Method and device for generating automatic business process based on fuzzy requirement semantic analysis

Abstract

The application discloses an automatic business process generation method and device based on fuzzy requirement semantic analysis. The scheme processes natural language business requirements through a preset requirement understanding intelligent agent, improving the accurate extraction capability of natural language business requirements to structured process instructions. Moreover, a multi-round context memory mechanism is introduced to continuously track the multi-round dialogue history of the user and the process construction state, realize the dynamic iteration and continuity support of process generation, and enhance the response capability of the system to complex multi-round business requirements. In addition, through multi-agent collaborative driving, an end-to-end automatic closed loop is formed from semantic analysis to process construction, technical mapping and deployment execution. Combined with the preset control and scheduling intelligent agent, task scheduling and abnormal self-repair are realized, which significantly improves the intelligent level and robustness of the business process generation system, can meet the dynamic changes and various expressions of business requirements, and thus improves the flexibility and accuracy of business process generation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to an automatic business process generation method and device based on fuzzy requirement semantic analysis. BACKGROUND

[0002] With the acceleration of enterprise digitization and intelligentization, business process requirement expression gradually presents the trend of natural language and unstructured. Users often describe business goals and operation logic in colloquial language, and express them in a mixed way of text and images. The language form has strong nesting and high ambiguity, which brings significant challenges to system understanding.

[0003] Currently, it is difficult to accurately identify task intent, condition relationship and trigger event by using keyword extraction or template matching method. At the same time, business process requirement input data types are increasingly diversified, and gradually evolve into multi-modal, multi-source and heterogeneous structure. In addition to text information, voice, image and other information have become important sources of process triggering and state perception, and come from different terminals, systems or business platforms.

[0004] Therefore, the traditional automatic business process generation method cannot meet the dynamic changes and diverse expressions of business requirements, and has the problems of poor flexibility and low accuracy. SUMMARY

[0005] The present application aims to at least solve the technical problems existing in the prior art. To this end, the first aspect of the present application provides an automatic business process generation method based on fuzzy requirement semantic analysis, which comprises:

[0006] The preset requirement understanding intelligent agent processes the received natural language business requirement, and generates a structured process instruction corresponding to the natural language business requirement;

[0007] The structured process instruction is processed by the multi-round dialogue context provided by the preset intent recognition intelligent agent and the preset context memory intelligent agent to generate an intent recognition result; wherein the intent recognition result includes business goal, trigger condition, key node and output requirement;

[0008] The structured process instruction and the intent recognition result are processed by the preset process construction intelligent agent calling the process graph generation engine to generate a preliminary process model;

[0009] The target node in the preliminary process model is bound to the target software technology component and the target data structure by the preset software flow mapping intelligent agent to generate a configuration file;

[0010] The target business process is deployed to the target execution environment based on the configuration file by the preset deployment intelligent agent to perform state monitoring and resource scheduling.

[0011] In a possible implementation, the preset demand understanding agent processes the received natural language business demand, generates a structured process instruction corresponding to the natural language business demand, including:

[0012] The preset demand understanding agent calls a preset large language model to perform context semantic analysis on the natural language business demand, and extracts a candidate semantic unit set; the candidate semantic unit set includes a plurality of candidate semantic units;

[0013] A semantic recognizer enhanced based on a preset rule template and a dictionary performs semantic labeling on each candidate semantic unit, and outputs a typed intermediate result; the typed intermediate result includes a task, a process node, a constraint condition, and a trigger event;

[0014] The typed intermediate result is standardized and confirmed based on a preset domain knowledge base, and a structured process instruction is generated.

[0015] In a possible implementation, the preset demand understanding agent calls a preset large language model to perform context semantic analysis on the natural language business demand, and extracts a candidate semantic unit set, including:

[0016] Obtain modal information corresponding to the natural language business demand; the modal information includes at least any one of text, table, image, and voice;

[0017] The preset demand understanding agent calls a preset large language model to determine a target modal information processing mode corresponding to the modal information;

[0018] The natural language business demand is contextually semantically analyzed based on the target modal information processing mode, and a candidate semantic unit set is extracted.

[0019] In a possible implementation, the typed intermediate result is standardized and confirmed based on a preset domain knowledge base, and a structured process instruction is generated, including:

[0020] The candidate entities in the typed intermediate result are encoded into candidate entity vectors by a preset semantic encoder, and each preset term entity in the preset domain knowledge base is encoded into a standard term vector;

[0021] The similarity between the candidate entity vectors and the standard term vectors is calculated to obtain a similarity score list;

[0022] Based on the similarity score list, a target term vector matching the candidate entity vector is determined from each standard term vector;

[0023] The typed intermediate result is standardized and confirmed according to the current similarity score between the candidate entity vector and the target term vector, and a structured process instruction is generated.

[0024] In a possible implementation, the standardized confirmation of the typed intermediate result is performed according to the current similarity score between the candidate entity vector and the target term vector, the structured process instruction is generated, and the generation includes:

[0025] obtaining a preset similarity threshold;

[0026] if the current similarity score is greater than the preset similarity threshold, the candidate entity corresponding to the candidate entity vector is replaced by the preset term entity corresponding to the target term vector, and the structured process instruction is generated;

[0027] if the current similarity score is less than or equal to the preset similarity threshold, the preset term entity corresponding to the target term vector is marked as an ambiguous item, and the structured process instruction is generated.

[0028] In a possible implementation, the structured process instruction and the intent recognition result are processed by calling the process graph generation engine through the preset process construction agent, and a preliminary process model is generated, including:

[0029] The process graph construction domain-specific language description file is constructed by calling the process graph construction agent through the preset process; wherein the domain-specific language description file includes multiple types of control structures, and the multiple types include approval, condition, parallel, and rollback;

[0030] The structured process instruction and the intent recognition result are processed based on the domain-specific language description file to generate a graphical process graph;

[0031] The graphical process graph is edited and processed through the preset visual editing interface to generate a preliminary process model.

[0032] In a possible implementation, the method further includes:

[0033] The other agents are scheduled and coordinated and the abnormal processing is performed by the preset control and scheduling agent, and the user context memory of the preset context memory agent is updated; wherein the other agents include the preset demand understanding agent, the preset intent recognition agent, the preset context memory agent, the preset process construction agent, the preset software flow mapping agent, and the preset deployment agent; the scheduling and coordination include managing the execution dependency and the scheduling order between the other agents; the abnormal processing includes detecting the task abnormal condition of the other agents and triggering the preset error recovery logic corresponding to the task abnormal condition.

