Conversation method, system, electronic device and storage medium for realizing target service

Abstract

The application relates to a conversation method and system for realizing a target service, an electronic device and a storage medium, the method comprising the following steps: analyzing a conversation request of a user for a target service initiated on a recruitment service platform to obtain a user identifier; obtaining an Agent interface based on the user identifier; sending the conversation request of the user to a Serverless platform through the Agent interface; routing the conversation request of the user to a dedicated Agent service instance of the user by the Serverless platform; processing the conversation request by the dedicated Agent service instance of the user and generating conversation response information; returning the conversation response information to the recruitment service platform and the user by the Serverless platform. The application can effectively isolate user sessions, improve the expansibility of the system, the flexibility of tool integration and the stability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a dialogue method, system, electronic device, and storage medium for realizing target business. Background Technology

[0002] With the rapid development of large-scale artificial intelligence (AI) model technology, AI dialogue systems have been widely applied to various internet service platforms. Among them, AI dialogue systems on recruitment service platforms have become a crucial interactive medium connecting the platform, job seekers, and corporate users. They can intelligently answer user inquiries, assist with resume screening, job matching, business queries, job postings, and other recruitment-related tasks, greatly improving the automation and intelligence of recruitment services and reducing the platform's manual service costs. Currently, AI dialogue systems built on traditional large-scale model interaction architectures mainly suffer from the following technical pain points: First, multi-user session isolation is challenging, resulting in low system resource utilization. Existing AI dialogue systems generally adopt a service instance sharing model, where all platform users share the same dialogue service instance and context storage system. In scenarios with large-scale simultaneous online interaction, data such as dialogue history, interaction context, and business preferences from different users are highly susceptible to confusion and crosstalk, leading to issues like erratic AI dialogue responses and incorrect business matching, impacting user experience and the accuracy of business processing. To address the session crosstalk problem, some solutions attempt to deploy a dedicated service instance for each user. However, this approach results in significant redundancy in server computing power, memory, and storage resources; the larger the user base, the more severe the resource consumption, while also significantly increasing the manpower and hardware costs of system deployment and daily maintenance.

[0003] Secondly, the business capabilities lack scalability and the flexibility for feature iteration is poor. Recruitment processes are characterized by numerous screening dimensions and high update frequency. In addition to standard screening dimensions such as education, work experience, and salary, personalized resume / job screening dimensions are often added as needed, such as certifications, years of experience, and place of residence. However, the business rules and screening logic of traditional AI dialogue systems are deeply coupled with the backend code. Whenever it is necessary to add, modify, or delete business screening dimensions and corresponding functional modules, developers need to modify a large amount of core backend code, and only after completing code debugging and overall testing can the system be deployed online. This process is cumbersome and has a long iteration cycle.

[0004] Third, large-scale models face challenges in integrating with business tools and exhibit poor compatibility. General-purpose large-scale models possess only basic cognitive capabilities such as semantic understanding, intelligent question answering, and text generation, lacking the ability to independently perform business operations and data processing. To accomplish practical tasks in recruitment scenarios, such as accurate resume retrieval, standardized conversion of regional codes, job information organization, and data validation, various business tools need to be integrated into the large-scale model. However, due to the different input parameter formats, interface call methods, and data return specifications of various recruitment business tools, separate adaptation code needs to be developed when the large-scale model interfaces with different business tools. This results in cumbersome integration processes, high code redundancy, and poor compatibility, leading to high integration costs for adding new business tools and modifying existing ones. Summary of the Invention

[0005] In view of the technical problems existing in the prior art, the present invention proposes a dialogue method, system, electronic device and storage medium for realizing target services, which can isolate user sessions and improve the scalability, tool integration flexibility and stability of the system.

[0006] To address the aforementioned technical problems, according to one aspect of the present invention, a dialogue method for realizing a target business is provided, which is applied to a recruitment service platform, the method comprising: Parse the dialogue requests initiated by users at the front-end application layer of the recruitment service platform to obtain user identifiers; The Agent interface is obtained based on the user identifier; wherein, the Agent is configured with a tool description for completing the target business and one or more independent skill files; The user's conversation request is sent to the Serverless platform via the Agent interface; The serverless platform routes the user's conversation request to the user's dedicated Agent service instance; The user's dedicated Agent service instance processes the dialogue request and generates dialogue response information; and The serverless platform returns the dialogue response information to the recruitment service platform and then back to the application layer to display it to the user. The steps for the user's dedicated Agent service instance to process the dialogue request and generate dialogue response information include: The large language model is invoked to understand user input information in dialogue requests based on tool descriptions and skill files, determine the task flow to complete the target task, and identify the tools required for each stage of the task flow; and The task flow is executed, and response information for the dialogue request is generated based on the execution result of the task flow.

