A method and apparatus for initiating dynamic processes based on agent orchestration
By using intelligent agent orchestration technology to automatically identify and assemble process data in collaborative office systems, the problem of users having to fill out a lot of forms in chat-style office work has been solved, realizing the automatic generation and automatic confirmation of process handling, thus improving efficiency.
Patent Information
- Application Number
- CN202611249269.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-17
AI Technical Summary
Existing collaborative office systems struggle to automatically generate business data that meets process requirements based on user input in chat-style office scenarios, resulting in users having to manually fill out and repeatedly confirm a large amount of information during the process, leading to low efficiency.
By using intelligent agent orchestration technology, user messages are obtained and process instructions are identified. Structured data items are extracted, forms are dynamically assembled, and rendered as embedded confirmation cards, reducing manual operations.
It enables the automatic generation of process processing data and the automatic determination of approval information in chat-based office scenarios, improving the automation level and interaction efficiency of process initiation.
Smart Images

Figure CN123117747A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of multi-agent systems, and more particularly to a method and apparatus for initiating dynamic processes based on agent orchestration. Background Technology
[0002] With the continuous development of large language models, intelligent agent orchestration technology, and collaborative office platforms, intelligent office based on natural language interaction is gradually becoming an important development direction for enterprise digital office. By combining natural language understanding, multimodal information extraction, business process orchestration, and collaborative office systems, users can directly use chat windows to complete various office operations such as business consultation, process initiation, and approval inquiries, thereby reducing the interactive complexity brought by traditional menu-based operations and improving office efficiency and intelligence.
[0003] Existing collaborative office systems typically handle business processes by pre-configuring business workflows, electronic forms, and approval rules, effectively enabling the information management of daily approval processes within enterprises. However, in chat-style office scenarios, due to the free-expression, incomplete information, and non-standard business descriptions inherent in user input, existing collaborative office systems struggle to automatically generate business data that meets the requirements for workflow processing based on user input, and to accurately complete information confirmation before workflow initiation. This results in users still needing to manually fill out and repeatedly confirm information during the workflow process, leading to low workflow efficiency.
[0004] Therefore, there is an urgent need for a dynamic process initiation method and device based on agent orchestration. Summary of the Invention
[0005] This application provides a dynamic process initiation method and apparatus based on intelligent agent orchestration, which solves the problem that existing collaborative office systems are unable to automatically generate business data that meets the process handling requirements based on user input and accurately complete the information confirmation before process initiation, resulting in users still needing to fill in a lot of manual information and repeatedly confirm during the process handling, leading to low process handling efficiency.
[0006] The first aspect of this application provides a dynamic process initiation method based on intelligent agent orchestration. The method includes: acquiring a user message sent by a target user and identifying whether the user message contains a process processing instruction; if it is confirmed that the user message contains a process processing instruction, acquiring the process type result corresponding to the process processing instruction through an intent recognition intelligent agent; based on the process type result, retrieving a process template that meets the target matching degree from a process template library and outputting the process template ID and basic process information; extracting structured information from the user message through a multimodal data extraction intelligent agent and outputting a list of structured data items; based on the process template ID, basic process information, and list of structured data items, acquiring the dynamic form component definition corresponding to the process template ID through a dynamic form assembly intelligent agent, and performing field mapping, type conversion, and data filling on the list of structured data items and the dynamic form component definition to obtain a form data object; based on the process template ID and the form data object, outputting the next node approval information through a flowchart parsing intelligent agent; rendering the form data object and the next node approval information as an embedded confirmation card and sending it to a chat window for the target user to confirm or modify.
[0007] Optionally, the process involves acquiring user messages sent by the target user and identifying whether the user messages contain process processing instructions. Specifically, this includes: inputting the process processing instructions into a natural language model and outputting the intent type corresponding to the user message through the natural language model; the intent type includes business submission initiation type, consultation type, data query type, approval query type, and ordinary chat type; when the intent type is any one of business submission initiation type, consultation type, data query type, or approval query type, it is confirmed that the user message contains process processing instructions; when the intent type is ordinary chat type, it is confirmed that the user message does not contain process processing instructions.
[0008] Optionally, the multimodal data extraction agent includes a text extraction agent, an image extraction agent, and a file extraction agent. The multimodal data extraction agent extracts structured information from user messages and outputs a list of structured data items. Specifically, this includes: using the text extraction agent to perform entity recognition and attribute extraction on the text content of user messages to obtain structured text data items; using the image extraction agent to perform optical character recognition and image content understanding on the image content of user messages to obtain structured image data items; using the file extraction agent to perform structured parsing on the file content of user messages to obtain structured file data items; and performing unified format conversion and integration processing on the text structured data items, image structured data items, and file structured data items to output a list of structured data items.
[0009] Optionally, based on the process template ID, basic process information, and a list of structured data items, a dynamic form assembly agent obtains the dynamic form component definition corresponding to the process template ID. Field mapping, type conversion, and data filling are then performed between the list of structured data items and the dynamic form component definition. Specifically, this includes: obtaining the dynamic form component definition based on the process template ID; the dynamic form component definition includes field names, field codes, field types, whether fields are required, and option data sources; constructing a field mapping relationship between the list of structured data items and the dynamic form component definition based on a multi-factor similarity model, and outputting the field mapping result corresponding to each structured data item; The field mapping result includes successful mapping results and failed mapping results. If the field mapping result is a successful mapping result and the corresponding field is not a multi-value field, then data type validation, data type conversion, and data filling are performed on the corresponding field. If the field mapping result is a successful mapping result and the corresponding dynamic form component is a multi-value field, then data filling is performed on the corresponding field in array form. If the field mapping result is a failed mapping result and the corresponding dynamic form component is a required field, then the corresponding field is marked as a field to be supplemented, and intelligent recommendation values are generated. Based on the data filling results corresponding to each dynamic form component, the form is assembled to obtain a form data object.
[0010] Optionally, a multi-factor similarity model is constructed, specifically including: constructing a semantic vector similarity factor based on the semantic association degree between each structured data item and each dynamic form component; constructing an ontology alignment factor based on the business ontology association relationship between each structured data item and each dynamic form component; constructing a type compatibility factor based on the compatibility relationship between the data type of each structured data item and the field type of each dynamic form component; constructing a context co-occurrence factor based on the historical mapping association relationship between each structured data item and each dynamic form component; constructing an entity parsing factor based on the matching relationship between the entity parsing result corresponding to each structured data item and the entity type corresponding to each dynamic form component; defining multi-factor similarity calculation rules based on the semantic vector similarity factor, ontology alignment factor, type compatibility factor, context co-occurrence factor, and entity parsing factor, and constructing a multi-factor similarity model.
[0011] Optionally, based on the process template ID and form data object, the flowchart parsing agent outputs the approval information for the next node. Specifically, this includes: obtaining the corresponding flowchart definition file based on the process template ID, parsing the flowchart definition file, and determining the process start node; determining the next node based on the form data object and the node exit conditions corresponding to the process start node; parsing the approver configuration rules corresponding to the next node, and determining the approver candidate list corresponding to the next node in conjunction with organizational structure data; the approver configuration rules include rules for specifying roles, rules for specifying personnel, rules for superiors, and dynamic calculation rules; if the approver candidate list includes multiple approvers, obtaining the process step configuration method corresponding to the next node, and determining the corresponding approval processing method based on the process step configuration method; the process step configuration methods include parallel countersigning and serial countersigning; and generating the approval information for the next node based on the approver candidate list and the approval processing method.
