Automatic process triggering method and device based on large language model and medium
By processing multimodal data through a large language model, the system automatically identifies scenarios and matches processes, solving the problems of high barriers to entry and development costs associated with traditional process triggering methods, and achieving flexible and efficient process triggering.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-04-03
AI Technical Summary
Traditional process triggering methods rely on users mastering technical details, resulting in a high barrier to entry, difficulty in quickly responding to dynamic changes in business scenarios, and high development costs.
By employing a large language model for multimodal data processing, and through scene recognition, intent parsing, and process matching, the system automatically filters and initiates target process instances, reducing user interaction difficulty and development costs.
It enables rapid response to business needs across multiple domains without requiring users to know process names and parameter formats, reducing the development cost and cycle time for expanding into new scenarios.
Smart Images
Figure CN121785656A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of computer software and business process management technology, and in particular to a method, device and medium for automatic process triggering based on a large language model. Background Technology
[0002] Process automation technology refers to the use of structured process design to achieve automatic or semi-automatic flow of business processes, reducing the cost of manual intervention. Current traditional process triggering methods have significant limitations.
[0003] Traditional process triggering methods often rely on standardized form filling or application programming interface (API) calls. This approach requires users to have a clear understanding of the specific process name, the format specifications of required parameters, and the meaning of fields, resulting in a high barrier to entry and a difficult learning curve. Furthermore, the process triggering logic is strongly tied to specific business scenarios; triggering rules and parameter validation logic are designed for a single initial scenario. When business requirements expand to new scenarios, the technical team must redevelop scenario-adaptive code and adjust the triggering logic, increasing development costs and time, and making it difficult to quickly respond to dynamic changes in business scenarios. Summary of the Invention
[0004] This application provides a method, device, and medium for automatic process triggering based on a large language model, which solves the problem that traditional process triggering methods are unable to quickly respond to dynamic changes in business scenarios.
[0005] The embodiments of this application adopt the following technical solutions: On the one hand, embodiments of this application provide an automatic process triggering method based on a large language model, the method comprising: The user-input multimodal data is matched and calculated with each scene label in a predefined multi-scene label library to obtain scene recognition results and the confidence scores corresponding to the scene recognition results; The multimodal data is semantically parsed and information extracted using a large language model to obtain user intent and associated entities; Based on the scene recognition results, initial candidate processes are selected from a preset multi-scene process knowledge graph; Based on the matching degree between the user intent, the entity association, and the initial candidate process, a target process instance is determined; The target process instance is started by using preset parameter mapping rules and the entity association.
[0006] In one example, the multimodal data includes: Text data, audio data, image data, and mixed data.
[0007] In one example, the process of matching and calculating the multimodal data input by the user with each scene label in a predefined multi-scene label library to obtain the scene recognition result and the confidence level corresponding to the scene recognition result specifically includes: The multimodal data is standardized to obtain unified text data; Scene-related keywords are extracted from the unified text data, and multi-dimensional feature vectors are constructed by combining the user's role information and user historical interaction records. The matching confidence of the multi-dimensional feature vector with each scene tag in the multi-scene tag library is calculated using a predefined confidence algorithm to obtain the scene confidence corresponding to each scene tag. When the confidence level of the scene is greater than or equal to a preset threshold, the scene label is taken as the scene recognition result.
[0008] In one example, the method further includes: When the confidence level of the scenario is less than a preset threshold, a scenario confirmation query instruction is generated.
[0009] In one example, the step of selecting initial candidate processes from a preset multi-scenario process knowledge graph based on the scene recognition results specifically includes: The preset multi-scenario process knowledge graph includes process definitions, scenario tags, startup parameters, and business rules; All process nodes associated with the scene recognition results are selected from the preset multi-scenario process knowledge graph as initial candidate processes.
