electronic device can send the retrieved second constraint information along with the received data processing instructions to the data processing model for parsing. Optionally, the electronic device can directly send the second constraint information and data processing instructions to the data processing model. Alternatively, the electronic device can fuse the second constraint information and data processing instructions and then send the fused data to the data processing model.

[0071] This application implements a process for determining constraint information based on intent, solving the problems of knowledge forgetting and rule inconsistency faced by data processing models when handling enterprise-level business. This application configures the constraint rules corresponding to common intents in a logical database. Using simple intents input by users, electronic devices can automatically complete missing constraint information. This not only significantly reduces the user's input cost, but more importantly, ensures the standardization of enterprise business execution. Regardless of the user's operation, as long as the intent is the same, the same set of verified constraint information is invoked from the logical database, avoiding operational errors caused by different individual interpretations.

[0072] In this embodiment, a dedicated logical database is built to uniformly store business rules, condition judgments, calculation logic, process templates, and interface parameters, providing version control, conflict detection, reuse, and atomic update capabilities. This solves the problems of scattered logic, poor reusability, and difficult maintenance in existing technologies, and achieves centralized rule management, stability, reliability, and consistent execution.

[0073] Example 3: To achieve dynamic and visualized process adjustments, based on the above embodiments, in this embodiment, after demonstrating the processing flow, the method further includes: If a modification instruction for the processing flow is received, the content of the sub-process in the processing flow is modified according to the information of the sub-process to be modified and the content to be modified carried in the modification instruction. The modified processing flow is shown.

[0074] In actual business operations, the processing flow generated by the data processing model may not fully meet the user's expectations, or the generated processing flow may become unsuitable for the changed business environment due to temporary changes. Therefore, in this embodiment, when the electronic device displays the processing on the display interface, the user can input modification commands through an external device or the display interface. The external device can be a mouse, keyboard, etc.

[0075] For example, a user is configuring an automated task for a "package promotion". Their input data processing instruction is: push 5G upgrade coupons to all users who have been with the network for more than 2 years. The electronic device calls the processing flow generated by the data processing model: filter users, generate coupons, and push messages.

[0076] However, when users reviewed the business process, they discovered that the default filtering condition for the user screening step was a network duration of >24 months. But they suddenly remembered the company's latest policy that this activity was only for non-5G plan users. If the current process was followed, it would result in repeated pushes to the group of people who are already 5G users, causing a waste of marketing resources.

[0077] Based on this, the user did not close the window and re-enter the data processing command, but instead directly entered the modification command on the displayed interface, thus modifying the displayed processing flow. For example, the user can click on the node corresponding to the user filtering step in the processing flow. After receiving this click, the electronic device pops up a parameter configuration panel on the display interface. The user enters the content to be modified through this parameter configuration panel, and the electronic device modifies the processing flow based on this content. For example, the content to be modified could be: the user's current plan is not a 5G plan.

[0078] In this embodiment of the application, the modification instruction received by the electronic device carries at least information about the sub-process to be modified and the content to be modified. The sub-process can be one or more nodes in the processing flow, and the content to be modified can be content to be replaced or content to be added.

[0079] Specifically, after receiving a modification instruction, the electronic device first parses the instruction to obtain information about the sub-process to be modified and the content to be modified. Based on this information, the electronic device then modifies the content within the sub-process.

[0080] The information for the subprocess to be modified can be the name of the subprocess, its position in the processing flow, or its order in the processing flow, or other information that can be used to locate the subprocess in the processing flow.

[0081] The following describes several possible ways in which electronic devices can modify data flows based on modification commands: Method 1: If the content to be modified in the modification instruction is the content to be replaced, the electronic device locates the sub-process to be modified in the processing flow according to the information of the sub-process to be modified carried in the modification instruction, and replaces the content recorded in the sub-process with the content to be modified.

[0082] Method 2: If the content to be modified in the modification instruction is content to be added, the electronic device locates the sub-process to be modified in the processing flow according to the information of the sub-process to be modified carried in the modification instruction, and adds the content to be modified to the content recorded in the sub-process.

[0083] Method 3: If the content to be modified in the modification instruction is content to be deleted, the electronic device locates the sub-process to be modified in the processing flow according to the information of the sub-process to be modified in the modification instruction, and deletes the content to be modified from the content recorded in the sub-process.