[0007] Optionally, when obtaining the Agent interface based on the user identifier, the dedicated Agent function and its interface for implementing the target business for the user are queried from the preset FC function binding table. The FC function binding table records the dedicated Agent function and its interface information for implementing the target business that are bound to the user identifier.

[0008] Optionally, in response to the absence of a function and its interface for the Agent that implements the target service for the user from a pre-configured FC function binding table, the method further includes: The function to create an Agent for the user's target business is created by calling the Serverless platform's function creation interface; an interface for calling the function is created; and... The function and its calling interface are persisted to the FC function binding table, and a binding relationship with the user is established.

[0009] Optionally, when calling the function creation interface of the serverless platform to create an Agent for implementing the target business for the user, at least the following information should be configured: The function name should be set with at least the target business information and user identifier; Configure the function's runtime environment as a runtime environment, and specify the runtime as the Agent container image; Set session affinity and include session fields and user identifiers in the session affinity information; and Configure the file system information to be mounted.

[0010] Optionally, the step of creating an interface for calling the function further includes: The serverless platform's function trigger creation interface is used to create a trigger for the function, and the trigger name, type, corresponding function name, and triggering rules are configured; and Receive the trigger URL returned by the Serverless platform, where the trigger URL is the Agent interface that calls the function.

[0011] Optionally, the Serverless platform routes the user's conversation requests to the user's corresponding dedicated Agent service instance based on the session affinity information carried by the Agent interface.

[0012] Optionally, after receiving the dialogue request information transmitted through the Agent interface, the Serverless platform further includes: Check if a pre-warmed dedicated Agent service instance for the user exists in the Serverless platform; and In response to the existence of a pre-warmed dedicated Agent service instance for the user in the Serverless platform, the pre-warmed dedicated Agent service instance is reused.

[0013] Optionally, in response to the absence of a pre-warmed dedicated agent service instance for the user in the serverless platform, the method further includes the step of creating a dedicated agent service instance for the user, including: Based on the function configuration information corresponding to the user, pull the Agent container image to start the Agent container; The agent container creates a session and loads the configured list of tools and skill files for the target business; and Establish a connection with the tools in the tool list.

[0014] Optionally, the step of creating a dedicated Agent service instance for the user further includes: The serverless platform returns a notification message about the creation of the user's dedicated Agent service instance to the recruitment service platform; and The recruitment service platform generates a loading animation based on the creation notification information of the user's exclusive Agent service instance and displays it to the user on the front end.

[0015] Optionally, when pulling the Agent container image based on the function configuration information corresponding to the user, the version of the Agent container image to be pulled is determined based on the Agent container image alias, and / or the Agent container image is optimized when pulling the Agent container image based on the optimization configuration information.

[0016] Optionally, when calling the function creation interface of the serverless platform to create an Agent for implementing the target business for the user, the idle timeout threshold information of the instance is also configured. Correspondingly, after the user's dedicated Agent service instance is created, the Serverless platform monitors the idle time of the user's dedicated Agent service instance, and when the idle timeout threshold is reached, the user's dedicated Agent service instance is reclaimed.

[0017] Optionally, after the user's dedicated Agent service instance is created, the user's dedicated Agent service instance is kept in a persistent state during working hours.

[0018] Optionally, the recruitment service platform sends heartbeat requests to the serverless platform periodically during working hours. The serverless platform responds to the heartbeat requests by maintaining the persistent state of the user's dedicated Agent service instance. The periodicity of sending heartbeat requests is less than the idle timeout threshold.

[0019] Optionally, it further includes calling the function alias creation interface of the Serverless platform to create a function alias for the Agent that implements the target business, and configuring the Agent container image version information and its traffic weight; and persisting the function alias and its configuration information of the Agent that implements the target business to the FC function binding table, and establishing a binding relationship with the corresponding user.