[0010] In one example, determining the target process instance based on the matching degree between the user intent, the entity association, and the initial candidate processes specifically includes: Extract the process definition description of each initial candidate process, and calculate the semantic similarity between the user intent and the process definition description using a semantic similarity algorithm to obtain the semantic matching degree corresponding to each initial candidate process; Extract the list of required startup parameters for each initial candidate process, and calculate the parameter matching degree by comparing the number of attribute entities in the associated entity with the total number of required startup parameters; Filter initial candidate processes that have a semantic matching degree greater than or equal to a preset semantic threshold and a parameter matching degree greater than or equal to a preset threshold to obtain a secondary candidate process set; When the set of secondary candidate processes includes at least two candidate processes, the comprehensive matching degree of each candidate process in the set of secondary candidate processes is calculated. The process with the highest overall matching degree is selected as the target process instance.
[0011] In one example, the method further includes: When the overall matching degree is similar, a scenario confirmation query instruction is generated.
[0012] In one example, starting the target process instance through preset parameter mapping rules and entity associations specifically includes: According to the preset parameter mapping rules, the entity association is converted into a parameter format that can be recognized by the target workflow instance; Call the target workflow instance creation interface to start the target workflow instance.
[0013] On the other hand, embodiments of this application provide an automatic process triggering device based on a large language model, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute any of the above-described automatic process triggering methods based on a large language model.
[0014] On the other hand, embodiments of this application provide a non-volatile computer storage medium for automatic process triggering based on a large language model, which stores computer-executable instructions that can execute any of the above-described automatic process triggering methods based on a large language model.
[0015] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: The method provided in this application, through multimodal data standardization and multi-dimensional scenario matching, eliminates the need for users to understand technical details such as process names and parameter formats, thus solving the high barrier to entry problem caused by standardized forms or API calls and significantly reducing the learning curve for non-technical personnel. By leveraging a large language model to automatically parse natural language semantics, extract user intent and related entities, it replaces the manual conversion of requirements into system instructions, further simplifying user interaction and improving the efficiency of requirement conversion.
[0016] Based on scene recognition results, initial candidate processes are automatically selected from a multi-scene process knowledge graph. Target process instances are determined through both semantic and parameter matching, eliminating the strong binding between processes and single scenes and eliminating the need for manual selection or redevelopment of selection logic. Adaptation between associated entities and target process instances is achieved through pre-defined parameter mapping rules, without the need to adjust parameter validation or triggering logic for new scenarios. This effectively reduces the development cost and cycle of expanding to new scenarios, enabling efficient and flexible triggering of processes across multiple domains. Attached Figure Description
[0017] To more clearly illustrate the technical solution of this application, some embodiments of this application will be described in detail below with reference to the accompanying drawings, in which: Figure 1 A flowchart illustrating an automatic process triggering method based on a large language model, provided in an embodiment of this application; Figure 2 A schematic diagram illustrating a cross-scenario intent parsing process provided in an embodiment of this application; Figure 3 A schematic diagram of intelligent process matching based on a large language model is provided for an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an automatic process triggering device based on a large language model, 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 of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] Some embodiments of this application will now be described in detail with reference to the accompanying drawings.
[0020] Figure 1 This is a flowchart illustrating an automatic process triggering method based on a large language model, provided as an embodiment of this application. This method can be applied to different business domains. The process can be executed by computing devices within the corresponding domain, and certain input parameters or intermediate results in the process can be manually adjusted to help improve accuracy.
[0021] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a server as an example.
[0022] It should be noted that the server can be a single device or a system composed of multiple devices, i.e., a distributed server. This application does not make any specific limitations on this.
[0023] Figure 1 The process includes the following steps: S101. The user-input multimodal data is matched and calculated with each scene label in the predefined multi-scene label library to obtain the scene recognition result and the confidence level corresponding to the scene recognition result.
[0024] Multimodal data refers to text, voice, and image data, as well as mixed data such as images, voice, and text, input by users through interactive entry points such as web, mobile, and smart terminals. Its core function is to carry users' business needs information.