[0084] It should be noted that modification instructions can include a modification type, which can be one of replacement, addition, or deletion. The electronic device can determine the appropriate modification operation based on the modification type specified in the modification instruction. If the modification instruction does not specify a modification type, the electronic device can determine the modification type based on the content to be modified in the modification instruction and the original content in the sub-process to be modified.

[0085] Specifically, the electronic device locates the sub-process to be modified within the processing flow based on the information of the sub-process to be modified carried in the modification instruction, and obtains the original content written within the corresponding node of the sub-process in the processing flow. The electronic device compares the original content with the content to be modified carried in the modification instruction, and determines the modification type based on the comparison result. For example, if the original content and the content to be modified are completely different, the electronic device determines the modification type as addition; if the content to be modified is a subset of the original content, the electronic device determines the modification type as deletion; if the original content is a subset of the content to be modified, the electronic device determines the modification type as replacement.

[0086] In this embodiment, after the electronic device modifies the processing flow according to the modification instruction, it does not directly control the data processing model to execute the modified processing flow. Instead, it displays the modified processing flow again for the user to confirm. If the user determines that the modified processing flow is correct, they can input a confirmation instruction. The electronic device can then control the processing model to respond to the data processing instruction based on the modified processing flow. The specific response process is the same as in the above embodiments and will not be repeated here. If the user believes that the modified processing flow still has errors, they can continue to input modification instructions until the modified data processing instruction is correct. The operation of the electronic device after receiving the modification instruction is the same as in the above embodiments and will not be repeated here.

[0087] Furthermore, in this embodiment, to improve data processing efficiency, the electronic device can be pre-configured with a modification count threshold. For a given processing flow, if the number of modification instructions received by the electronic device reaches the threshold, the data processing instruction corresponding to that flow can be considered inaccurate. The electronic device can then terminate its response to the data processing instruction and output a failure message. Additionally, this message can also include a notification that the data processing instruction is inaccurate, allowing the user to modify the instruction accordingly.

[0088] Furthermore, in this embodiment, if multiple modifications are involved, the content of each modification can be identified in the processing flow using different formats. These formats can include font, font color, highlight color, etc.

[0089] Compared with related technologies, the embodiments of this application have the following significant advantages: Zero compilation iteration: Users do not need to understand code. They can simply modify the node parameters of the processing flow on the display interface of the electronic device like building blocks, and the electronic device will take effect immediately without recompiling or re-deploying.

[0090] Full traceability: Every modification is reflected in the visual processing flow, allowing users to clearly see the changes in logic and avoiding errors or omissions that may occur when modifying code text.

[0091] Human-machine collaboration: This application's embodiments perfectly combine the data processing model's generation capabilities with human judgment. The data processing model is responsible for generating the initial draft of the processing flow and executing the processing flow, while humans are responsible for logically calibrating the processing flow generated by the data processing model. This collaborative model greatly improves the accuracy and flexibility of complex business processing.

[0092] Example 4: To simplify the data processing workflow, reduce costs, and enable natural language processing, structured data storage, and reuse of complex business requirements, based on the above embodiments, in this application embodiment, after obtaining the data processing results output by the data processing model, the method further includes: The processing flow and the target intent carried in the data processing instructions are saved to the logical database.

[0093] In practical applications, the constraints imposed by different intentions are often subject to frequent changes. For example, the pricing policies of government and enterprise clients, the approval processes of internal employees, or the troubleshooting procedures for network failures may all change with adjustments to the market environment or company strategies. In IT systems using related technologies, once a user completes a complex business process through interaction with a large language model using natural language, this process is usually only a one-time execution record. If another user needs to handle the same business the next time, the large language model may generate a slightly different process, resulting in inconsistent business execution standards and requiring re-debugging each time, leading to extremely low efficiency.

[0094] Based on this, the logical database constructed in this application embodiment can evolve dynamically. Specifically, when the electronic device detects that a data processing instruction has been successfully executed, the electronic device can save the target intent corresponding to the data processing instruction and the processing flow corresponding to the data processing instruction into the logical database.