[0020] Optionally, when routing a user's conversation request to the user's dedicated Agent service instance, the Serverless platform determines the corresponding version of the dedicated Agent service instance based on the function alias and traffic weight carried in the Agent interface.

[0021] Optionally, the method further includes: creating a public function and its interface for the Agent that implements the target business, and persisting it to a function binding table stored in the FC. Correspondingly, the Serverless platform creates instances of the public function as reserved public instances during working hours.

[0022] Optionally, in response to the absence of a function and its interface for the Agent that implements the target business for the user from the preset FC function binding table, a public function interface is obtained as the Agent interface; correspondingly, the Serverless platform routes the user's dialogue request to a reserved public instance.

[0023] Optionally, the step of the user's dedicated Agent service instance processing the dialogue request and generating dialogue response information further includes: Load the user's target business historical session information from the file system and send the historical session information to the large language model, enabling the large language model to understand the user input information in the dialogue request based on the historical session information, tool descriptions, and skill files; and The current session information is stored in the user's session information set in the file system, and the session information is isolated by the target business identifier. The current session information includes user input information and corresponding response information in the user request.

[0024] Optionally, when loading the user's target service's historical session information from the file system, the complete session information of the most recent preset round is loaded; if the round of the historical session information is greater than the preset round, a summary of the complete session information greater than the preset round is obtained and the summary is loaded.

[0025] Optionally, the target business is a resume screening business corresponding to the target position, and there are multiple skill files to complete the target business, with each skill file corresponding to a screening dimension; the corresponding tools include search engine clients, standardized clients, and large language models.

[0026] Optionally, the step of executing the task flow to obtain the response information of the dialogue request includes: Use a search engine client to obtain job information for the target position; The large language model is used to extract the corresponding screening criteria from the job information of the target job based on the skill file corresponding to the screening dimension. The extracted filtering criteria are standardized using a standardized client to obtain standardized filtering criteria; A large language model is used to generate search criteria based on filtering conditions extracted from user input and job information. The search engine client is used to search a pre-set number of resumes based on search criteria. A large language model is used to analyze the preset number of resumes to obtain the analysis results for each resume; and The system uses a search engine client or a standardized client to output resume search results in a pre-formatted format. The resume search results include resume information and resume analysis results; wherein, the resume search results serve as response information to the dialogue request.

[0027] Optionally, the analysis results of the resume may include resume scoring instructions and / or highlights descriptions.

[0028] Optionally, after generating search criteria using a large language model based on filtering conditions extracted from user input information and job information, the search criteria are returned to the user; after obtaining confirmation information from the user regarding the search criteria, the search engine client is invoked, enabling the search engine client to search for a preset number of resumes based on the search criteria.

[0029] Optionally, the search criteria include multiple search options, which correspond to at least two or more different priorities.

[0030] According to another aspect of the present invention, the present invention also provides a dialogue system for realizing target business, which is applied to a recruitment service platform, comprising: The front-end interface layer is configured to receive and parse the dialogue request initiated by the user in the front-end application layer of the recruitment service platform to the target business in order to obtain the user identifier, and return the dialogue response information to the front-end application layer of the recruitment service platform to be displayed to the user. The AI ​​service layer is configured to obtain the Agent interface based on the user identifier; The user's dialogue request is sent to the function computing layer hosted by the Serverless platform through the Agent interface. The Agent is configured with tool descriptions and one or more skill files for completing the target business. The function computation layer is configured to route the user's dialogue request to the user's dedicated Agent service instance; the user's dedicated Agent service instance processes the dialogue request and generates dialogue response information, which is then returned to the AI ​​service layer. The large language model layer, configured to provide a large language model, responds to calls from Agent service instances, understands user input in dialogue requests based on tool descriptions and skill files, determines the task flow to complete the target task and the tools required for each stage of the task flow, and generates dialogue response information based on the execution results of the task flow; and The tool layer is configured to respond to tool calls made by the Agent service instance when performing phase tasks, and provides the corresponding tools to complete the phase tasks.

[0031] According to another aspect of the present invention, an electronic device is also provided, comprising a processor and a memory, wherein a set of computer program instructions is stored in the memory, and the aforementioned dialogue method for realizing the target service is implemented when the processor executes the set of computer program instructions in the memory.