[0025] Text-based data can include plain text, colloquial text with punctuation, and multi-line text with line breaks. Audio-based data can include multiple languages such as Mandarin Chinese, Cantonese, and English.
[0026] Image data can include images in formats such as JPG, PNG, and PDF.
[0027] The multi-scenario tag library refers to a predefined collection of scenario tags covering multiple fields such as enterprise office, government services, financial approval, and intelligent manufacturing. Each tag is associated with scenario keywords, user roles, common entities, and other features.
[0028] In some embodiments of this application, the multimodal data input by the user is matched and calculated with each scene label in a predefined multi-scene label library to obtain the scene recognition result and the confidence level corresponding to the scene recognition result, specifically including: Multimodal data is standardized to obtain unified text data. Scene-related keywords are extracted from the unified text data and combined with user role information and user historical interaction records to construct a multi-dimensional feature vector. Using a predefined confidence algorithm, the matching confidence of the multi-dimensional feature vector with each scene tag in a multi-scene tag library is calculated to obtain the scene confidence corresponding to each scene tag. When the scene confidence is greater than or equal to a preset threshold, the scene tag is taken as the scene recognition result.
[0029] Furthermore, the method also includes generating a scenario confirmation query instruction when the scenario confidence level is less than a preset threshold.
[0030] Among them, the predefined confidence algorithm can be, for example, the TF-IDF algorithm, which calculates the degree of matching between the feature vector and the core features of each scene label and outputs a scene confidence score of 0-100%.
[0031] S102. Perform semantic parsing and information extraction processing on multimodal data through a large language model to obtain user intent and associated entities.
[0032] For example, the multimodal data is first preprocessed, including cleaning up meaningless interjections, correcting typos, performing domain-specific word segmentation, and labeling candidate entities such as dates, amounts, and numbers. A finely tuned large model is then used to jointly extract intent and entities, avoiding semantic fragmentation caused by separate extraction. Related entities are clearly categorized into three types: core entities, attribute entities, and constraint entities, ensuring information integrity.
[0033] S103. Based on the scene recognition results, select initial candidate processes from the preset multi-scene process knowledge graph.
[0034] In some embodiments of this application, based on scene recognition results, initial candidate processes are selected from a preset multi-scene process knowledge graph, specifically including: The pre-defined multi-scenario process knowledge graph includes process definitions, scenario tags, startup parameters, and business rules. All process nodes associated with the scenario recognition results are selected from the pre-defined multi-scenario process knowledge graph as initial candidate processes.
[0035] The process definition may include, for example, a process identification number (ID), a process name, a workflow engine to which it is compatible, and a list of required parameters. Startup parameters may include, for example, a parameter ID, name, data type, and whether it is required. Business rules may include, for example, a rule ID and a rule description.
[0036] Based on the scene recognition results, by using the scene edge relationship between process definition and scene tag in the multi-scene process knowledge graph, all process definition nodes associated with the scene tag are accurately filtered to avoid interference from irrelevant processes. Extract the basic information corresponding to the selected process definition nodes to form an initial candidate process set. At the same time, extract the required start-up parameters and business rules of each process to prepare data in advance for subsequent matching steps.
[0037] S104. Based on the matching degree between user intent, entity association and initial candidate process, determine the target process instance.
[0038] In some embodiments of this application, a target process instance is determined based on the matching degree between user intent, entity association, and initial candidate processes, specifically including: Extract the process definition descriptions of each initial candidate process, and calculate the semantic similarity between the user intent and the process definition description using a semantic similarity algorithm to obtain the semantic matching degree for each initial candidate process. Extract the list of required startup parameters for each initial candidate process, and calculate the parameter matching degree by comparing the number of attribute entities in the associated entity with the total number of required startup parameters. Filter the initial candidate processes whose semantic matching degree is greater than or equal to a preset semantic threshold, and whose parameter matching degree is greater than or equal to a preset threshold, to obtain a secondary candidate process set. When the secondary candidate process set includes at least two candidate processes, calculate the comprehensive matching degree of each candidate process in the secondary candidate process set. Select the process with the highest comprehensive matching degree as the target process instance.