[0095] For example, suppose the scenario is as follows: A user is handling a new customer request. Through the display interface of an electronic device, they input a data processing instruction: design a "high-speed cloud network" activation process for a newly established technology park company, requiring a fiber optic cable route survey first, resource pre-allocation only after the survey is approved, and finally, the generation of a formal order, skipping the conventional initial review stage. The electronic device sends this data processing instruction to a data processing model, which parses the instruction and generates a processing flow including route survey, resource pre-allocation, and order generation. The user then fine-tunes this processing flow through the electronic device's display interface, including adding a "fiber optic signal test" node. The electronic device responds to the data processing instruction based on the adjusted processing flow, and if the execution result is successful, the electronic device can save the target intent corresponding to the data processing instruction and the adjusted processing flow to a logical database.

[0096] In one possible approach, the logical database may already contain the historical processing flow corresponding to the target intent. When the electronic device stores the processing flow corresponding to the target intent in the logical database, it can disregard this historical processing flow and directly write the processing flow corresponding to the target intent into a blank field in the logical database.

[0097] Furthermore, in this embodiment, the electronic device can also receive a modification instruction for certain information stored in the logical database. This information may be a processing flow corresponding to an intent, constraint information corresponding to an intent, etc. After receiving the modification instruction, the electronic device can modify the corresponding information stored in the logical database based on the content to be modified carried in the modification instruction. After modifying the flowchart or logical database, there is no need for transcoding, compilation, or deployment; the changes take effect immediately and return to the execution stage, reducing iteration costs and processing cycles, and enabling faster response to business changes.

[0098] To improve the efficiency of the data processing flow, based on the above embodiments, in this embodiment, before sending the data processing instruction to the data processing model for parsing, the method further includes: Determine whether the logical database contains candidate processing flows corresponding to the target intent; If the logical database does not store the candidate processing flow corresponding to the target intent, then the subsequent step of sending the data processing instruction to the data processing model for parsing is executed; If the logical database stores candidate processing flows corresponding to the target intent, then the candidate processing flow is determined as the processing flow, and the subsequent steps of displaying the processing flow are executed.

[0099] In this embodiment, the logical database can store intents, corresponding constraint information, and processing flows. Based on this, when an electronic device receives a data processing request, it can first parse the request to determine the target intent, and then directly search for the processing flow in the logical database based on that target intent. This eliminates the need for parsing the data processing model, further improving data processing efficiency, avoiding wasted computing power, and balancing speed and accuracy. The overall efficiency, stability, and adaptability of data processing are significantly improved.

[0100] The process by which the electronic device determines the target intent corresponding to the data processing request is the same as the process of determining the target intent in the above embodiments, and will not be described again here.

[0101] Specifically, the electronic device determines whether the logical database contains candidate processing flows corresponding to the target intent. If the electronic device determines that the logical database contains candidate processing flows corresponding to the target intent, it skips the step of generating the processing flow corresponding to the data processing instruction based on the data processing model and directly displays the candidate processing flow corresponding to the target intent in the logical database on the display interface. If the electronic device determines that the logical database does not contain candidate processing flows corresponding to the target intent, it executes the step of generating the processing flow corresponding to the data processing instruction based on the data processing model.

[0102] In this embodiment, the electronic device can query a logical database according to the target intent corresponding to the data processing instruction. If the electronic device finds a processing flow corresponding to the target intent in the logical database, it is determined that the logical database stores a candidate processing flow corresponding to the target intent; if the electronic device does not find a processing flow corresponding to the target intent in the logical database, it is determined that the logical database does not store a candidate processing flow corresponding to the target intent.

[0103] It should be noted that the logical database may contain two or more processing flows for a single intent. If the electronic device finds two or more candidate processing flows corresponding to the target intent in the logical database, the electronic device can simultaneously display these two or more candidate processing flows on the display interface, and then the user can select which candidate processing flow to use as the processing flow for the target intent.

[0104] Users can input selection commands via external devices or a display interface. These commands specify candidate processing flows for the target intent. For example, if a user selects a candidate processing flow with the mouse, the electronic device will receive this operation and use the selected candidate processing flow as the processing flow for the target intent.