[0012] Optionally, after rendering the form data object and the next approval step information as an embedded confirmation card and sending it to the chat window for the target user to confirm or modify, the method further includes: in response to receiving the target user's confirmation instruction for the embedded confirmation card, submitting the confirmed form data object and the confirmed next approval step information to the collaborative office system to create a process instance and return the process initiation result.
[0013] A second aspect of this application provides a dynamic process initiation device based on intelligent agent orchestration. The device includes an acquisition module and a processing module, wherein... The acquisition module is used to acquire user messages sent by the target user and identify whether the user messages contain process instructions.
[0014] The acquisition module is used to obtain the process type result corresponding to the process processing instruction if it is confirmed that the user message contains a process processing instruction, through the intent recognition agent.
[0015] The acquisition module is used to retrieve process templates that meet the target matching degree from the process template library based on the process type result, and output the process template ID and basic process information.
[0016] The acquisition module is used to extract structured information from user messages through a multimodal data extraction agent and output a list of structured data items.
[0017] The processing module is used to obtain the dynamic form component definition corresponding to the process template ID through a dynamic form assembly agent based on the process template ID, basic process information, and a list of structured data items. It then performs field mapping, type conversion, and data filling on the list of structured data items and the dynamic form component definition to obtain a form data object.
[0018] The processing module is used to output the next approval step information through the flowchart parsing agent based on the process template ID and form data object.
[0019] The processing module renders the form data object and the next approval step information into an embedded confirmation card and sends it to the chat window for the target user to confirm or modify.
[0020] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described above.
[0021] A fourth aspect of this application provides a non-transitory computer-readable storage medium storing a computer program, the computer program being executed by a processor using any of the methods described above.
[0022] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By acquiring target user messages and identifying whether they contain process processing instructions, if so, the process type result is obtained through the intent recognition agent, the corresponding process template is retrieved, and the process template ID and basic process information are output. The structured data item list is output by the multimodal data extraction agent. The form data object is generated by the dynamic form assembly agent. The next node approval information is output by the flowchart parsing agent and rendered as an embedded confirmation card, which is sent to the chat window for the target user to confirm or modify. This achieves automatic generation of process processing data and automatic determination of approval information in chat-style office scenarios, reducing manual filling and repeated confirmation operations by users, and improving the automation level and interaction efficiency of process initiation.
[0023] 2. Input the process processing instructions into the natural language model, and output the intent type corresponding to the user message through the natural language model; when the intent type is any one of the business submission initiation type, consultation type, data query type, or approval query type, confirm that the user message contains process processing instructions; when the intent type is ordinary chat type, confirm that the user message does not contain process processing instructions, thereby accurately identifying user messages that need to be processed and filtering non-business messages, providing a reliable data foundation for the automatic initiation of subsequent processes.
[0024] 3. Obtain the dynamic form component definition based on the process template ID; construct the field mapping relationship between the structured data item list and the dynamic form component definition based on the multi-factor similarity model, and output the field mapping result corresponding to each structured data item; if the field mapping result is a successful mapping result and the corresponding field is not a multi-value field, perform data type validation, data type conversion, and data filling processing on the corresponding field; if the field mapping result is a successful mapping result and the corresponding dynamic form component is a multi-value field, perform data filling processing on the corresponding field in array form; if the field mapping result is a failed mapping result and the corresponding dynamic form component is a required field, mark the corresponding field as a field to be supplemented, and generate intelligent recommendation values; assemble the form based on the data filling results corresponding to each dynamic form component to obtain the form data object, thereby realizing the automatic conversion of user business data to process form data, reducing manual filling and repetitive supplementation, and improving the automation level of process initiation. Attached Figure Description
[0025] Figure 1 This is a flowchart illustrating a dynamic process initiation method based on intelligent agent orchestration provided in an embodiment of this application; Figure 2 This is a schematic diagram of a module of a dynamic process initiation device based on intelligent agent orchestration provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.
[0026] Explanation of reference numerals in the attached figures: 21. Acquisition module; 22. Processing module; 301. Processor; 302. Communication bus; 303. User interface; 304. Network interface; 305. Memory. Detailed Implementation
[0027] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.
[0028] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.
[0029] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.
[0030] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.
[0031] Please refer to Figure 1 The diagram illustrates a flowchart of a dynamic process initiation method based on intelligent agent orchestration provided in an embodiment of this application. The flowchart mainly includes the following steps: S101 to S107.
[0032] Step S101: Obtain the user message sent by the target user and identify whether the user message contains process processing instructions.
[0033] Specifically, the system acquires user messages sent by the target user, including text messages, image messages, and file messages; it then inputs these user messages into a natural language model for semantic analysis to identify the intent type corresponding to the user message; the intent types include business submission initiation type, consultation type, data query type, approval query type, and ordinary chat type; when the intent type is any one of the business submission initiation type, consultation type, data query type, or approval query type, it confirms that the user message contains process processing instructions; when the intent type is ordinary chat type, it confirms that the user message does not contain process processing instructions.
[0034] In one possible implementation, step S101 further includes: inputting the process processing instruction into the natural language model, and outputting the intent type corresponding to the user message through the natural language model; the intent type includes business submission initiation type, consultation type, data query type, approval query type, and ordinary chat type; when the intent type is any one of business submission initiation type, consultation type, data query type, or approval query type, confirming that the user message contains the process processing instruction; when the intent type is ordinary chat type, confirming that the user message does not contain the process processing instruction.
[0035] Specifically, the natural language model is constructed using a large language model based on the Transformer architecture. It combines historical dialogue data, process initiation data, process approval data, process query data, and general dialogue data from the collaborative office domain to build training corpus. The pre-trained model is trained through supervised fine-tuning, enabling the natural language model to learn the semantic expression features and corresponding intent classification rules in different office scenarios. It is then deployed in the intent recognition agent to perform semantic understanding and process type recognition of user messages.
[0036] In the specific identification process, the user message corresponding to the process handling instruction is sent to the natural language model (NLP), and preset intent recognition prompts are input into the NLP to constrain the NLP to perform semantic reasoning according to a unified intent classification rule. The prompts instruct the NLP, acting as an intent recognition expert, to analyze the user message, determine the intent type corresponding to the user message, and extract the target process type corresponding to the user message. Intent types include business submission initiation, consultation, data query, approval query, and general chat. Specifically, a business submission initiation indicates that the user wishes to initiate, apply for, or submit a business process, thereby causing a change in the business status; a consultation indicates that the user is consulting about knowledge, systems, policies, or operational methods; a data query indicates that the user is querying business data, business status, or historical records; an approval query indicates that the user is querying the approval status, progress, or history of an existing approval process; and general chat indicates greetings, casual conversation, or communication unrelated to the business process.