[0034] The second aspect of the present application proposes an automatic business process generation device based on fuzzy demand semantic analysis, which includes:

[0035] The demand understanding module is configured to process the received natural language business demand by using a preset demand understanding agent, and generate a structured process instruction corresponding to the natural language business demand.

[0036] The intent recognition module is configured to process the structured process instruction by using a multi-round dialogue context provided by a preset intent recognition agent and a preset context memory agent, and generate an intent recognition result. The intent recognition result includes a business target, a trigger condition, a key node, and an output requirement.

[0037] The process construction module is configured to process the structured process instruction and the intent recognition result by using a preset process construction agent and a process graph generation engine, and generate a preliminary process model.

[0038] The software stream mapping module is configured to bind a target node in the preliminary process model with a target software technology component and a target data structure by using a preset software stream mapping agent, and generate a configuration file.

[0039] The deployment module is configured to deploy a target business process to a target execution environment based on the configuration file by using a preset deployment agent, so as to perform state monitoring and resource scheduling.

[0040] In a possible implementation, the demand understanding module is specifically configured to:

[0041] The preset demand understanding agent is configured to call a preset large language model to perform context semantic analysis on the natural language business demand, and extract a candidate semantic unit set. The candidate semantic unit set includes a plurality of candidate semantic units.

[0042] A semantic recognizer enhanced based on a preset rule template and a dictionary is configured to perform semantic labeling on each candidate semantic unit, and output a typed intermediate result. The typed intermediate result includes a task, a process node, a constraint condition, and a trigger event.

[0043] The preset domain knowledge base is configured to perform standardization confirmation on the typed intermediate result, and generate a structured process instruction.

[0044] In a possible implementation, the demand understanding module is further configured to:

[0045] Modal information corresponding to the natural language business demand is acquired. The modal information includes at least any one of text, a table, an image, and voice.

[0046] The preset demand understanding agent is configured to call a preset large language model to determine a target modal information processing mode corresponding to the modal information.

[0047] Based on the target modality information processing method, the contextual semantics of natural language business requirements are parsed to extract a set of candidate semantic units.

[0048] In one possible implementation, the aforementioned requirement understanding module is further used for:

[0049] The candidate entities in the typed intermediate results are encoded into candidate entity vectors by a preset semantic encoder, and each preset term entity in the preset domain knowledge base is encoded into a standard term vector.

[0050] The similarity between the candidate entity vectors and the standard term vectors is calculated to obtain a list of similarity scores;

[0051] Based on the similarity score list, the target term vector that matches the candidate entity vector is determined from each standard term vector;

[0052] The typified intermediate results are standardized and confirmed based on the current similarity score between the candidate entity vector and the target term vector, and structured process instructions are generated.

[0053] In one possible implementation, the aforementioned requirement understanding module is further used for:

[0054] Obtain the preset similarity threshold;

[0055] If the current similarity score is greater than the preset similarity threshold, the candidate entity corresponding to the candidate entity vector is replaced by the preset term entity corresponding to the target term vector, and a structured process instruction is generated.

[0056] If the current similarity score is less than or equal to the preset similarity threshold, the preset term entity corresponding to the target term vector will be marked as an ambiguous item, and a structured process instruction will be generated.

[0057] In one possible implementation, the above process construction module is specifically used for:

[0058] The intelligent agent is constructed by calling the flowchart through a pre-defined process to build a domain-specific language description file; the domain-specific language description file includes various types of control structures, including approval, condition, parallel, and rollback.

[0059] The structured process instructions and intent recognition results are processed based on domain-specific language description files to generate graphical flowcharts.

[0060] The graphical flowchart can be edited through a preset visual editing interface to generate a preliminary process model.

[0061] In one possible implementation, the aforementioned automated business process generation device based on fuzzy demand semantic parsing is further used for:

[0062] The preset control and scheduling agent respectively schedules and coordinates other agents and handles exceptions, and updates the user context memory of the preset context memory agent; wherein the other agents include a preset demand understanding agent, a preset intention recognition agent, a preset context memory agent, a preset flow construction agent, a preset software flow mapping agent, and a preset deployment agent; the scheduling and coordination includes managing the execution dependencies and scheduling sequences between the other agents, and the exception handling includes detecting task exception conditions of the other agents and triggering preset error recovery logic corresponding to the task exception conditions.

[0063] The third aspect of the present application provides an electronic device, the electronic device includes a processor and a memory, the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to realize the automatic business process generation method based on fuzzy demand semantic analysis as described in the first aspect.

[0064] The fourth aspect of the present application provides a computer readable storage medium, the storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to realize the automatic business process generation method based on fuzzy demand semantic analysis as described in the first aspect.

[0065] The embodiments of the present application have the following beneficial effects:

[0066] The embodiment of the application provides a method and device for generating an automatic business process based on fuzzy requirement semantic analysis, which comprises the following steps: processing a received natural language business requirement by a preset requirement understanding agent to generate a structured process instruction corresponding to the natural language business requirement; processing the structured process instruction by a multi-round dialogue context provided by a preset intention recognition agent and a preset context memory agent to generate an intention recognition result; processing the structured process instruction and the intention recognition result by a preset process construction agent to generate a preliminary process model; binding a target node in the preliminary process model with a target software technology component and a target data structure by a preset software flow mapping agent to generate a configuration file; and deploying a target business process to a target execution environment based on the configuration file by a preset deployment agent to perform state monitoring and resource scheduling. The natural language business requirement is processed by the preset requirement understanding agent, and the accurate extraction capability of the natural language business requirement to the structured process instruction is improved. In addition, the multi-round context memory mechanism is introduced, the multi-round dialogue history of a user and the process construction state are continuously tracked, the dynamic iteration and continuity support of the process generation are realized, and the response capability of the system to complex multi-round business requirements is enhanced. In addition, the multi-agent collaborative driving is used from semantic analysis to process construction, technology mapping and deployment execution to form an end-to-end automatic closed loop, the preset control and scheduling agent is combined to realize task scheduling and abnormal self-repairing, the intelligent level and robustness of the agent-driven business process generation system based on fuzzy requirement semantic analysis are significantly improved, the dynamic change and various expressions of business requirements can be met, and the flexibility and accuracy of the business process generation are improved. In addition, a non-technical user can directly describe the business process requirement in natural language, and the system can automatically complete process design, component mapping and online deployment to realize the "zero code" generation and execution of the business process, significantly reduce the modeling threshold and improve the business automation level. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 A block diagram of a computer device provided by the embodiment of the application;