[0105] Example 5: To avoid version conflicts during the iteration process of the logical database and to ensure that the information stored in the logical database is always the latest and most accurate, based on the above embodiments, in this embodiment of the application, before saving the processing flow and the target intent carried in the data processing instruction to the logical database, the method further includes: Determine whether the logical database contains the initial processing flow corresponding to the target intent; If the logical database does not contain the initial processing flow corresponding to the target intent, then the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed. If the logical database contains an initial processing flow corresponding to the target intent, then the initial processing flow stored in the logical database is deleted, and the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed.

[0106] In order to better manage the logical database and avoid version conflicts, in this embodiment of the application, for any intention, the logical database stores at most one processing flow corresponding to that intention.

[0107] Based on this, in the embodiments of this application, before the electronic device saves the processing flow and the target intent carried in the data processing instruction to the logical database, the electronic device will first determine whether the initial processing flow corresponding to the target intent is stored in the logical database. If it exists, the electronic device will use the processing flow to replace the initial processing flow. If it does not exist, the electronic device will directly save the processing flow in the logical database.

[0108] Specifically, if the initial processing flow corresponding to the target intent is not stored in the logical database, the electronic device will perform the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database; if the initial processing flow corresponding to the target intent is stored in the logical database, the electronic device will delete the initial processing flow stored in the logical database and perform the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database.

[0109] In this embodiment, the electronic device uses a logical database as its underlying support and a processing flow as its visualization carrier. It integrates dual input and dual output, layered processing, and real-time iteration to form a closed loop of the entire process. It also supports version control and atomic updates, thus constructing a complete and self-consistent technical system. This achieves a closed loop of logic construction, execution, debugging, and iteration, improving its practicality and scalability.

[0110] Compared with related technologies, the overall system constructed in this application's embodiments features a closed-loop process for the construction, execution, debugging, and iteration of the tourism bureau's timeline. The various templates coordinate and cooperate, ensuring both the integrity of the technical solution and improving overall operational efficiency, thus possessing stronger feasibility and promotion potential.

[0111] Furthermore, users can use natural language to organize, break down, and structurally describe complex business requirements; electronic devices automatically parse, extract, and summarize them into standardized logical units, i.e., processing flows, and automatically store them in the logical database; when executing similar business in the future, the accumulated logic can be directly called without repeated organization and description, realizing the one-time organization, permanent accumulation, multiple reuse, and rapid execution of complex business, greatly improving efficiency, reducing repetitive workload, and strengthening business standardization and consistency.

[0112] Example 6: In order to meet the needs of manual operation and system API integration, based on the above embodiments, in this application embodiment, the data processing instructions are in API instruction format or natural language format; After obtaining the data processing result output by the data processing model, the method further includes: If the data processing instruction is in API instruction format, the data processing result is converted into API structured data, and the API structured data is sent to the sending device of the data processing instruction. If the data processing instruction is in natural language format, the data processing result will be displayed in the input window corresponding to the data processing instruction.

[0113] To address the issue that related technologies often employ a single input / output mode, failing to simultaneously meet the needs of manual operation and system API integration, this application proposes a dual-input, dual-output core architecture. Input supports natural language dialogue and API requests; output supports interface display and structured API data. This dual-channel parallel processing simultaneously satisfies both manual operation and system integration, covering all scenarios from personal office work to enterprise collaboration and third-party integration.

[0114] In the digital architecture of telecommunications companies, there are two distinct types of users: ordinary employees who can input data processing commands through electronic device displays; and other automated systems or third-party applications that require structured data (such as JSON / XML) for subsequent processing. Traditional automation tools often can only handle one of these tasks. For example, RPA tools typically only simulate human interfaces and cannot provide APIs; while traditional APIs cannot be directly used by employees who do not understand coding. This results in companies needing to maintain two parallel systems, leading to high costs and difficulties in data integration.

[0115] This application demonstrates how a unified underlying engine can dynamically switch output modes based on the format of data processing requests, thereby meeting the needs of manual operation and system API integration.

[0116] The following describes the relationship between several possible application scenarios and the format of data processing instructions: Scenario A: Data processing instructions in natural language format: When an employee enters "Check my to-do list for today" on the web interface of an electronic device, the electronic device recognizes that the request originates from the web front-end and that the content is natural language text. The electronic device then determines that the received data processing instructions are in natural language format.