[0032] According to another aspect of the present invention, the present invention also provides a computer-readable storage medium, wherein a computer program instruction set is stored on the computer-readable storage medium, and the computer program instruction set, when executed by a processor, performs the aforementioned dialogue method for implementing the target service.

[0033] According to another aspect of the present invention, the present invention also provides a computer program product comprising a computer program instruction set, which, when executed by a processor, implements the aforementioned dialogue method for achieving the target business.

[0034] This invention effectively isolates user sessions and allows for convenient addition, deletion, or modification of the Agent's skill files as needed, thereby improving the system's scalability and flexibility. The Agent in this invention integrates multiple tools, and the tools communicate with the large language model using the MCP protocol, resulting in good tool integration and strong adaptability. Attached Figure Description

[0035] The preferred embodiments of the present invention will now be described in further detail with reference to the accompanying drawings, wherein: Figure 1 This is an architecture diagram of a recruitment service platform according to an embodiment of the present invention; Figure 2 This is a block diagram illustrating the principle of an AI dialogue system applied to a recruitment service platform according to an embodiment of the present invention. Figure 3 This is a flowchart of a dialogue method for implementing a target service according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the entry page of a dialogue system according to an embodiment of the present invention; Figure 5 This is a schematic diagram of a front-end application page of a dialogue system according to an embodiment of the present invention; Figure 6 This is a flowchart of a dialogue method for implementing resume search according to an embodiment of the present invention; Figure 7 This is a flowchart of a method for a user's dedicated Agent service instance to process the dialogue request and generate dialogue response information according to an embodiment of the present invention; Figure 8 This is a schematic diagram of the front-end application page of the dialogue system during the second stage of the task flow execution according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the front-end application page of the dialogue system during stage six of the task flow execution, according to an embodiment of the present invention. Figure 10 This is a schematic diagram of the front-end application page of a dialogue system after the task flow has been completed, according to an embodiment of the present invention; and Figure 11 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] In the following detailed description, reference can be made to the accompanying drawings, which form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Specific embodiments of the present application are described in sufficient detail below to enable those skilled in the art to implement the technical solutions of the present application. It should be understood that other embodiments may also be utilized, or structural, logical, or electrical changes may be made to the embodiments of the present application.

[0038] See Figure 1 , Figure 1 This is an architecture diagram of a recruitment service platform according to an embodiment of the present invention. The recruitment service platform includes a user terminal and a server terminal 3. The user terminal is installed as a client on a user terminal device. The user terminal is divided into a recruitment terminal 1 and a job seeker terminal 2, as shown in the figure. The recruitment terminal 1 is installed on different recruitment user devices, such as the first recruitment user device 1a, the second recruitment user device 1b, etc., as shown in the figure. The job seeker terminal 2 is installed on job seeker user devices, such as the first job seeker user device 2a, the second job seeker user device 2b, the third job seeker user device 2c, etc., as shown in the figure. The server terminal 3 includes multiple servers, such as the first server 3b, the second server 3c, and the service management terminal device 3a, etc. The user devices and terminal devices here are, for example, desktop computers, smartphones, laptops, tablet computers, etc.

[0039] See Figure 2 , Figure 2 This is a schematic diagram of an AI dialogue system applied to a recruitment service platform according to an embodiment of the present invention. The dialogue system interfaces with the front-end application layer of the recruitment service platform. Users initiate dialogue requests through the front-end application layer, inputting relevant information about the business to be completed in natural language, such as the resumes required by recruiters or the desired positions by job seekers. The dialogue system receives user requests, processes them to complete resume or job searches, and returns the corresponding processing results to the user. Specifically, the dialogue system includes a front-end interface layer 100, an AI service layer 200, a function computation layer 300, a large language model layer 400, and a tool layer 500. The front-end interface layer 100 receives and parses the dialogue request initiated by the user in the front-end application layer to obtain the user identifier, and returns the dialogue response information to the front-end application layer for the user.