[0068] Figure 2 A step flowchart of a method for generating an automatic business process based on fuzzy requirement semantic analysis provided by the embodiment of the application;

[0069] Figure 3 A step flowchart of generating a structured process instruction provided by the embodiment of the application;

[0070] Figure 4 A step flowchart of extracting a candidate semantic unit set provided by the embodiment of the application;

[0071] Figure 5Another step flow chart for generating a structured flow instruction provided by an embodiment of the present application;

[0072] Figure 6 Another step flow chart for generating a structured flow instruction provided by an embodiment of the present application;

[0073] Figure 7 A step flow chart for generating a preliminary flow model provided by an embodiment of the present application;

[0074] Figure 8 A structural block diagram of an automatic business process generation device based on fuzzy requirement semantic analysis provided by an embodiment of the present application. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0076] With the acceleration of enterprise digitization and intelligentization, business process requirement expression gradually presents a natural language and unstructured trend. Users often describe business goals and operation logic in colloquial language, and express them in a mixed form of text and images. The language form has strong nesting and high ambiguity, which brings significant challenges to system understanding. Traditional keyword extraction or template matching methods are difficult to accurately identify task intent, condition relationship and trigger event, and strong semantic analysis capability is urgently needed to support high-precision process modeling. At the same time, the input data types of business process requirements are increasingly diversified, gradually evolving into multi-modal, multi-source and heterogeneous structures. In addition to text information, voice, image and other information have become important sources of process triggering and state perception, and come from different terminals, systems or business platforms. This requires the system to have cross-modal semantic alignment and fusion capability, and to be able to realize unified data representation, event analysis and process response, and adapt to complex information flow in real business scenarios.

[0077] In the traditional business process construction and deployment process, developers and enterprises generally face the following challenges:

[0078] (1) Natural language requirement expression is fuzzy and lacks structured ability. Business personnel usually describe process goals and rules in free language, and the semantic structure is complex and highly uncertain. Existing systems mostly rely on keyword matching, rule engine and other shallow methods for analysis, which are difficult to identify key elements such as task action, trigger condition and constraint logic, resulting in low system understanding accuracy and high misjudgment rate.

[0079] (2) Process modeling relies heavily on manual participation, and the development threshold is high. The mainstream process platform adopts visual modeling and explicit configuration methods, which requires engineers with development experience to manually draw process diagrams, configure task nodes, and execute logic. Non-technical users cannot directly participate in process creation and maintenance, the process construction cycle is long, and the cross-team collaboration cost is high.

[0080] (3) The process automation link is difficult to close the loop, and lacks intelligent deployment and evolution capabilities. The current system lacks unified semantic support and intelligent linkage mechanisms in stages such as demand acquisition, process modeling, component configuration, and deployment execution. Version switching, context migration, and dynamic adjustment during process operation still require manual intervention, which cannot achieve end-to-end automation, seriously affecting iteration efficiency and system maintainability.

[0081] (4) The support capability for multi-language and multi-modal business input is insufficient. In actual business, user input may contain mixed expressions of Chinese and English, voice instructions, image materials, and other multi-modal data. The existing system has limited processing capability for unstructured input, lacks a unified semantic understanding framework and modal fusion mechanism, resulting in limited business scope and weak adaptability.

[0082] (5) The process scheduling and decision of the current automation platform rely on static rule templates, and the reaction to dynamic environmental changes is slow. It lacks intelligent decision-making mechanisms based on context awareness and task dependency modeling, making it difficult to support abnormal processing, branch jumping, and collaborative execution of complex processes.

[0083] Therefore, the present application proposes an automatic business process generation method based on fuzzy requirement semantic analysis. The preset requirement understanding agent processes natural language business requirements, improving the accurate extraction of structured process instructions from natural language business requirements. In addition, a multi-round context memory mechanism is introduced to continuously track user multi-round dialogue history and process construction state, enabling dynamic iteration and continuity support for process generation, enhancing the system's response capability to complex multi-round business requirements. Furthermore, through multi-agent collaborative driving, an end-to-end automation loop is formed from semantic analysis to process construction, technology mapping, and deployment execution. Combined with preset control and scheduling agents, task scheduling and abnormal self-repair are achieved, significantly improving the intelligent level and robustness of the agent-driven business process generation system based on fuzzy requirement semantic analysis. It can meet the dynamic changes and diverse expressions of business requirements, improve the flexibility and accuracy of business process generation. Moreover, non-technical users can directly describe business process requirements in natural language, and the system automatically completes process design, component mapping, and online deployment, realizing "zero-code" generation and execution of business processes, significantly reducing the modeling threshold and improving business automation.

[0084] The terms "first", "second", "third", etc. are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise stated, the meaning of "a plurality of" is two or more. In addition, the use of "based on" or "according to" means open and inclusive, because the process, step, calculation or other action based on one or more stated conditions or values can be based on additional conditions or values beyond those stated in practice.

[0085] The automatic business process generation method based on fuzzy requirement semantic analysis provided in the present application can be applied to a computer device (electronic device). The computer device can be a server or a terminal. The server can be a single server or a server cluster composed of multiple servers. The embodiments of the present application do not make specific limitations on this. The terminal can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices.

[0086] Taking the computer device as a server for example, Figure 1 A block diagram of a server is shown, as Figure 1 As shown, the server can include a processor and a memory connected by a system bus. The processor of the server is used to provide computing and control capabilities. The memory of the server includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The computer program is executed by the processor to implement an automatic business process generation method based on fuzzy requirement semantic analysis.

[0087] Those skilled in the art can understand, Figure 1 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the server to which the scheme of the present application is applied. Alternatively, the server can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0088] It should be noted that the execution subject of the embodiments of the present application can be a computer device or an automatic business process generation device based on fuzzy requirement semantic analysis. The following method embodiments are described with the computer device as the execution subject.

[0089] Figure 2 A step flowchart of an automatic business process generation method based on fuzzy requirement semantic analysis provided by the embodiments of the present application. As Figure 2As shown, the method comprises the following steps:

[0090] Step 202, the agent processes the received natural language business requirement through a preset requirement understanding, and generates a structured process instruction corresponding to the natural language business requirement.