[0117] Scenario B: Data processing instructions in API instruction format: When the company's OA system needs to retrieve automated data, it initiates an HTTP POST request. If the electronic device recognizes that the HTTP POST request originates from a server within the IP whitelist and is in standard API format, then the electronic device will determine that the received data processing instructions are in API instruction format.

[0118] After acquiring the data processing results, the electronic device will output the data processing results in different formats according to the format corresponding to the data processing instructions.

[0119] Specifically, if the data processing instruction is in API instruction format, the data processing result is converted into API structured data and sent to the sending device of the data processing instruction; if the data processing instruction is in natural language format, the data processing result is displayed in the input window corresponding to the data processing instruction.

[0120] For example, if the data processing instruction is in API instruction format, the electronic device can call a visualization renderer to convert the data processing result into HTML / CSS code, building a visually appealing front-end display page. Alternatively, the electronic device can call a structured data serializer to clean and standardize the data processing result, removing all HTML tags, styles, and natural language descriptions, retaining only the core data fields, and then converting the remaining core data fields into standard JSON objects or XML-formatted API structured data. The electronic device can then send this API structured data directly to the callback address of the data processing request sender via HTTPS.

[0121] This application's embodiments implement dual input and dual output for manual operation and system API integration. Humans can process data via chat, while the IT system can process data through standard API access. Both share the same underlying business logic, eliminating the need for redundant development. Furthermore, users see an intuitive and easy-to-understand interface; developers receive standardized and regulated data.

[0122] It should be noted that the embodiments of this application can be widely applied to all telecommunications service scenarios and intelligent upgrades in various industries, focusing on: telecommunications operation decision-making: business rule verification, process approval, demand breakdown, and tariff package configuration; telecommunications front-line services: customer service work orders, fault diagnosis, service activation, and rapid demand response; internal complex business processing: employees can automate the entire process of cross-system data query, statistics, approval, distribution, archiving, and notification through the Xingchen Big Data Model intelligent agent with zero programming; system integration and docking: seamless connection with BSS / OSS / MSS, OA, work order, report, and to-do system APIs.

[0123] Furthermore, the data processing method provided in this application embodiment can be deployed on electronic devices such as local devices, cloud devices, edge devices, and telecom cloud platforms.

[0124] Figure 2 The data processing flowchart provided in this application embodiment is as follows: Figure 2As shown, the process includes: S201: Construct the underlying support system.

[0125] This underlying support system includes, but is not limited to, a logical database, and may also include built-in modules configured in electronic devices, lightweight algorithms, and trained data processing models.

[0126] S202: Receive input data processing instructions.

[0127] The data processing instruction can be input in the form of natural language dialogue or in the form of API request.

[0128] S203: Forward the data processing instructions to the data processing model for parsing, generating and displaying the processing flow.

[0129] When the data processing model parses the data processing instructions, it does not transcode, compile, or require the deployment of a programming environment.

[0130] This processing flow has a hierarchical structure and is suitable for routine calculations, information processing, or complex needs.

[0131] S204: The data processing model is based on the processing flow and responds to data processing instructions.

[0132] If the electronic device receives a confirmation instruction for the processing flow, it invokes the data processing model to respond to the data processing instruction based on the processing flow.

[0133] If the electronic device receives a modification instruction for the processing flow, it modifies the processing flow based on the modification instruction and displays the modified processing flow again until it receives a confirmation instruction for the processing flow. Then, it calls the data processing model to respond to the data processing instruction based on the final modified processing flow.

[0134] S205: Obtain the data processing results and output them in the set format.

[0135] If the data processing instruction is in natural language format, the data processing result is displayed on the electronic device's display interface; if the data processing instruction is in API instruction format, the data processing result is converted into API structured data and sent to the device that sent the data processing instruction.

[0136] Example 7: Based on the above embodiments, this application also provides a data processing device based on a large language model. Figure 3 A schematic diagram of a data processing device based on a large language model is provided in this application embodiment. The device includes: The receiving module 301 is used to receive input data processing instructions; The logic parsing module 302 is used to send the data processing instruction to the data processing model for parsing, and to obtain the processing flow output by the data processing model when responding to the data processing instruction; The processing module 303 is used to display the processing flow. If a confirmation instruction for the processing flow is received, it controls the data processing model to respond to the data processing instruction based on the processing flow and obtains the data processing result output by the data processing model.