[0037] After completing semantic reasoning, the natural language model outputs the intent type and target process type in a structured manner. For example, when the user message is "I want to request leave," the natural language model identifies the intent type as a business submission initiation type and extracts the corresponding target process type. When the user message is "What materials are needed for leave?", the natural language model identifies the intent type as an inquiry type. When the user message is "Check my attendance," the natural language model identifies the intent type as a data query type. When the user message is "Has my leave been approved?", the natural language model identifies the intent type as an approval query type. When the user message is a greeting or casual conversation, the natural language model identifies the intent type as a regular chat type. Furthermore, the natural language model encapsulates the recognition results according to a preset data format, such as outputting structured data containing intent type and target process type fields. When a user is identified as having a process initiation requirement, the model can output the intent type as a business submission initiation type and the corresponding target process type, such as a travel expense reimbursement process or a leave request process, providing a basis for subsequent process template matching.
[0038] After obtaining the structured recognition results, the system determines whether the user message contains process processing instructions based on the output intent type. If the intent type is any of the following: business submission initiation type, consultation type, data query type, or approval query type, the system confirms that the user message contains process processing instructions and proceeds to the subsequent intelligent agent orchestration process. If the intent type is a normal chat type, the system confirms that the user message does not contain process processing instructions and responds as a normal chat message.
[0039] Step S102: If it is confirmed that the user message contains a process processing instruction, the process type result corresponding to the process processing instruction is obtained through the intent recognition agent.
[0040] Specifically, the intent recognition agent semantically parses the business requirements expressed by the user based on the business objects, business behaviors, and business scenarios in the user message, identifies the process type corresponding to the user's current business requirement, and generates a process type result. The process type result describes the business process category corresponding to the user's request, so that a matching process template can be retrieved from the process template library later.
[0041] When the user message contains a clear process name, the intent recognition agent directly extracts the process name as the process type result. For example, when the user message is "Help me initiate a travel expense reimbursement process", the travel expense reimbursement process is directly identified as the process type result; when the user message is "Submit a leave application", the leave application process is directly identified as the process type result.
[0042] For user messages that do not contain explicit process names but only describe business matters, the intent is to identify the intelligent agent by combining the business semantics in the user message to infer the process type. For example, when the user message is "I want to reimburse the expenses incurred from last week's business trip," the corresponding process type is identified as a travel expense reimbursement process; when the user message is "I want to apply to purchase an office computer," the corresponding process type is identified as a purchase application process; and when the user message is "I want to take three days off," the corresponding process type is identified as a leave application process.
[0043] To improve the accuracy of process type identification, the intent recognition agent maps abbreviations, aliases, synonyms, or colloquial expressions in user messages to corresponding standard process types through semantic normalization. For example, different expressions such as "reimburse business trip expenses," "travel expense reimbursement," and "business trip reimbursement" are uniformly identified as a travel expense reimbursement process, and different expressions such as "request annual leave," "apply for leave," and "I want to rest for a few days" are uniformly identified as a leave request process. This results in a uniformly formatted process type and is sent to the process matching agent for subsequent process template matching.
[0044] Step S103: Based on the process type result, retrieve the process template that meets the target matching degree from the process template library, and output the process template ID and basic process information.
[0045] Specifically, the process template library stores various pre-configured business process templates in the collaborative office system. Each process template corresponds to a unique process template identifier and is associated with basic process information. The basic process information includes the process name, scope of application, process category, process description, and dynamic form definition information corresponding to the process template ID. This information describes the basic attributes of the process template and provides basic data for subsequent dynamic form assembly.
[0046] Upon receiving the process type result, the process matching agent receives the result and generates a process retrieval request based on it. This request is then sent to the process template library, where candidate process templates matching the process type result are searched. Each process template in the library has a pre-defined process type index. Upon receiving a retrieval request, the agent first filters candidate process templates associated with the process type result based on the process type index, thus narrowing the search scope and improving retrieval efficiency.
[0047] For each candidate process template obtained, the process matching agent extracts the process type description information corresponding to each candidate process template ID and performs semantic matching with the process type result. It calculates the matching degree between the process type result and each candidate process template, constructs process template matching results based on the matching degrees, and selects the target process template that meets the target matching degree as the process template corresponding to the user's current business. The target matching degree represents the degree of matching between the candidate process template and the process type result that meets preset matching conditions. A preset matching threshold or the maximum matching degree can be used as the filtering criterion to ensure that the selected process template is consistent with the user's business requirements.
[0048] For example, when the process type result is a travel expense reimbursement process, the process matching agent retrieves multiple candidate process templates related to reimbursement business from the process template library, including travel expense reimbursement process, daily expense reimbursement process, and business reception reimbursement process, and calculates the matching degree between each candidate process template and the process type result. Finally, the travel expense reimbursement process with a matching degree that meets the target matching degree is selected as the target process template; when the process type result is a leave application process, the leave application process template is retrieved as the target process template.
[0049] After determining the target process template, the process matching agent reads the process template identifier and basic process information corresponding to the target process template ID, and outputs the process template identifier and basic process information. The process template identifier uniquely identifies the target process template, allowing the subsequent dynamic form assembly agent to obtain the corresponding dynamic form component definition; the basic process information describes the basic business attributes of the target process template and serves as the foundational data for subsequent dynamic form assembly and flowchart parsing.
[0050] Step S104: The multimodal data extraction agent extracts structured information from user messages and outputs a list of structured data items.
[0051] Specifically, the multimodal data extraction agent performs parallel extraction and structuring of business information from text, images, and files based on the data composition of user messages, and merges the extraction results of each modality to output a structured list of data items in a unified format, providing a data foundation for subsequent dynamic form assembly and field mapping.
[0052] In one possible implementation, step S104 further includes: performing entity recognition and attribute extraction on the text content in the user message using a text extraction agent to obtain text structured data items; performing optical character recognition and image content understanding on the image content in the user message using an image extraction agent to obtain image structured data items; performing structured parsing on the file content in the user message using a file extraction agent to obtain file structured data items; and performing unified format conversion and integration processing on the text structured data items, image structured data items, and file structured data items to output a list of structured data items.
[0053] Specifically, the multimodal data extraction agent includes text extraction sub-agents, image extraction sub-agents, and file extraction sub-agents. The agent orchestration engine schedules each sub-agent in parallel to perform information extraction tasks based on the data types contained in the user message. When the user message contains only one type of data, the corresponding sub-agent is scheduled to perform information extraction; when the user message contains multiple data types such as text, images, and files, multiple corresponding sub-agents are invoked in parallel to complete information extraction, thereby improving the processing efficiency of multimodal data.
[0054] The text extraction sub-agent performs natural language parsing on the text content of user messages, identifies entity and attribute information in the text, and generates corresponding structured text data. Entity information includes business entities such as process type, personnel name, department name, project name, and expense type; attribute information includes attributes describing business entities such as amount, date, time, location, quantity, and approval comments.