[0091] Among them, the user can input the natural language business requirement through a preset visual editing interface, and can support input of multiple languages and multiple modal information. The natural language business requirement can include business goals, constraint conditions, etc. Optionally, the preset visual editing interface can also support user identity recognition, input verification, context state maintenance and multi-round dialogue management, to ensure the coherence and integrity of the input.

[0092] After obtaining the natural language business requirement input by the user, the received natural language business requirement is processed by a preset requirement understanding agent, so as to generate a structured process instruction corresponding to the natural language business requirement. In some optional embodiments, as shown in Figure 3 , Figure 3 A step flow chart for generating a structured process instruction provided by an embodiment of the present application comprises:

[0093] Step 302, the agent calls a preset large language model to perform context semantic analysis on the natural language business requirement, and extracts a candidate semantic unit set.

[0094] Step 304, based on the semantic recognizer enhanced by the preset rule template and the dictionary, each candidate semantic unit is semantically labeled, and a typed intermediate result is output.

[0095] Step 306, based on the preset domain knowledge base, the typed intermediate result is standardized and confirmed, and a structured process instruction is generated.

[0096] Among them, the candidate semantic unit set comprises a plurality of candidate semantic units, the preset large language model can be denoted as , the input natural language business requirement is denoted as , and the extracted candidate semantic unit set can be denoted as , , wherein , is a candidate semantic unit, i The value range of is 1-n.

[0097] In some optional embodiments, as shown in Figure 4 , Figure 4 A step flow chart for extracting a candidate semantic unit set provided by an embodiment of the present application comprises:

[0098] At step 402, modal information corresponding to the natural language business requirement is obtained.

[0099] At step 404, a preset large language model is called by a preset requirement understanding agent to determine a target modal information processing mode corresponding to the modal information.

[0100] At step 406, context semantic analysis is performed on the natural language business requirement based on the target modal information processing mode, and a candidate semantic unit set is extracted.

[0101] The modal information can include at least any one of text, table, image, and voice. The user can input the natural language business requirement through a preset visual editing interface supporting multiple languages. For example, the natural language business requirement can be "I want to apply for a database server, which is used in a production environment and needs to open access permission to the external network."

[0102] Thus, the preset large language model can be called by the preset requirement understanding agent to determine the target modal information processing mode corresponding to the modal information, so as to perform context semantic analysis on the natural language business requirement based on the target modal information processing mode and extract a candidate semantic unit set. The target modal information processing mode corresponding to the natural language business requirement of different modal information such as text, table, image, and voice can be preset in advance. By supporting semantic analysis and process construction in a multi-language environment, the multi-modal information processing capability is integrated, the business requirement of cross-language and multi-type data input is met, and the universality and application range of the system are improved.

[0103] Then, the semantic recognizer enhanced based on the preset rule template and the dictionary can be used to perform semantic labeling on each candidate semantic unit in the candidate semantic unit set, thereby outputting a typed intermediate result. The typed intermediate result includes tasks, process nodes, constraint conditions, and trigger events.

[0104] Specifically, .

[0105] For example, when the natural language business requirement input by the user is "after the customer completes registration, the system needs to send a welcome email and start a 7-day free trial process", the output typed intermediate result can include:

[0106] {

[0107] "task": ["send welcome email", "start free trial process"],

[0108] ​​​"Process Nodes": ["Registration Completed", "Email Sent", "Trial Process Started"],

[0109] "Constraints": ["Customer completes registration"],

[0110] Triggered event: ["Registration successful"]

[0111] }

[0112] After obtaining the typified intermediate results, they can be standardized and validated based on a preset domain knowledge base to generate structured process instructions. In some optional embodiments, such as... Figure 5 As shown, Figure 5 A flowchart illustrating another step in generating structured process instructions provided in this application embodiment includes:

[0113] Step 502: Encode the candidate entities in the typed intermediate results into candidate entity vectors using a preset semantic encoder, and encode each preset term entity in the preset domain knowledge base into a standard term vector.

[0114] Step 504: Calculate the similarity between the candidate entity vector and the standard term vector to obtain a list of similarity scores.

[0115] Step 506: Based on the similarity score list, determine the target term vector that matches the candidate entity vector from each standard term vector.

[0116] Step 508: Standardize and confirm the typified intermediate results based on the current similarity score between the candidate entity vector and the target term vector, and generate structured process instructions.

[0117] In this context, candidate entities in the typed intermediate results are denoted as... Each preset term entity in the preset domain knowledge base is denoted as The predefined semantic encoder can be a bidirectional encoder representation from Transformers (BERT). This predefined semantic encoder can encode candidate entities and predefined term entities into high-dimensional semantic vectors, thus obtaining the candidate entity vector. Vector of Standard Terminology .

[0118] Next, for each standard term vector, the similarity between the candidate entity vector and the standard term vector can be calculated, resulting in a similarity score list composed of multiple current similarity scores. Current similarity score It can be calculated using formula (1).

[0119] (1)

[0120] Thus, the target term vector matching the candidate entity vector can be determined from each standard term vector based on the list of similarity scores. Finally, the typed intermediate result is standardized and confirmed according to the current similarity score between the candidate entity vector and the target term vector, and the structured process instruction is generated.

[0121] In some optional embodiments, as shown in Figure 6 , Figure 6 Another step flowchart for generating a structured process instruction provided by the embodiments of the present application includes:

[0122] Step 602, obtaining a preset similarity threshold.

[0123] Step 604, if the current similarity score is greater than the preset similarity threshold, replacing the candidate entity corresponding to the candidate entity vector with the preset term entity corresponding to the target term vector, and generating a structured process instruction.

[0124] Step 606, if the current similarity score is less than or equal to the preset similarity threshold, marking the preset term entity corresponding to the target term vector as an ambiguous item, and generating a structured process instruction.

[0125] The preset similarity threshold may be pre-defined, and optionally, can be adaptively adjusted according to the corpus complexity and business fault tolerance. For example, the preset similarity threshold can be set between 0.75-0.9.

[0126] If the current similarity score is greater than the preset similarity threshold, it is considered that the matching is valid, and the candidate entity corresponding to the candidate entity vector can be replaced by the preset term entity corresponding to the target term vector, and a structured process instruction is generated. If the current similarity score is less than or equal to the preset similarity threshold, the preset term entity corresponding to the target term vector is marked as an ambiguous item, and a structured process instruction is generated after triggering a subsequent manual confirmation process.