[0137] In one possible implementation, the logic parsing module 302 is specifically used to send the first constraint information to the data processing model for parsing if the data processing instruction carries first constraint information; wherein, the first constraint information carries user requirements, data processing logic, and data information to be called.

[0138] In one possible implementation, the logic parsing module 302 is further configured to: if the data processing instruction does not carry the first constraint information, obtain the target intent carried in the data processing instruction; search for the second constraint information corresponding to the target intent in a pre-configured logic database; and send the second constraint information and the data processing instruction to the data processing model for parsing.

[0139] In one possible implementation, the processing module 303 is further configured to, if a modification instruction for the processing flow is received, modify the content of the sub-process in the processing flow according to the information of the sub-process to be modified and the content to be modified carried in the modification instruction; and display the modified processing flow.

[0140] In one possible implementation, the processing module 303 is further configured to save the processing flow and the target intent carried in the data processing instruction into the logical database.

[0141] In one possible implementation, the logic parsing module 302 is further configured to determine whether the logic database contains a candidate processing flow corresponding to the target intent; if the logic database does not contain a candidate processing flow corresponding to the target intent, then the subsequent step of sending the data processing instruction to the data processing model for parsing is executed; if the logic database contains a candidate processing flow corresponding to the target intent, then the candidate processing flow is determined as the processing flow, and the subsequent step of displaying the processing flow is executed.

[0142] In one possible implementation, the processing module 303 is further configured to determine whether the initial processing flow corresponding to the target intent is stored in the logical database; if the initial processing flow corresponding to the target intent is not stored in the logical database, then the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed; if the initial processing flow corresponding to the target intent is stored in the logical database, then the initial processing flow stored in the logical database is deleted, and the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed.

[0143] In one possible implementation, the data processing instructions are in API instruction format or natural language format; The processing module 303 is further configured to convert the data processing result into API structured data and send the API structured data to the sending device of the data processing instruction if the data processing instruction is in API instruction format; and to display the data processing result in the input window corresponding to the data processing instruction if the data processing instruction is in natural language format.

[0144] Example 8: Based on the above embodiments, this application also provides an electronic device. Figure 4 This application provides a schematic diagram of an electronic device structure, such as... Figure 4 As shown, it includes: processor 401, communication interface 402, memory 403 and communication bus 404, wherein processor 401, communication interface 402 and memory 403 communicate with each other through communication bus 404. The memory 403 stores a computer program. When the program is executed by the processor 401, the processor 401 performs the following steps: Receive input data processing instructions; The data processing instruction is sent to the data processing model for parsing, and the processing flow output by the data processing model in response to the data processing instruction is obtained. The processing flow is demonstrated. If a confirmation instruction for the processing flow is received, the data processing model is controlled to respond to the data processing instruction based on the processing flow, and the data processing result output by the data processing model is obtained.

[0145] In one possible implementation, if the data processing instruction carries first constraint information, sending the data processing instruction to the data processing model for parsing includes: The first constraint information is sent to the data processing model for parsing; The first constraint information contains user requirements, data processing logic, and data information to be invoked.

[0146] In one possible implementation, if the data processing instruction does not carry the first constraint information, before sending the data processing instruction to the data processing model for parsing, the method further includes: Obtain the target intent carried in the data processing instructions; In a pre-configured logical database, search for the second constraint information corresponding to the target intent; The step of sending the data processing instructions to the data processing model for parsing includes: The second constraint information and the data processing instructions are sent to the data processing model for parsing.

[0147] In one possible implementation, after demonstrating the processing flow, the method further includes: If a modification instruction for the processing flow is received, the content of the sub-process in the processing flow is modified according to the information of the sub-process to be modified and the content to be modified carried in the modification instruction. The modified processing flow is shown.

[0148] In one possible implementation, after obtaining the data processing result output by the data processing model, the method further includes: The processing flow and the target intent carried in the data processing instructions are saved to the logical database.