[0055] The image extraction sub-agent performs optical character recognition and image content understanding processing on images in user messages, converting text information and business objects in the images into structured data. For scenarios containing business images such as tickets, invoices, and certificates, it first identifies the text information in the image, then combines the ticket format and business semantics to extract the corresponding data items. For example, when a user message contains a photo of a train ticket for travel expense reimbursement, it can identify information such as the travel date, departure station, arrival station, ticket price, insurance fee, and total price, and convert it into corresponding structured data items.
[0056] The document extraction sub-agent performs structured parsing on the business documents in the user message, reading the file content by calling the corresponding parsing strategy according to the file format. Specifically, for spreadsheet files, it extracts the data content corresponding to each row and column; for text documents, it extracts chapter content and business fields; and for portable documents, it extracts text and table information, all of which are then uniformly converted into structured data. For example, when a user message contains a spreadsheet detailing travel expenses, it can parse out data items such as expense type, expense amount, date of occurrence, location of occurrence, and reason for expense, forming the corresponding structured data for the file.
[0057] After extracting information from each sub-agent, the multimodal data extraction agent performs unified encoding and format standardization on text structured data, image structured data, and file structured data, eliminating data format differences between different data sources. It then integrates these data according to a unified data item description format to form a list of structured data items. Each data item in the list includes at least a data item name and a data item value. When a corresponding data item contains a unit, source type, or additional attributes, the agent synchronously records the corresponding unit information, source information, and attribute information, providing a unified data input for the subsequent dynamic form assembly agent to perform field mapping, type conversion, and data filling.
[0058] Step S105: Based on the process template ID, basic process information, and structured data item list, obtain the dynamic form component definition corresponding to the process template ID through the dynamic form assembly agent, and perform field mapping, type conversion, and data filling on the structured data item list and the dynamic form component definition to obtain the form data object.
[0059] Specifically, the dynamic form assembly agent receives the process template identifier, basic process information, and a list of structured data items. Based on the process template identifier, it retrieves the dynamic form component definition for the corresponding process template. The dynamic form component definition describes the form structure corresponding to the target process, including configuration information such as the field name, field code, field type, whether it is required, and option data source for each form field, thereby establishing the association between the process template and the dynamic form.
[0060] After obtaining the dynamic form component definitions, the dynamic form assembly agent calculates the matching relationship between structured data items and each dynamic form component based on a multi-factor similarity model, and maps each structured data item to the corresponding dynamic form component according to the matching results. Specifically, for successfully mapped dynamic form components, data type validation, data type conversion, and data filling are performed based on the corresponding field type; for multi-value dynamic form components, data filling is completed in array form; for dynamic form components that are not successfully mapped and have corresponding required fields, they are marked as fields to be supplemented, and corresponding intelligent recommendation values are generated based on basic process information and user message context.
[0061] After processing the data of each dynamic form component, the dynamic form assembly agent assembles the processing results of each component into a form data object. This form data object is used to uniformly describe the form content corresponding to the target process and serves as the input data for the subsequent flowchart parsing agent to identify the approval path and approver.
[0062] In one possible implementation, step S105 further includes: obtaining the dynamic form component definition based on the process template ID; the dynamic form component definition includes field name, field code, field type, whether it is required, and option data source; constructing a field mapping relationship between the structured data item list and the dynamic form component definition based on a multi-factor similarity model, and outputting the field mapping result corresponding to each structured data item; the field mapping result includes mapping success result and mapping failure result; if the field mapping result is a mapping success result and the corresponding field is not a multi-value field, then performing data type validation processing, data type conversion processing, and data filling processing on the corresponding field; if the field mapping result is a mapping success result and the corresponding dynamic form component is a multi-value field, then performing data filling processing on the corresponding field in array form; if the field mapping result is a mapping failure result and the corresponding dynamic form component is a required field, then marking the corresponding field as a field to be supplemented and generating intelligent recommendation values; assembling the form based on the data filling results corresponding to each dynamic form component to obtain a form data object.
[0063] Specifically, the dynamic form assembly agent first retrieves the corresponding dynamic form component definition from the collaborative office system based on the process template ID. The dynamic form component definition describes the form structure corresponding to the target process, including configuration information such as field names, field codes, field types, whether a field is required, and option data sources. Specifically, the field name describes the business meaning of the form field; the field code uniquely identifies the corresponding field; the field type describes the data type of the field; whether a field is required indicates whether the corresponding field must be filled in; and the option data source describes the optional data source for components such as dropdown lists, radio buttons, or checkboxes. By retrieving the dynamic form component definition, the dynamic form assembly agent can obtain the form structure information corresponding to the target process, providing a basis for subsequent field mapping.
[0064] Subsequently, the dynamic form assembly agent, based on the list of structured data items and the definition of dynamic form components, uses a multi-factor similarity model to calculate the matching relationship between each structured data item and each dynamic form component, and outputs the corresponding field mapping results. The multi-factor similarity model comprehensively considers the semantic association, business association, and data feature consistency between the structured data items and dynamic form components, matching each structured data item with the most suitable dynamic form component, and dividing the field mapping results into successful mapping results and unsuccessful mapping results. A successful mapping result indicates that a valid mapping relationship can be established between the structured data item and the corresponding dynamic form component; a unsuccessful mapping result indicates that no dynamic form component matching the structured data item was found.
[0065] For cases where the field mapping result is successful and the corresponding field is not a multi-value field, the dynamic form assembly agent first performs data type validation on the corresponding field to determine whether the data type of the structured data item meets the data type requirements of the corresponding field. When the data types match, data filling is performed directly; when the data types do not match but type conversion can be performed, the corresponding data type conversion is performed before data filling is completed. For example, a text-based amount is converted to a numeric type, and a date string is converted to a standard date format before being written to the corresponding field.
[0066] When the field mapping result is successful and the corresponding dynamic form component is a multi-value field, the dynamic form assembly agent organizes multiple structured data items into an array based on the characteristic that the corresponding field supports multiple data values, and then fills them into the corresponding field sequentially. For example, when the expense details field contains multiple expense records, each expense record is written into the expense details field as an array element; when the invoice information field corresponds to multiple invoices, the information of each invoice is uniformly filled into an array to ensure the complete preservation of multi-value data.
[0067] When a field mapping fails and the corresponding field in the dynamic form component is a required field, the dynamic form assembly agent marks the corresponding field as a field to be supplemented and generates a corresponding intelligent recommendation value based on the basic process information and the contextual semantics in the user message. The field to be supplemented serves as a notification to the user that the corresponding information is incomplete; the intelligent recommendation value is used to infer the most likely field value based on existing business information. For example, when the reimbursement type is a required field but the user has not explicitly filled it in, travel expense reimbursement can be recommended as a candidate value based on the business trip semantics in the user message; when the start and end dates are required fields but the user only describes "last week," the corresponding time range can be inferred based on the current time, and the inferred result can be used as a recommendation value for the user's confirmation.
[0068] After the above processing is completed, the dynamic form assembly agent performs form assembly processing based on the data filling results corresponding to each dynamic form component, organizing each field into a unified form data object according to the definition of the dynamic form component. The form data object includes the field value and the corresponding filling status of each field. Among them, the fields that have been filled have stored the corresponding data content, and the fields to be filled have stored the to be filled identifier and intelligent recommendation value, thus forming a complete process form data, providing a unified data foundation for the subsequent flowchart parsing agent to identify the approval path and approver.