[0039] Furthermore, to avoid incorrect process selection due to ambiguous matching results, the method also includes: When the overall matching degree is similar, a scenario confirmation query instruction is generated.
[0040] S105. Start the target process instance through the preset parameter mapping rules and entity associations.
[0041] In some embodiments of this application, a target process instance is initiated through preset parameter mapping rules and entity associations, specifically including: Based on preset parameter mapping rules, entity associations are converted into a parameter format recognizable by the target workflow instance. The target workflow instance creation interface is then called to start the target workflow instance.
[0042] The method provided in this application, through multimodal data standardization and multi-dimensional scenario matching, eliminates the need for users to understand technical details such as process names and parameter formats, thus solving the high barrier to entry problem caused by standardized forms or API calls and significantly reducing the learning curve for non-technical personnel. By leveraging a large language model to automatically parse natural language semantics, extract user intent and related entities, it replaces the manual conversion of requirements into system instructions, further simplifying user interaction and improving the efficiency of requirement conversion.
[0043] Based on scene recognition results, initial candidate processes are automatically selected from a multi-scene process knowledge graph. Target process instances are determined through both semantic and parameter matching, eliminating the strong binding between processes and single scenes and eliminating the need for manual selection or redevelopment of selection logic. Adaptation between associated entities and target process instances is achieved through pre-defined parameter mapping rules, without the need to adjust parameter validation or triggering logic for new scenarios. This effectively reduces the development cost and cycle of expanding to new scenarios, achieving high efficiency and flexibility in triggering processes across multiple domains.
[0044] It should be noted that, although the embodiments in this application are based on... Figure 1 Steps S101 to S105 will be described sequentially, but this does not mean that steps S101 to S105 must be performed in a strict order. The reason this embodiment follows this order is... Figure 1 The order in which steps S101 to S105 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S101 to S105 can be appropriately adjusted according to actual needs.
[0045] based on Figure 1 In addition to the method described herein, this specification also provides some specific implementation schemes and extension schemes of this method, which will be further explained below.
[0046] Figure 2 This application provides a flowchart illustrating a cross-scenario intent parsing process, as shown in the embodiments below. Figure 2The method begins by receiving standardized input text, first loading scene labels to constrain the text within a specific scene, and then optimizing the semantics through secondary text enhancement. Subsequently, a large model is used to simultaneously extract user intent and related entities. After verifying the confidence level of the results, if the confidence level meets the standard, the entities are standardized; otherwise, the text is re-inputted after conflict resolution and information supplementation. Finally, a structured parsing result containing scene identifiers, intent, and standardized entities is output, completing cross-scene intent parsing and achieving efficient conversion from natural language requirements to precise structured information.
[0047] Figure 3 A schematic diagram of intelligent process matching based on a large language model is provided for an embodiment of this application, such as... Figure 3 As shown, starting with the intent parsing results, the system first filters by invoking a multi-scenario process knowledge graph, selecting all processes within a scenario and obtaining candidate processes through semantic similarity matching. After validity assessment, if no candidate process is found, conflict resolution and large-scale model queries are used to obtain supplementary user scenario information before re-filtering. If a candidate process exists, parameter matching is checked. This checks the matching degree between the startup parameters and attribute entities. If the matching degree is <80%, conflict resolution and large-scale model queries are used to obtain supplementary user parameter information before re-verification. If the matching degree is ≥80%, the target process is determined. This process achieves accurate selection and determination of the target process across scenarios through dual matching of scenarios and parameters and a closed-loop supplementation mechanism.
[0048] Based on the same idea, some embodiments of this application also provide devices and non-volatile computer storage media corresponding to the above methods.
[0049] Figure 4 A schematic diagram of a process automatic triggering device based on a large language model, provided for embodiments of this application, includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform any of the above-described automatic process triggering methods based on a large language model.