[0149] In one possible implementation, before sending the data processing instructions to the data processing model for parsing, the method further includes: Determine whether the logical database contains candidate processing flows corresponding to the target intent; If the logical database does not store the candidate processing flow corresponding to the target intent, then the subsequent step of sending the data processing instruction to the data processing model for parsing is executed; If the logical database stores candidate processing flows corresponding to the target intent, then the candidate processing flow is determined as the processing flow, and the subsequent steps of displaying the processing flow are executed.

[0150] In one possible implementation, before saving the processing flow and the target intent carried in the data processing instructions to the logical database, the method further includes: Determine whether the logical database contains the initial processing flow corresponding to the target intent; If the logical database does not contain the initial processing flow corresponding to the target intent, then the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed. If the logical database contains an initial processing flow corresponding to the target intent, then the initial processing flow stored in the logical database is deleted, and the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed.

[0151] In one possible implementation, the data processing instructions are in API instruction format or natural language format; After obtaining the data processing result output by the data processing model, the method further includes: If the data processing instruction is in API instruction format, the data processing result is converted into API structured data, and the API structured data is sent to the sending device of the data processing instruction. If the data processing instruction is in natural language format, the data processing result will be displayed in the input window corresponding to the data processing instruction.

[0152] Since the principle of the above-mentioned electronic device in solving the problem is similar to that of the data processing method based on the large language model, the implementation of the above-mentioned electronic device can be found in the embodiments of the method, and repeated details will not be repeated.

[0153] The communication bus mentioned in the above-mentioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not indicate that there is only one bus or one type of bus. Communication interface 402 is used for communication between the above-mentioned electronic device and other devices. The memory can include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0154] The processors mentioned above can be general-purpose processors, including central processing units, network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0155] Example 9: Based on the above embodiments, this invention also provides a computer-readable storage medium storing a computer program executable by a processor. When the program runs on the processor, it causes the processor to perform the following steps: Receive input data processing instructions; The data processing instruction is sent to the data processing model for parsing, and the processing flow output by the data processing model in response to the data processing instruction is obtained. The processing flow is demonstrated. If a confirmation instruction for the processing flow is received, the data processing model is controlled to respond to the data processing instruction based on the processing flow, and the data processing result output by the data processing model is obtained.

[0156] In one possible implementation, if the data processing instruction carries first constraint information, sending the data processing instruction to the data processing model for parsing includes: The first constraint information is sent to the data processing model for parsing; The first constraint information contains user requirements, data processing logic, and data information to be invoked.

[0157] In one possible implementation, if the data processing instruction does not carry the first constraint information, before sending the data processing instruction to the data processing model for parsing, the method further includes: Obtain the target intent carried in the data processing instructions; In a pre-configured logical database, search for the second constraint information corresponding to the target intent; The step of sending the data processing instructions to the data processing model for parsing includes: The second constraint information and the data processing instructions are sent to the data processing model for parsing.

[0158] In one possible implementation, after demonstrating the processing flow, the method further includes: If a modification instruction for the processing flow is received, the content of the sub-process in the processing flow is modified according to the information of the sub-process to be modified and the content to be modified carried in the modification instruction. The modified processing flow is shown.

[0159] In one possible implementation, after obtaining the data processing result output by the data processing model, the method further includes: The processing flow and the target intent carried in the data processing instructions are saved to the logical database.

[0160] In one possible implementation, before sending the data processing instructions to the data processing model for parsing, the method further includes: Determine whether the logical database contains candidate processing flows corresponding to the target intent; If the logical database does not store the candidate processing flow corresponding to the target intent, then the subsequent step of sending the data processing instruction to the data processing model for parsing is executed; If the logical database stores candidate processing flows corresponding to the target intent, then the candidate processing flow is determined as the processing flow, and the subsequent steps of displaying the processing flow are executed.

[0161] In one possible implementation, before saving the processing flow and the target intent carried in the data processing instructions to the logical database, the method further includes: Determine whether the logical database contains the initial processing flow corresponding to the target intent; If the logical database does not contain the initial processing flow corresponding to the target intent, then the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed. If the logical database contains an initial processing flow corresponding to the target intent, then the initial processing flow stored in the logical database is deleted, and the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed.