[0127] For example, in the IT service management scenario of an enterprise, employees often need to initiate service processes such as “opening VPN”, “applying for database”, “allocating virtual machines”, etc. These processes usually involve multi-department approval, security compliance verification, automatic execution and log recording, and the following structured process instruction can be obtained after processing:

[0128] {

[0129] "flow_name": "database resource application process",

[0130] "actors": ["Employee", "Direct Manager", "IT Security", "Automation System"],

[0131] "steps": [

[0132] {"name": "Submit Application", "actor": "Employee"},

[0133] {"name": "Manager Approval", "actor": "Direct Manager"},

[0134] {

[0135] "name": "Security Review",

[0136] "actor": "IT Security",

[0137] "condition": "Resource is production environment or involves external network access"

[0138] },

[0139] {

[0140] "name": "Resource Allocation",

[0141] "actor": "Automation System",

[0142] "action": "Execute database deployment script"

[0143] },

[0144] {

[0145] "name": "Notification and Logging",

[0146] "actor": "System",

[0147] "action": "Send deployment notification + write operation log"

[0148] }

[0149] ],

[0150] "constraints": {

[0151] "auto_notify": true,

[0152] "record_all_steps": true

[0153] }

[0154] }

[0155] In this embodiment, by combining the context semantic understanding of the large language model, the semantic recognition based on the rule and the entity dictionary, and the standardized verification of the domain knowledge base, high-precision extraction of business process tasks, nodes, constraints and events is realized, greatly enhancing the accuracy and professionalism of the structured instruction.

[0156] Step 204, the structured process instruction is processed by the multi-round dialogue context provided by the preset intent recognition agent and the preset context memory agent, and an intent recognition result is generated.

[0157] The intent recognition result includes business target, trigger condition, key node and output requirement, and the output requirement can include format requirement of output content. The preset context memory agent can model the context of the current input. For example, the process can specifically include: identifying the current user identity as "IT process administrator" with process publishing authority; determining the conversation context as "database service automation application"; loading the user's last 3 historical application process templates as prompt word supplementary input; constructing a structured context prompt word for guiding the preset large language model to generate stable output.

[0158] Optionally, the preset context memory agent can also continuously record the multi-round input, interaction history, process construction intermediate state and interaction data between the agents of the user. And maintain the vector database and context cache to realize multi-round conversation association and semantic consistency maintenance. It can also provide real-time context query and historical data support for the preset demand understanding agent and the preset intent recognition agent, assist business process optimization and iterative generation. By introducing the preset context memory agent, the multi-round dialogue state and task history are continuously maintained and dynamically updated, supporting demand continuation and process reuse in complex business scenarios, effectively overcoming the limitations of traditional one-time business process generation, and improving the adaptability and intelligent level of the system.

[0159] Then, the structured process instruction can be processed by the multi-round dialogue context provided by the preset intent recognition agent and the preset context memory agent to generate an intent recognition result. Specifically, the preset intent recognition agent can combine the multi-round dialogue context provided by the context memory agent, perform deep semantic understanding and function decomposition, identify the business target, trigger condition, key node and output requirement, form a complete business semantic structure, and finally output a perfect intent recognition result for further processing by the preset process construction agent.

[0160] Step 206, the structured process instruction and the intent recognition result are processed by the preset process construction agent calling the process graph generation engine to generate a preliminary process model.

[0161] The preset process construction agent can generate a preliminary process model according to the structured process instruction and the intention recognition result. In some optional embodiments, as shown in Figure 7 Figure 7 A step flowchart for generating a preliminary process model provided in the embodiment of the application includes the following steps.

[0162] In step 702, the preset process construction agent calls a process graph construction domain specific language description file.

[0163] In step 704, the structured process instruction and the intention recognition result are processed based on the domain specific language description file to generate a graphical process graph.

[0164] In step 706, the graphical process graph is edited through a preset visual editing interface to generate a preliminary process model.

[0165] The preset process construction agent can call a process graph construction domain specific language (Domain Specific Language, DSL) description file. The domain specific language description file includes multiple types of control structures, which can include but are not limited to approval, condition, parallel, and rollback. The structured process instruction and the intention recognition result can be processed based on the domain specific language description file to generate a graphical process graph, which can support interactive modification.

[0166] Then, the graphical process graph can be edited through a preset visual editing interface to generate a preliminary process model. Alternatively, the user can modify the node order, modify the branch condition and the executor type, and perform other editing processes through the preset visual editing interface. The system can automatically synchronize and update the user's editing operation to finally obtain a preliminary process model.

[0167] In this embodiment, the process modeling and visualization platform based on the domain specific language supports interactive process editing, parameter adjustment and abnormal rollback of the user, drives process optimization through the user feedback closed loop, realizes human-computer collaborative construction and intelligent iteration, and greatly improves the user operation experience and the process execution quality.

[0168] In step 208, the target node in the preliminary process model is bound to the target software technology component and the target data structure through a preset software flow mapping agent to generate a configuration file.

[0169] ​The target node in the abstract preliminary flow model can be automatically and accurately mapped to the target software technology component and the target data structure based on a preset semantic matching algorithm through the preset software flow mapping agent, and is bound with the corresponding target software technology component and target data structure, so that a configuration file conforming to the target running environment can be automatically generated, and syntax and compatibility checking can be automatically completed. The user can also preset parameters of the target software technology component to perform running environment adaptation detection and dynamically adjust the configuration to meet the platform requirements.

[0170] In step 210, the target business flow is deployed to the target execution environment based on the configuration file by the preset deployment agent to perform state monitoring and resource scheduling.

[0171] The preset deployment agent can automatically complete software flow deployment based on container technology to deploy the target business flow to the preset flow orchestration platform, realize end-to-end integration, that is, the target business flow can be deployed to the target execution environment based on the configuration file to perform state monitoring and resource scheduling.

[0172] In addition, automated integration testing and log monitoring can also be performed to track the flow running state and exceptions in real time. The preset intelligent exception diagnosis module is used to trigger a self-repair or rollback mechanism to ensure high availability and stability of the flow. At the same time, the deployment result, execution state and exception information can be fed back to the preset control and scheduling agent to support real-time display on the user interface.