[0050] Some embodiments of this application provide a non-volatile computer storage medium for automatic process triggering based on a large language model, which stores computer-executable instructions capable of executing any of the above-described automatic process triggering methods based on a large language model.
[0051] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.
[0052] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.
[0053] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention 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.
[0054] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will 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 and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0055] 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.
[0056] 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 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0057] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0058] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0059] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0060] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0061] The above description is merely an embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the technical principles of this application should fall within the protection scope of this application.
Claims
1. A method for automatically triggering processes based on a large language model, characterized in that, The method includes: The user-input multimodal data is matched and calculated with each scene label in a predefined multi-scene label library to obtain scene recognition results and the confidence scores corresponding to the scene recognition results; The multimodal data is semantically parsed and information extracted using a large language model to obtain user intent and associated entities; Based on the scene recognition results, initial candidate processes are selected from a preset multi-scene process knowledge graph; Based on the matching degree between the user intent, the entity association, and the initial candidate process, a target process instance is determined; The target process instance is started by using preset parameter mapping rules and the entity association.
2. The method according to claim 1, characterized in that, The multimodal data includes: Text data, audio data, image data, and mixed data.
3. The method according to claim 1, characterized in that, The step of matching and calculating the multimodal data input by the user with each scene label in a predefined multi-scene label library to obtain the scene recognition result and the confidence level corresponding to the scene recognition result specifically includes: The multimodal data is standardized to obtain unified text data; Scene-related keywords are extracted from the unified text data, and multi-dimensional feature vectors are constructed by combining the user's role information and user historical interaction records. The matching confidence of the multi-dimensional feature vector with each scene tag in the multi-scene tag library is calculated using a predefined confidence algorithm to obtain the scene confidence corresponding to each scene tag. When the confidence level of the scene is greater than or equal to a preset threshold, the scene label is taken as the scene recognition result.
4. The method according to claim 3, characterized in that, The method further includes: When the confidence level of the scenario is less than a preset threshold, a scenario confirmation query instruction is generated.
5. The method according to claim 1, characterized in that, The step of selecting initial candidate processes from a preset multi-scene process knowledge graph based on the scene recognition results specifically includes: The preset multi-scenario process knowledge graph includes process definitions, scenario tags, startup parameters, and business rules; All process nodes associated with the scene recognition results are selected from the preset multi-scenario process knowledge graph as initial candidate processes.
6. The method according to claim 1, characterized in that, The step of determining the target process instance based on the matching degree between the user intent, the entity association, and the initial candidate processes specifically includes: Extract the process definition description of each initial candidate process, and calculate the semantic similarity between the user intent and the process definition description using a semantic similarity algorithm to obtain the semantic matching degree corresponding to each initial candidate process; Extract the list of required startup parameters for each initial candidate process, and calculate the parameter matching degree by comparing the number of attribute entities in the associated entity with the total number of required startup parameters; Filter initial candidate processes that have a semantic matching degree greater than or equal to a preset semantic threshold and a parameter matching degree greater than or equal to a preset threshold to obtain a secondary candidate process set; When the set of secondary candidate processes includes at least two candidate processes, the comprehensive matching degree of each candidate process in the set of secondary candidate processes is calculated. The process with the highest overall matching degree is selected as the target process instance.
7. The method according to claim 6, characterized in that, The method further includes: When the overall matching degree is similar, a scenario confirmation query instruction is generated.
8. The method according to claim 1, characterized in that, The step of initiating the target process instance through preset parameter mapping rules and entity associations specifically includes: According to the preset parameter mapping rules, the entity association is converted into a parameter format that can be recognized by the target workflow instance; Call the target workflow instance creation interface to start the target workflow instance.
9. A process automatic triggering device based on a large language model, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the automatic process triggering method based on a large language model as described in any one of claims 1-8.
10. A non-volatile computer storage medium for automatic process triggering based on a large language model, storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of executing the automatic process triggering method based on a large language model as described in any one of claims 1-8.