[0162] In one possible implementation, the data processing instructions are in API instruction format or natural language format; After obtaining the data processing result output by the data processing model, the method further includes: If the data processing instruction is in API instruction format, the data processing result is converted into API structured data, and the API structured data is sent to the sending device of the data processing instruction. If the data processing instruction is in natural language format, the data processing result will be displayed in the input window corresponding to the data processing instruction.

[0163] Since the principle of the computer-readable storage medium in solving the problem is similar to that of the data processing method based on a large language model, the implementation of the computer-readable storage medium can be found in the embodiments of the method, and repeated details will not be described again.

[0164] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0165] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0166] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0167] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0168] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A data processing method based on a large language model, characterized in that, The method includes: Receive input data processing instructions; The data processing instruction is sent to the data processing model for parsing, and the processing flow output by the data processing model in response to the data processing instruction is obtained. The processing flow is demonstrated. If a confirmation instruction for the processing flow is received, the data processing model is controlled to respond to the data processing instruction based on the processing flow, and the data processing result output by the data processing model is obtained.

2. The method of claim 1, wherein, If the data processing instruction carries first constraint information, the step of sending the data processing instruction to the data processing model for parsing includes: The first constraint information is sent to the data processing model for parsing; The first constraint information contains user requirements, data processing logic, and data information to be invoked.

3. The method according to claim 1 or 2, characterized in that, If the data processing instruction does not carry the first constraint information, before sending the data processing instruction to the data processing model for parsing, the method further includes: Obtain the target intent carried in the data processing instructions; In a pre-configured logical database, search for the second constraint information corresponding to the target intent; The step of sending the data processing instructions to the data processing model for parsing includes: The second constraint information and the data processing instructions are sent to the data processing model for parsing.

4. The method of claim 1, wherein, After demonstrating the processing flow, the method further includes: If a modification instruction for the processing flow is received, the content of the sub-process in the processing flow is modified according to the information of the sub-process to be modified and the content to be modified carried in the modification instruction. The modified processing flow is shown.

5. The method of claim 3, wherein, After obtaining the data processing result output by the data processing model, the method further includes: The processing flow and the target intent carried in the data processing instructions are saved to the logical database.

6. The method of claim 5, wherein, Before sending the data processing instructions to the data processing model for parsing, the method further includes: Determine whether the logical database contains candidate processing flows corresponding to the target intent; If the logical database does not store the candidate processing flow corresponding to the target intent, then the subsequent step of sending the data processing instruction to the data processing model for parsing is executed; If the logical database stores candidate processing flows corresponding to the target intent, then the candidate processing flow is determined as the processing flow, and the subsequent steps of displaying the processing flow are executed.

7. The method of claim 5, wherein, Before saving the processing flow and the target intent carried in the data processing instruction to the logical database, the method further includes: Determine whether the logical database contains the initial processing flow corresponding to the target intent; If the logical database does not contain the initial processing flow corresponding to the target intent, then the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed. If the logical database contains an initial processing flow corresponding to the target intent, then the initial processing flow stored in the logical database is deleted, and the subsequent step of saving the processing flow and the target intent carried in the data processing instruction to the logical database is executed.

8. The method of claim 1, wherein, The data processing instructions are in API instruction format or natural language format; After obtaining the data processing result output by the data processing model, the method further includes: If the data processing instruction is in API instruction format, the data processing result is converted into API structured data, and the API structured data is sent to the sending device of the data processing instruction. If the data processing instruction is in natural language format, the data processing result will be displayed in the input window corresponding to the data processing instruction. 9.A data processing apparatus based on a large language model, characterized by, The device includes: The receiving module is used to receive input data processing instructions; The logic parsing module is used to send the data processing instructions to the data processing model for parsing, and to obtain the processing flow output by the data processing model when responding to the data processing instructions. The processing module is used to display the processing flow. If a confirmation instruction for the processing flow is received, it controls the data processing model to respond to the data processing instruction based on the processing flow and obtains the data processing result output by the data processing model.

10. An electronic device, comprising: The electronic device includes at least a processor and a memory, wherein the processor is used to execute a computer program stored in the memory to implement the steps of the data processing method based on a large language model as described in any one of claims 1-8.

Images