[0069] In one possible implementation, step S105 further includes: constructing a semantic vector similarity factor based on the semantic association degree between each structured data item and each dynamic form component; constructing an ontology alignment factor based on the business ontology association relationship between each structured data item and each dynamic form component; constructing a type compatibility factor based on the compatibility relationship between the data type of each structured data item and the field type of each dynamic form component; constructing a context co-occurrence factor based on the historical mapping association relationship between each structured data item and each dynamic form component; constructing an entity parsing factor based on the matching relationship between the entity parsing result corresponding to each structured data item and the entity type corresponding to each dynamic form component; defining multi-factor similarity calculation rules based on the semantic vector similarity factor, ontology alignment factor, type compatibility factor, context co-occurrence factor, and entity parsing factor, and constructing a multi-factor similarity model.
[0070] Specifically, a multi-factor similarity model is used to calculate the overall matching degree between structured data items and dynamic form components to improve the accuracy of field mapping. For the i-th structured data item and the j-th dynamic form component, the corresponding overall alignment score is calculated using the following formula:
[0071] in, This represents the overall alignment score between the i-th structured data item and the j-th dynamic form component. A higher overall alignment score indicates a higher degree of matching between the two. This represents the i-th structured data item; This represents the j-th dynamic form component; This represents the Sigmoid activation function, which maps the weighted calculation results of multiple similarity factors to between 0 and 1, thereby suppressing the impact of abnormally high or low values on the final matching results. This represents the weight parameter of the k-th similarity factor. All weight parameters satisfy the condition that the sum of their weights is 1. This parameter is used to adjust the degree of influence of different similarity factors on the overall alignment score. This represents the calculation result of the k-th similarity factor; This represents the positional prior decay factor, used to reflect the degree of consistency between the arrangement order of structured data items and the arrangement order of dynamic form components; This represents the required field gain coefficient, used to increase the matching priority of the dynamic form component corresponding to the required field; This function indicates whether a field is required. It takes the value 1 when the corresponding dynamic form component is a required field, and 0 otherwise. This represents the entity parsing gating factor, used to further adjust the overall alignment score based on the entity parsing results.
[0072] The semantic vector similarity factor is used to measure the semantic similarity between the names of structured data items and the names of dynamic form components. Its calculation formula is as follows: in, This represents the semantic vector similarity factor. A semantic vector representing the name of a structured data item; This represents the semantic vector corresponding to the name of the dynamic form component; both semantic vectors are generated by the Sentence-BERT or BGE model. The cosine similarity between two semantic vectors is represented by a value ranging from -1 to 1. By adding 1 and then dividing by 2, the cosine similarity is normalized to between 0 and 1. This represents a temperature compression parameter used to control the distribution of semantic similarity. As the number of samples increases, the difference between highly similar and low-similarity samples becomes even stronger. This calculation method enables data items with similar business semantics to obtain higher semantic matching scores.
[0073] The ontology alignment factor is used to integrate business domain knowledge and improve the recognition capabilities of business codes, business aliases, and business abbreviations. Its calculation formula is as follows: in, Indicates the ontology alignment factor; This indicates the exact match between the structured data item code and the dynamic form component code. The value is 1 when the two are completely identical, and 0 otherwise. This indicates the Jaccard character matching results between data item names and field names, used to measure the degree of character overlap between short texts; This indicates the business alias matching results, which are determined by querying the business alias dictionary for the corresponding field to determine if a synonym exists. This indicates the matching results for business abbreviations, used to support matching of abbreviations such as the first letter of Pinyin and English abbreviations; finally, the maximum value among the four matching methods is selected as the ontology alignment factor to fully utilize business knowledge and improve field matching accuracy.
[0074] The type compatibility factor describes the compatibility between the data type of a structured data item and the field type of a dynamic form component. When the data types are completely identical, the type compatibility factor is 1. When the mapping can be completed through type conversion, the type compatibility factor is 0.5 (e.g., string to numeric type, date string to date object). When the corresponding field is an enumeration type and the value of the structured data item exists in the corresponding option data source, the score is increased by 0.4. When the data types are incompatible, the type compatibility factor is 0. In this case, the edge weight between the corresponding data item and the dynamic form component is directly set to negative infinity, and it is directly excluded from the field mapping process, thus avoiding cross-type mismapping.
[0075] The context co-occurrence factor is used to reflect the field mapping patterns in historical process instances, and its calculation formula is as follows: in, Indicates the context co-occurrence factor; This indicates the cumulative number of times that structured data items in the historical process instance have been successfully mapped to the corresponding dynamic form components; This indicates the cumulative number of times the corresponding dynamic form component has been filled; the historical mapping frequency is compressed using a logarithmic function to reduce the bias caused by high-frequency mapping, so that the historical mapping relationship can reasonably participate in the current field mapping calculation.
[0076] The entity resolution factor is used to process dynamic form components with entity type identifiers, and its calculation formula is as follows: in, Represents the entity resolution factor; This represents the entity resolution confidence score output by the entity resolution service. When the entity type corresponding to a structured data item matches the entity type corresponding to a dynamic form component and entity resolution is successful, the entity resolution confidence score is used as the value of the entity resolution factor. For example, if optical character recognition identifies "Hankou" and the station city resolution service identifies the corresponding city as "Wuhan City" with a resolution confidence score of 1.0, then the entity resolution factor is set to 1.0, thereby increasing the matching priority of the corresponding field.
[0077] The positional prior decay factor describes the correlation between the order of structured data items in a user message and the order of dynamic form components in a form. Its calculation formula is as follows: in, Indicates the prior attenuation factor of the location; This represents the position attenuation parameter, which ranges from 0 to 1 and is used to control the degree of influence of sequence differences on the overall matching result. Indicates the position of a structured data item in the list of structured data items; This indicates the position of the dynamic form component within its definition. The smaller the difference between the two positions, the closer the prior position decay factor is to 1, indicating that the order of user description is more consistent with the order of form design.
[0078] For required fields, a required field gain coefficient is used to further improve the matching priority of the corresponding field. When a dynamic form component is a required field, its overall alignment score is multiplied by [a factor]. ,in, This represents the required field gain coefficient, used to increase the matching weight of required fields. When a dynamic form component has an entity type identifier, the entity parsing gating factor is further used to correct the overall alignment score; if entity parsing is successful, the overall alignment score is multiplied by [the factor]. ,in, Indicates the entity enhancement parameters, This represents the entity resolution confidence score; if entity resolution fails, the overall alignment score is multiplied by the penalty base. ,in, A value of 0.6 is used to reduce the impact of entity parsing failures on field matching. Finally, the above similarity factors are fused according to the preset multi-factor similarity calculation rules to construct a multi-factor similarity model, achieving comprehensive matching between structured data items and dynamic form components.
[0079] Step S106: Based on the process template ID and form data object, the flowchart parsing agent outputs the approval information for the next node.