[0173] Optionally, the running log of the target business flow can be collected in real time, and the execution state can be displayed. If the target business flow fails to be deployed, such as resource shortage, script exception, etc., a rollback process is automatically triggered, and the user is notified of the error details.

[0174] In some optional embodiments, the preset control and scheduling agent can also schedule and coordinate other agents and handle exceptions, and update the user context memory of the preset context memory agent. The other agents can include a preset demand understanding agent, a preset intent recognition agent, a preset context memory agent, a preset flow construction agent, a preset software flow mapping agent and a preset deployment agent. The scheduling and coordination include managing the execution dependencies and scheduling order among the other agents. The exception handling includes detecting task exception conditions of the other agents and triggering preset error recovery logic corresponding to the task exception conditions.

[0175] Specifically, the preset control and scheduling agent plays a core role in the entire system, can manage the execution dependencies and scheduling order among the agents, dynamically detect task failure, interruption or state inconsistency, trigger preset error recovery logic such as model switching, prompt word rewriting, rollback process, etc., and update the user context memory to realize long-term knowledge retention.

[0176] In the embodiment of the present application, the method comprises: processing the received natural language business requirement by a preset requirement understanding intelligent agent to generate a structured process instruction corresponding to the natural language business requirement; processing the structured process instruction by a multi-round dialogue context provided by a preset intention recognition intelligent agent and a preset context memory intelligent agent to generate an intention recognition result; processing the structured process instruction and the intention recognition result by a preset process construction intelligent agent to generate a preliminary process model; binding the target node in the preliminary process model with a target software technology component and a target data structure by a preset software flow mapping intelligent agent to generate a configuration file; and deploying the target business process to a target execution environment based on the configuration file by a preset deployment intelligent agent to perform state monitoring and resource scheduling. The natural language business requirement is processed by the preset requirement understanding intelligent agent, which improves the accurate extraction capability of the natural language business requirement to the structured process instruction. In addition, the multi-round context memory mechanism is introduced to continuously track the multi-round dialogue history of the user and the process construction state, to realize the dynamic iteration and continuity support of the process generation, and to enhance the response capability of the system to the complex multi-round business requirement. Furthermore, the multi-intelligent agent collaborative driving is performed from the semantic analysis to the process construction, the technology mapping, the deployment and execution to form an end-to-end automatic closed loop, the preset control and scheduling intelligent agent is combined to realize the task scheduling and the abnormal self-repairing, which significantly improves the intelligent level and the robustness of the intelligent agent driven business process generation system based on the fuzzy requirement semantic analysis, can meet the dynamic changes and the various expressions of the business requirement, and improves the flexibility and the accuracy of the business process generation. Moreover, the non-technical user can directly describe the business process requirement in the natural language, and the system automatically completes the process design, the component mapping and the online deployment to realize the "zero code" generation and execution of the business process, significantly reduces the modeling threshold, and improves the business automation degree.

[0177] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or steps or stages in other steps.

[0178] Figure 8A structural block diagram of an automatic business process generation device based on fuzzy requirement semantic analysis is provided for an embodiment of the present application.

[0179] As shown in Figure 8 The automatic business process generation device 800 based on fuzzy requirement semantic analysis includes the following:

[0180] A requirement understanding module 802 is configured to process a received natural language business requirement by a preset requirement understanding agent, and generate a structured process instruction corresponding to the natural language business requirement.

[0181] An intent recognition module 804 is configured to process the structured process instruction by a multi-round dialogue context provided by a preset intent recognition agent and a preset context memory agent, and generate an intent recognition result; wherein the intent recognition result includes a business target, a trigger condition, a key node, and an output requirement.

[0182] A process construction module 806 is configured to process the structured process instruction and the intent recognition result by a preset process construction agent calling a process graph generation engine, and generate a preliminary process model.

[0183] A software stream mapping module 808 is configured to bind a target node in the preliminary process model with a target software technology component and a target data structure by a preset software stream mapping agent, and generate a configuration file.

[0184] A deployment module 810 is configured to deploy a target business process to a target execution environment based on the configuration file by a preset deployment agent, to perform state monitoring and resource scheduling.

[0185] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment related to the method, and will not be described in detail here. Each module in the above automatic business process generation device based on fuzzy requirement semantic analysis can be realized by software, hardware, and combinations thereof, in whole or in part. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform the operations of each module.

[0186] In an embodiment of the present application, a computer device is provided, which includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0187] processing a received natural language business requirement by a preset requirement understanding agent, and generating a structured process instruction corresponding to the natural language business requirement;

[0188] The multi-round dialogue context provided by the preset intention recognition agent and the preset context memory agent is used to process the structured process instruction, and an intention recognition result is generated; the intention recognition result includes a business target, a trigger condition, a key node, and an output requirement;

[0189] The structured process instruction and the intention recognition result are processed by a preset process construction agent calling a flowchart generation engine, and a preliminary process model is generated;

[0190] The target node in the preliminary process model is bound to a target software technology component and a target data structure by a preset software flow mapping agent, and a configuration file is generated;

[0191] A target business process is deployed to a target execution environment based on the configuration file by a preset deployment agent to perform state monitoring and resource scheduling.

[0192] In an embodiment of the present application, when the processor executes the computer program, the following steps are also implemented:

[0193] The natural language business requirement is contextually semantically parsed by a preset requirement understanding agent calling a preset large language model to extract a candidate semantic unit set; the candidate semantic unit set includes multiple candidate semantic units;

[0194] The semantic recognizer based on the preset rule template and the dictionary enhancement performs semantic labeling on each candidate semantic unit, and outputs a typed intermediate result; the typed intermediate result includes a task, a process node, a constraint condition, and a trigger event;

[0195] The typed intermediate result is standardized and confirmed based on a preset domain knowledge base to generate a structured process instruction.

[0196] In an embodiment of the present application, when the processor executes the computer program, the following steps are also implemented:

[0197] Modal information corresponding to the natural language business requirement is obtained; the modal information includes at least any one of text, table, image, and voice;

[0198] The preset requirement understanding agent calls a preset large language model to determine a target modal information processing mode corresponding to the modal information;

[0199] The natural language business requirement is contextually semantically parsed based on the target modal information processing mode to extract a candidate semantic unit set.