[0080] Specifically, based on the process template ID and the form data object, the flowchart definition file corresponding to the target process is parsed by the flowchart parsing intelligent agent. The target process flow path and the next node are determined according to the node exit conditions in the form data object and the flowchart definition file. Combined with the approver configuration rules and organizational structure data corresponding to the next node, the approval process information of the next node is output.
[0081] In one possible implementation, step S106 further includes: obtaining the corresponding flowchart definition file based on the process template ID, parsing the flowchart definition file, and determining the process start node; determining the next node based on the form data object and the node exit conditions corresponding to the process start node; parsing the approver configuration rules corresponding to the next node, and determining the approver candidate list corresponding to the next node in conjunction with the organizational structure data; the approver configuration rules include rules for specifying roles, rules for specifying personnel, rules for superiors, and dynamic calculation rules; if the approver candidate list includes multiple approvers, obtaining the process step configuration method corresponding to the next node, and determining the corresponding approval processing method based on the process step configuration method; the process step configuration method includes parallel countersigning and serial countersigning; and generating the approval step information for the next node based on the approver candidate list and the approval processing method.
[0082] Specifically, after receiving the process template ID and form data object, the flowchart parsing agent retrieves the corresponding flowchart definition file based on the process template ID, parses the flowchart definition file, and identifies the process start node and the connection relationships between each process node. The flowchart definition file describes the process nodes of the target process, the node connection relationships, the node exit conditions, and the approval rules corresponding to each approval node, providing a basis for process flow path parsing.
[0083] Subsequently, the flowchart parsing agent reads the node exit conditions corresponding to the starting node of the process and matches them with the business data execution conditions in the form data object to determine the target process flow path that meets the node exit conditions and the corresponding next node. The node exit conditions are configured on the flowchart connection lines and describe the business conditions that must be met for the process to flow from the current node to different nodes. For example, in a project approval process, when the project stage in the form data object is the pre-sales stage, the flowchart parsing agent determines that the process flows to the sales approval node based on the corresponding node exit conditions; when the project stage is the implementation stage, it determines that the process flows to the project manager approval node; and when the business type is departmental expenses, it determines that the process flows to the department head approval node, thus achieving dynamic determination of the process flow.
[0084] After determining the next node, the flowchart parsing agent analyzes the approver configuration rules corresponding to that node and, in conjunction with organizational structure data, determines the candidate list of approvers for that node. The approver configuration rules include role-specifying rules, personnel-specifying rules, superior-leader rules, and dynamic calculation rules. Role-specifying rules determine approvers based on preset roles; personnel-specifying rules directly designate approvers; superior-leader rules determine the initiator's direct superior as the approver based on the organizational structure; and dynamic calculation rules dynamically calculate approvers by combining business fields in the form data object and organizational structure data.
[0085] If the list of approvers includes multiple approvers, the flowchart parsing agent further obtains the configuration method for the next node's process steps. These configuration methods include parallel review and sequential review. When the process step configuration method is parallel review, multiple approvers from the approver candidate list are simultaneously assigned to the current approval node, allowing multiple approvers to complete the approval process in parallel. When the process step configuration method is sequential review, the approval process is completed sequentially according to the approval order pre-configured in the flowchart definition file.
[0086] After completing the process flow path parsing, determining the next node, and determining the approval processing method, the flowchart parsing agent combines the next node, the list of approver candidates, and the approval processing method to generate the next node approval process information. The next node approval process information includes at least the name of the next approval node, the approval processing method, and the corresponding approver information, which is used for subsequent embedded confirmation card display and process instance creation.
[0087] Step S107: Render the form data object and the next approval step information into an embedded confirmation card and send it to the chat window for the target user to confirm or modify.
[0088] Specifically, the human-computer confirmation interaction module receives form data objects and next-step approval information, and performs visual rendering of both based on a preset confirmation card template to generate an embedded confirmation card. This embedded confirmation card displays the process form content and approval flow information uniformly in the chat window, allowing the target user to complete the process confirmation without switching to the collaborative office system page.
[0089] The embedded confirmation card uses a structured layout. The form display area shows the names and values of each field in the form data object, allowing users to directly modify the corresponding field content. The approval information display area shows the name of the next approval node, the approval processing method, and the next approval step information. When the approver can be changed, users can also reselect or add approvers. By integrating process information with approval information, users can verify process content and confirm the approval path within the chat window.
[0090] After the embedded confirmation card is generated, the human-computer interaction module sends the embedded confirmation card to the chat window and listens for the target user's interaction. When the target user modifies the form fields or approver information, the human-computer interaction module receives the modified content and synchronously updates the corresponding form data object or the next approval stage information, then re-renders the embedded confirmation card for display. When the target user confirms that the form content and approval information are correct, the module receives the confirmation operation triggered by the target user and sends the confirmed form data object and the next approval stage information to the process initiation interface to execute subsequent process initiation processing.
[0091] In one possible implementation, step S107 further includes: in response to receiving a confirmation instruction from the target user for the embedded confirmation card, submitting the confirmed form data object and the confirmed next-node approval information to the collaborative office system to create a process instance and return the process initiation result.
[0092] Specifically, after the target user confirms the form content and approval information, the chat window receives the confirmation command triggered by the target user and sends the confirmation command to the process initiation interface. The process initiation interface obtains the form data object finally confirmed by the target user and the next approval step information based on the confirmation command, and encapsulates the two to form process initiation request data that meets the interface specifications of the collaborative office system.
[0093] Subsequently, the process initiation interface calls the process creation interface provided by the collaborative office system, submitting the process template ID, the confirmed form data object, and the confirmed next node approval information to the collaborative office system. The collaborative office system then creates a target process instance based on the corresponding process template and completes the allocation of the first approval node approval task according to the confirmed next node approval information, thereby completing the process initiation.
[0094] After a workflow instance is created, the collaborative office system returns the corresponding workflow initiation result. The workflow initiation result includes at least the workflow instance identifier, workflow number, current approval node, and workflow access identifier. The workflow instance identifier uniquely identifies the current workflow instance; the workflow number is used for workflow querying and management; the current approval node identifies the current approval stage of the workflow; and the workflow access identifier allows access to the corresponding workflow details. The workflow initiation interface receives the workflow initiation result and sends it to the chat window to notify the target user that the workflow initiation was successful. It also allows the target user to view the corresponding workflow details using the workflow access identifier.
[0095] It should be noted that the construction methods of each intelligent agent in the embodiments of this application include, but are not limited to: the intent recognition intelligent agent is constructed based on a large language model, and encapsulated by combining intent recognition prompts, process type recognition rules, and an intelligent agent orchestration framework; the process matching intelligent agent is constructed based on a large language model and a process template retrieval interface, and encapsulated by combining a process template library and process matching rules; the multimodal data extraction intelligent agent is constructed based on a large language model, a multimodal model, and an intelligent agent orchestration framework, and integrates text extraction sub-intelligent agents, image extraction sub-intelligent agents, and file extraction sub-intelligent agents; wherein, the text extraction sub-intelligent agent is constructed based on a natural language processing model. The image extraction sub-agent is built based on an optical character recognition model and a visual language model, while the document extraction sub-agent is built based on a document parsing model. The dynamic form assembly agent is built based on a large language model, a dynamic form engine, and field mapping rules, and integrates a multi-factor similarity model to complete the dynamic form assembly. The flowchart parsing agent is built based on a large language model, a flowchart engine parsing module, and a rule reasoning engine, and is encapsulated and built by combining flowchart definition files, approval rule configurations, and organizational structure data. The embedded confirmation card rendering agent is built based on a large language model, a card template engine, and a front-end rendering framework, and is encapsulated and built by combining a chat window interaction interface.
[0096] Please refer to Figure 2 This illustration shows a schematic diagram of a dynamic process initiation device based on intelligent agent orchestration provided in an embodiment of this application. The device includes an acquisition module 21 and a processing module 22, wherein... Get module 21, used for.
[0097] Processing module 22 is used to acquire user messages sent by the target user and identify whether the user messages contain process processing instructions.
[0098] The acquisition module 21 is used to obtain the process type result corresponding to the process processing instruction through the intent recognition intelligent agent if it is confirmed that the user message contains a process processing instruction.
[0099] The acquisition module 21 is used to retrieve process templates that meet the target matching degree from the process template library based on the process type result, and output the process template ID and basic process information.
[0100] The acquisition module 21 is used to extract structured information from user messages through a multimodal data extraction agent and output a list of structured data items.
[0101] Processing module 22 is used to obtain the dynamic form component definition corresponding to the process template ID through a dynamic form assembly agent based on the process template ID, basic process information, and a list of structured data items. It then performs field mapping, type conversion, and data filling on the list of structured data items and the dynamic form component definition to obtain a form data object.
[0102] Processing module 22 is used to output the next node approval information through the flowchart parsing agent based on the process template ID and form data object.
[0103] Processing module 22 is used to render the form data object and the next approval step information into an embedded confirmation card and send it to the chat window for the target user to confirm or modify.
[0104] In one possible implementation, the acquisition module 21 is used to acquire user messages sent by the target user and identify whether the user messages contain process processing instructions. Specifically, this includes: inputting the process processing instructions into a natural language model and outputting the intent type corresponding to the user message through the natural language model; the intent type includes business submission initiation type, consultation type, data query type, approval query type, and ordinary chat type; when the intent type is any one of business submission initiation type, consultation type, data query type, or approval query type, it is confirmed that the user message contains process processing instructions; when the intent type is ordinary chat type, it is confirmed that the user message does not contain process processing instructions.
[0105] In one possible implementation, the acquisition module 21 is used for a multimodal data extraction agent, including a text extraction agent, an image extraction agent, and a file extraction agent. The multimodal data extraction agent extracts structured information from user messages and outputs a list of structured data items. Specifically, this includes: using the text extraction agent to perform entity recognition and attribute extraction on the text content of the user message to obtain text structured data items; using the image extraction agent to perform optical character recognition and image content understanding on the image content of the user message to obtain image structured data items; using the file extraction agent to perform structured parsing on the file content of the user message to obtain file structured data items; and performing unified format conversion and integration processing on the text structured data items, image structured data items, and file structured data items to output a list of structured data items.
[0106] In one possible implementation, the processing module 22 is used to obtain the dynamic form component definition corresponding to the process template ID through a dynamic form assembly agent based on the process template ID, basic process information, and a list of structured data items. It then performs field mapping, type conversion, and data filling between the list of structured data items and the dynamic form component definition. Specifically, this includes: obtaining the dynamic form component definition based on the process template ID; the dynamic form component definition includes field names, field codes, field types, whether fields are required, and option data sources; constructing a field mapping relationship between the list of structured data items and the dynamic form component definition based on a multi-factor similarity model; and outputting the field mapping relationship corresponding to each structured data item. Field mapping results; field mapping results include successful mapping results and failed mapping results; if the field mapping result is a successful mapping result and the corresponding field is not a multi-value field, then data type validation processing, data type conversion processing, and data filling processing are performed on the corresponding field; if the field mapping result is a successful mapping result and the corresponding dynamic form component is a multi-value field, then data filling processing is performed on the corresponding field in array form; if the field mapping result is a failed mapping result and the corresponding dynamic form component is a required field, then the corresponding field is marked as a field to be supplemented, and intelligent recommendation values are generated; based on the data filling results corresponding to each dynamic form component, the form is assembled to obtain a form data object.
[0107] In one possible implementation, the processing module 22 is used to construct a multi-factor similarity model, specifically including: constructing a semantic vector similarity factor based on the semantic association degree between each structured data item and each dynamic form component; constructing an ontology alignment factor based on the business ontology association relationship between each structured data item and each dynamic form component; constructing a type compatibility factor based on the compatibility relationship between the data type of each structured data item and the field type of each dynamic form component; constructing a context co-occurrence factor based on the historical mapping association relationship between each structured data item and each dynamic form component; constructing an entity parsing factor based on the matching relationship between the entity parsing result corresponding to each structured data item and the entity type corresponding to each dynamic form component; defining multi-factor similarity calculation rules based on the semantic vector similarity factor, ontology alignment factor, type compatibility factor, context co-occurrence factor, and entity parsing factor, and constructing a multi-factor similarity model.
[0108] In one possible implementation, the processing module 22 is used to output the next node approval information through a flowchart parsing agent based on the process template ID and form data object. Specifically, this includes: obtaining the corresponding flowchart definition file based on the process template ID, parsing the flowchart definition file, and determining the process start node; determining the next node based on the form data object and the node exit conditions corresponding to the process start node; parsing the approver configuration rules corresponding to the next node, and determining the approver candidate list corresponding to the next node in combination with organizational structure data; the approver configuration rules include rules for specifying roles, rules for specifying personnel, rules for superiors, and dynamic calculation rules; if the approver candidate list includes multiple approvers, obtaining the process step configuration method corresponding to the next node, and determining the corresponding approval processing method based on the process step configuration method; the process step configuration method includes parallel countersigning and serial countersigning; and generating the next node approval information based on the approver candidate list and the approval processing method.
[0109] In one possible implementation, after rendering the form data object and the next approval step information into an embedded confirmation card and sending it to the chat window for the target user to confirm or modify, the processing module 22 further includes: in response to receiving the target user's confirmation instruction for the embedded confirmation card, submitting the confirmed form data object and the confirmed next approval step information to the collaborative office system to create a process instance and return the process initiation result.
[0110] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.
[0111] This application also provides an electronic device. (See reference...) Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 301, at least one communication bus 302, a user interface 303, at least one network interface 304, and a memory 305.
[0112] The communication bus 302 is used to enable communication between these components.
[0113] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.
[0114] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).
[0115] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 305, and by calling data stored in memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.
[0116] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory 305 may include a non-transitory computer-readable storage medium. The memory 305 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. (Refer to...) Figure 3 The memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a dynamic process initiation application based on agent orchestration.
[0117] exist Figure 3 In the illustrated electronic device, the user interface 303 is primarily used to provide an input interface for the user and acquire user input data; while the processor 301 can be used to call the dynamic process based on intelligent agent orchestration stored in the memory 305 to initiate an application. When executed by one or more processors 301, the electronic device performs one or more of the methods described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0118] This application also provides a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.