[0200] In an embodiment of the present application, when the processor executes the computer program, the following steps are also implemented:

[0201] encoding the candidate entity in the typed intermediate result into a candidate entity vector by a preset semantic encoder, and encoding each preset terminology entity in a preset domain knowledge base into a standard terminology vector;

[0202] performing similarity calculation on the candidate entity vector and the standard terminology vector to obtain a list of similarity scores;

[0203] determining a target terminology vector matching the candidate entity vector from each standard terminology vector based on the list of similarity scores;

[0204] standardizing the typed intermediate result according to a current similarity score between the candidate entity vector and the target terminology vector, and generating a structured process instruction.

[0205] In an embodiment of the present application, the processor, when executing the computer program, further implements the following steps:

[0206] obtaining a preset similarity threshold;

[0207] if the current similarity score is greater than the preset similarity threshold, replacing the candidate entity corresponding to the candidate entity vector with a preset terminology entity corresponding to the target terminology vector, and generating the structured process instruction;

[0208] if the current similarity score is less than or equal to the preset similarity threshold, marking the preset terminology entity corresponding to the target terminology vector as an ambiguous item, and generating the structured process instruction.

[0209] In an embodiment of the present application, the processor, when executing the computer program, further implements the following steps:

[0210] calling a flowchart construction domain-specific language description file by a preset process construction intelligent agent, wherein the domain-specific language description file includes multiple types of control structures, and the multiple types include approval, condition, parallel, and rollback;

[0211] processing the structured process instruction and the intent recognition result based on the domain-specific language description file to generate a graphical flowchart;

[0212] editing the graphical flowchart through a preset visual editing interface to generate a preliminary process model.

[0213] In an embodiment of the present application, the preset control and scheduling agent respectively schedules and coordinates other agents and handles exceptions, updates the user context memory of the preset context memory agent; wherein the other agents include a preset demand understanding agent, a preset intention recognition agent, a preset context memory agent, a preset flow construction agent, a preset software flow mapping agent and a preset deployment agent; the scheduling and coordination includes managing the execution dependencies and scheduling sequences between the other agents, and the exception handling includes detecting task exception conditions of the other agents and triggering preset error recovery logic corresponding to the task exception conditions.

[0214] The computer device provided by the embodiments of the present application has similar implementation principles and technical effects to the above-mentioned method embodiments, and thus will not be described here.

[0215] In an embodiment of the present application, a computer readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the following steps are implemented:

[0216] The preset demand understanding agent processes the received natural language business demand to generate a structured flow instruction corresponding to the natural language business demand;

[0217] The preset intention recognition agent and the preset context memory agent provide a multi-round dialogue context to process the structured flow instruction to generate an intention recognition result; wherein the intention recognition result includes a business target, a trigger condition, a key node and an output requirement;

[0218] The preset flow construction agent calls a flowchart generation engine to process the structured flow instruction and the intention recognition result to generate a preliminary flow model;

[0219] The preset software flow mapping agent binds the target node in the preliminary flow model with a target software technology component and a target data structure to generate a configuration file;

[0220] The preset deployment agent deploys the target business flow to a target execution environment based on the configuration file to perform state monitoring and resource scheduling.

[0221] In an embodiment of the present application, the computer program is executed by the processor to further implement the following steps:

[0222] The preset demand understanding agent calls a preset large language model to perform context semantic analysis on the natural language business demand to extract a candidate semantic unit set; wherein the candidate semantic unit set includes a plurality of candidate semantic units;

[0223] The semantic recognizer based on the preset rule template and the dictionary enhancement performs semantic labeling on each candidate semantic unit, and outputs a typed intermediate result; wherein the typed intermediate result includes a task, a process node, a constraint condition, and a trigger event;

[0224] The typed intermediate result is standardized and confirmed based on a preset domain knowledge base, and a structured process instruction is generated.

[0225] In an embodiment of the present application, the computer program is executed by the processor to further implement the following steps:

[0226] Obtain modal information corresponding to the natural language business requirement; wherein the modal information includes at least any one of text, table, image, and voice;

[0227] The preset requirement understanding agent calls a preset large language model to determine a target modal information processing mode corresponding to the modal information;

[0228] Based on the target modal information processing mode, the natural language business requirement is contextually semantically analyzed to extract a set of candidate semantic units.

[0229] In an embodiment of the present application, the computer program is executed by the processor to further implement the following steps:

[0230] The candidate entities in the typed intermediate result are encoded into candidate entity vectors by a preset semantic encoder, and each preset term entity in the preset domain knowledge base is encoded into a standard term vector;

[0231] The similarity between the candidate entity vector and the standard term vector is calculated to obtain a list of similarity scores;

[0232] Based on the list of similarity scores, a target term vector matching the candidate entity vector is determined from each standard term vector;

[0233] The typed intermediate result is standardized and confirmed according to the current similarity score between the candidate entity vector and the target term vector, and a structured process instruction is generated.

[0234] In an embodiment of the present application, the computer program is executed by the processor to further implement the following steps:

[0235] Obtain a preset similarity threshold;

[0236] If the current similarity score is greater than the preset similarity threshold, the candidate entity corresponding to the candidate entity vector is replaced by the preset term entity corresponding to the target term vector, and a structured process instruction is generated;

[0237] If the current similarity score is less than or equal to the preset similarity threshold, the preset term entity corresponding to the target term vector is marked as an ambiguous item, and a structured flow instruction is generated.

[0238] In an embodiment of the present application, the computer program, when executed by the processor, further implements the following steps:

[0239] The agent calling flowchart construction domain-specific language description file is constructed through a preset flow;

[0240] The structured flow instruction and the intent recognition result are processed based on the domain-specific language description file to generate a graphical flowchart.

[0241] The graphical flowchart is edited through a preset visual editing interface to generate a preliminary flow model.

[0242] In an embodiment of the present application, the preset control and scheduling agent respectively schedules and coordinates other agents and handles exceptions, and updates the user context memory of the preset context memory agent; wherein the other agents include a preset demand understanding agent, a preset intent recognition agent, a preset context memory agent, a preset flow construction agent, a preset software stream mapping agent, and a preset deployment agent; the scheduling and coordination include managing the execution dependencies and scheduling sequences between the other agents, and the exception handling includes detecting task exception conditions of the other agents and triggering preset error recovery logic corresponding to the task exception conditions.

[0243] The computer readable storage medium provided in the embodiment has similar implementation principles and technical effects to the above method embodiments, and will not be described here.

[0244] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, databases, or other media in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0245] Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the features disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure that are deemed to fall within the general principles of the disclosure and include commonly known or customary practice in the art. The specification and examples are to be considered exemplary only, with the true scope and spirit of the disclosure being indicated by the following claims.

[0246] It should be understood that the present disclosure is not limited to the precise structures herein described and illustrated in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the claims that follow.

Claims

1. An automatic business process generation method based on fuzzy requirement semantic parsing, characterized in that, The method includes: The system processes received natural language business requirements through a pre-defined requirement understanding agent to generate structured process instructions corresponding to those requirements. This includes: the agent invoking a pre-defined large language model to perform contextual semantic parsing on the natural language business requirements, extracting a set of candidate semantic units; wherein the set of candidate semantic units includes multiple candidate semantic units; semantically labeling each candidate semantic unit based on a pre-defined rule template and a dictionary-enhanced semantic recognizer, outputting a typified intermediate result; wherein the typified intermediate result includes tasks, process nodes, constraints, and triggering events; and standardizing and confirming the typified intermediate result based on a pre-defined domain knowledge base to generate the structured process instructions. The structured process instructions are processed by a pre-set intent recognition agent and a pre-set context memory agent to generate intent recognition results; wherein, the intent recognition results include business objectives, triggering conditions, key nodes and output requirements. The intelligent agent is constructed by constructing a pre-defined process and calling a flowchart generation engine to process the structured process instructions and the intent recognition results to generate a preliminary process model. By using a pre-defined software flow mapping agent, the target nodes in the preliminary process model are bound to the target software technology components and the target data structure to generate a configuration file; By pre-deploying intelligent agents based on the configuration file, the target business process is deployed to the target execution environment for execution status monitoring and resource scheduling.

2. The method of claim 1, wherein, The process involves the intelligent agent, through which the preset requirements are understood, invoking a preset large language model to perform contextual semantic analysis on the natural language business requirements, and extracting a set of candidate semantic units, including: Obtain modal information corresponding to the natural language service requirements; wherein, the modal information includes at least one of text, tables, images, and speech; The intelligent agent, based on the preset requirements, calls a preset large language model to determine the target modality information processing method corresponding to the modality information. Based on the target modality information processing method, the natural language business requirements are analyzed using contextual semantic parsing to extract the candidate semantic unit set.

3. The method of claim 2, wherein, The standardization and confirmation of the typed intermediate results based on a preset domain knowledge base, and the generation of the structured process instructions, include: The candidate entities in the typed intermediate results are encoded into candidate entity vectors by a preset semantic encoder, and each preset term entity in the preset domain knowledge base is encoded into a standard term vector. The similarity between the candidate entity vector and the standard term vector is calculated to obtain a list of similarity scores; Based on the similarity score list, a target term vector that matches the candidate entity vector is determined from each of the standard term vectors; The typified intermediate results are standardized and confirmed based on the current similarity score between the candidate entity vector and the target term vector, and the structured process instructions are generated.

4. The method of claim 3, wherein, The standardization confirmation of the typed intermediate result according to the current similarity score between the candidate entity vector and the target term vector generates the structured process instruction, including: obtaining a preset similarity threshold; if the current similarity score is greater than the preset similarity threshold, replacing the candidate entity corresponding to the candidate entity vector with a preset term entity corresponding to the target term vector, and generating the structured process instruction; if the current similarity score is less than or equal to the preset similarity threshold, marking the preset term entity corresponding to the target term vector as an ambiguous item, and generating the structured process instruction.

5. The method of claim 1, wherein, The preliminary process model is generated by processing the structured process instruction and the intent recognition result through a preset process construction agent calling a process graph generation engine, including: The domain language description file is constructed through the preset process construction agent calling the process graph construction domain language description file; wherein the domain language description file includes multiple types of control structures, and the multiple types include approval, condition, parallel, rollback; The structured process instruction and the intent recognition result are processed based on the domain language description file to generate a graphical process graph; The graphical process graph is edited and processed through a preset visual editing interface to generate the preliminary process model.

6. The method of claim 1, wherein, The method further includes: The user context memory of the preset context memory agent is updated by scheduling and coordinating other agents and handling exceptions through a preset control and scheduling agent; wherein the other agents include the preset intent recognition agent, the preset context memory agent, the preset process construction agent, the preset software stream mapping agent and the preset deployment agent; the scheduling and coordination includes managing the execution dependency and scheduling order between the other agents, and the exception handling includes detecting the task exception of the other agents and triggering the preset error recovery logic corresponding to the task exception.

7. An apparatus for automatic business process generation based on fuzzy requirement semantic parsing, characterized in that, The device includes: The demand understanding module is used for processing the received natural language business demand through a preset demand understanding agent to generate a structured process instruction corresponding to the natural language business demand, including: performing context semantic analysis on the natural language business demand through the preset demand understanding agent calling a preset large language model to extract a candidate semantic unit set; wherein the candidate semantic unit set includes multiple candidate semantic units; performing semantic labeling on each candidate semantic unit based on a semantic recognizer enhanced by a preset rule template and a dictionary to output a typed intermediate result; wherein the typed intermediate result includes tasks, process nodes, constraint conditions and trigger events; performing standardization confirmation on the typed intermediate result based on a preset domain knowledge base to generate the structured process instruction; An intent recognition module is configured to process the structured flow instruction based on a multi-round dialogue context provided by a preset intent recognition agent and a preset context memory agent, and generate an intent recognition result, wherein the intent recognition result includes a business target, a trigger condition, a key node, and an output requirement. A flow construction module is configured to process the structured flow instruction and the intent recognition result based on a flow graph generation engine invoked by a preset flow construction agent, and generate a preliminary flow model. A software stream mapping module is configured to bind a target node in the preliminary flow model with a target software technology component and a target data structure based on a preset software stream mapping agent, and generate a configuration file. A deployment module is configured to deploy a target business flow to a target execution environment based on the configuration file by a preset deployment agent, and perform state monitoring and resource scheduling.

8. An electronic device, comprising: The electronic device includes a processor and a memory, and the memory stores at least one instruction, at least one program, a code set, or an instruction set, which are loaded and executed by the processor to implement the automatic business flow generation method based on fuzzy requirement semantic parsing according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, a code set, or an instruction set, which are loaded and executed by the processor to implement the automatic business flow generation method based on fuzzy requirement semantic parsing according to any one of claims 1-6.

Images