[0119] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0120] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.
[0121] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0122] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0123] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.
[0124] The above description is merely an exemplary embodiment disclosed in this application and should not be construed as limiting the scope of this application. Any equivalent changes and modifications made in accordance with the teachings of this application shall still fall within the scope of this application.
[0125] This application is intended to cover any variations, uses, or adaptations disclosed herein that follow the general principles disclosed herein and include common knowledge or customary technical means in the art that are not described in this application.
Claims
1. A dynamic process initiation method based on agent orchestration, characterized in that, The method includes: Obtain user messages sent by the target user and identify whether the user messages contain process processing instructions; If it is confirmed that the user message contains the process processing instruction, the process type result corresponding to the process processing instruction is obtained through the intent recognition intelligent agent; Based on the process type result, retrieve the process template that meets the target matching degree from the process template library, and output the process template ID and basic process information; The multimodal data extraction agent extracts structured information from the user messages and outputs a list of structured data items. Based on the process template ID, the basic process information, and the list of structured data items, the dynamic form component definition corresponding to the process template ID is obtained through the dynamic form assembly agent, and field mapping, type conversion, and data filling are performed on the list of structured data items and the dynamic form component definition to obtain a form data object; Based on the process template ID and the form data object, the flowchart parsing agent outputs the next node approval information. The form data object and the next approval step information are rendered as an embedded confirmation card and sent to the chat window for the target user to confirm or modify.
2. The method of claim 1, wherein, The step of acquiring user messages sent by the target user and identifying whether the user messages contain process processing instructions specifically includes: The process instructions are input into a natural language model, and the intent type corresponding to the user message is output through the natural language model; the intent type includes business submission initiation type, consultation type, data query type, approval query type, and ordinary chat type; When the intent type is any one of the business submission initiation type, the consultation type, the data query type, or the approval query type, it is confirmed that the user message contains the process processing instruction. When the intent type is the normal chat type, it is confirmed that the user message does not contain the process processing instruction.
3. The method of claim 1, wherein, The multimodal data extraction agent includes a text extraction agent, an image extraction agent, and a file extraction agent. The process of extracting structured information from the user message using the multimodal data extraction agent and outputting a list of structured data items specifically includes: The text extraction agent performs entity recognition and attribute extraction on the text content of the user message to obtain structured text data items. The image extraction agent performs optical character recognition and image content understanding on the image content in the user message to obtain image structured data items. The file extraction agent performs structured parsing of the file content in the user message to obtain structured file data items. The text structured data items, image structured data items, and file structured data items are uniformly formatted and integrated to output a list of structured data items.
4. The method of claim 1, wherein, The process, based on the process template ID, the basic process information, and the structured data item list, obtains the dynamic form component definition corresponding to the process template ID through a dynamic form assembly agent, and performs field mapping, type conversion, and data filling on the structured data item list and the dynamic form component definition. Specifically, this includes: The dynamic form component definition is obtained based on the process template ID; the dynamic form component definition includes field name, field code, field type, whether it is required, and option data source; A field mapping relationship between the structured data item list and the dynamic form component definition is constructed based on a multi-factor similarity model, and the field mapping result corresponding to each structured data item is output; the field mapping result includes mapping success result and mapping failure result; If the field mapping result is a successful mapping result and the corresponding field is not a multi-value field, then perform data type validation, data type conversion, and data filling processing on the corresponding field. If the field mapping result is a successful mapping result and the corresponding dynamic form component is a multi-value field, then the corresponding field is filled with data in array form; If the field mapping result is a mapping failure result and the corresponding dynamic form component is a required field, then the corresponding field will be marked as a field to be supplemented, and a smart recommendation value will be generated; The form is assembled based on the data filling results corresponding to each dynamic form component to obtain the form data object.
5. The method of claim 4, wherein, The construction of the multi-factor similarity model specifically includes: Based on the semantic association between each structured data item and each dynamic form component, a semantic vector similarity factor is constructed; Based on the business ontology relationship between each structured data item and each dynamic form component, an ontology alignment factor is constructed; Based on the compatibility relationship between the data types of each structured data item and the field types of each dynamic form component, a type compatibility factor is constructed. Based on the historical mapping relationship between each structured data item and each dynamic form component, a context co-occurrence factor is constructed; Based on the matching relationship between the entity parsing results corresponding to each structured data item and the entity type corresponding to each dynamic form component, an entity parsing factor is constructed; Based on the semantic vector similarity factor, the ontology alignment factor, the type compatibility factor, the context co-occurrence factor, and the entity parsing factor, a multi-factor similarity calculation rule is defined, and the multi-factor similarity model is constructed.
6. The method of claim 1, wherein, The step of outputting the next approval stage information through the flowchart parsing agent based on the process template ID and the form data object specifically includes: Based on the process template ID, obtain the corresponding flowchart definition file, parse the flowchart definition file, and determine the process start node; Based on the form data object and the node exit conditions corresponding to the process start node, determine the next node; The approver configuration rules corresponding to the next node are analyzed, and the approver candidate list corresponding to the next node is determined in combination with the organizational structure data; the approver configuration rules include rules for specifying roles, rules for specifying personnel, rules for superiors, and dynamic calculation rules; If the list of approvers includes multiple approvers, then the process step configuration method corresponding to the next node is obtained, and the corresponding approval processing method is determined based on the process step configuration method; the process step configuration method includes parallel countersigning and serial countersigning. Based on the list of approvers and the approval process, the next approval step information is generated.
7. The method of claim 1, wherein, After rendering the form data object and the next approval step information into an embedded confirmation card and sending it to the chat window for the target user to confirm or modify, the method further includes: In response to receiving a confirmation instruction from the target user for the embedded confirmation card, the confirmed form data object and the confirmed next approval step information are submitted to the collaborative office system to create a process instance and return the process initiation result.
8. An apparatus for initiating a dynamic process based on agent orchestration, comprising: The device includes an acquisition module and a processing module, wherein, The acquisition module is used to acquire user messages sent by the target user and identify whether the user messages contain process processing instructions; The acquisition module is used to, if it is confirmed that the user message contains the process processing instruction, obtain the process type result corresponding to the process processing instruction through the intent recognition intelligent agent; The acquisition module is used to retrieve process templates that meet the target matching degree from the process template library based on the process type result, and output the process template ID and basic process information; The acquisition module is used to extract structured information from the user message through a multimodal data extraction agent and output a list of structured data items. The processing module is used to obtain the dynamic form component definition corresponding to the process template ID through a dynamic form assembly agent based on the process template ID, the basic process information, and the structured data item list, and to perform field mapping, type conversion, and data filling on the structured data item list and the dynamic form component definition to obtain a form data object. The processing module is used to output the next node approval information through the flowchart parsing agent based on the process template ID and the form data object; The processing module is used to render the form data object and the next node approval information into an embedded confirmation card and send it to the chat window for the target user to confirm or modify.
9. An electronic device, comprising: It includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the dynamic process initiation method based on agent orchestration as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium, comprising: The non-transitory computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.