[0040] The AI ​​service layer 200 consists of a controller layer and an agent service layer. For example, the controller layer, as a Java microservice layer, provides multiple controllers, such as: a dialogue controller responsible for the full lifecycle management of dialogue sessions, completing basic session management tasks such as user session creation, context storage, session state maintenance, and session permission verification, connecting the entire AI dialogue business chain; a communication mechanism controller (SseController) relying on SSE long-connection technology to realize AI streaming dialogue interaction, pushing the content output by the large model in segments to the front-end page in real time; business controllers, such as the job requirement controller (JobRequirementCtrl), targeting recruitment business scenarios, undertaking the full lifecycle management operations of adding, editing, querying, and archiving job recruitment requirements, and uniformly managing basic recruitment job data; and a report controller (ReportCtrl) supporting AI report generation capabilities, generating relevant business reports after agent processing. The agent service layer is used to connect to the Serverless platform interface and the agents deployed on the Serverless platform, and maintains an FC function binding table, which records user-specific agent functions and their interfaces. After the AI ​​service layer 200 receives the user identifier sent by the front-end interface layer 100, it obtains the Agent interface based on the user identifier; then, through the Agent interface, it sends the user's dialogue request to the function computing layer 300 hosted by the Serverless platform. The Agent is configured with tool descriptions and skill files for completing the target business.

[0041] The function computation layer 300 is hosted by a serverless platform, which routes the user's dialogue request to the user's dedicated agent service instance; the user's dedicated agent service instance processes the dialogue request and generates dialogue response information, and returns the dialogue response information to the AI ​​service layer.

[0042] The Large Language Model Layer 400 provides a large language model. In response to calls from Agent service instances, the large language model understands user input information in dialogue requests based on tool descriptions and skill files, determines the task flow to complete the target task and the tools required for each stage of the task flow, and generates dialogue response information based on the execution results of the task flow.

[0043] The tool layer 500 includes the MCP Server and provides tools that conform to the standardized Model Context Protocol (MCP) to complete various tasks. The MCP Server responds to tool calls made by the Agent service instance when performing phase tasks and provides the corresponding tools to complete the phase tasks.

[0044] See Figure 3, Figure 3 This is a flowchart of a dialogue method for implementing a target service according to an embodiment of the present invention.

[0045] Step S11: Analyze the dialogue request initiated by the user on the recruitment service platform for the target business to obtain the user identifier.

[0046] Step S12: Obtain the Agent interface based on the user identifier; wherein the Agent is configured with a tool description for completing the target business and one or more independent skill files.

[0047] Step S13: Send the user's conversation request to the Serverless platform through the Agent interface.

[0048] In step S14, the Serverless platform routes the user's conversation request to the user's dedicated Agent service instance.

[0049] Step S15: The user's dedicated Agent service instance processes the dialogue request and generates dialogue response information.

[0050] In step S16, the Serverless platform returns the dialogue response information to the recruitment service platform and then to the user.

[0051] In step S15, when the user's dedicated Agent service instance processes the dialogue request and generates dialogue response information, it first calls the large language model, inputting the user's input information, tool description, skill file, and prompt words into the large language model. The large language model understands the task based on the prompt words and comprehends the user's input information in the dialogue request based on the tool description and skill file, determines the task flow to complete the target task and the tools required for each stage of the task flow, and then the user's dedicated Agent service instance executes the task flow and generates the response information for the dialogue request based on the execution result of the task flow.

[0052] This invention deploys the Agent on a Serverless platform independent of the recruitment service platform, thus ensuring that the Agent's operation does not affect the recruitment service platform's server computing power, memory, or other resources. Therefore, this invention deploys a dedicated Agent service instance for each user, effectively solving the problem of user session isolation. This invention configures the skills required by the Agent to complete the target business as independent files, separating them from business rules and logic. Therefore, when expanding business capabilities, only skill files need to be added according to the expanded capabilities, without modifying the Agent code, thereby effectively improving the scalability of business capabilities and the flexibility of functional iteration. Furthermore, this invention configures task flows for the target business and provides various tools conforming to the MCP protocol required to complete related tasks. A standardized calling interface is provided between the large language model and the tools. Based on the capabilities of the large language model, it understands the user's intent, determines the task flow and the tools required at each stage, and only needs to issue standardized tool call requests. The MCP server at the tool layer then converts these standardized tool call requests into API calls for specific tools. Therefore, the large language model and tools have strong application adaptability and high integration.

[0053] In one embodiment, the dialogue system performs the task of screening resumes for recruiting users. In the resume screening scenario, recruiting users log into the recruitment client (which is the front-end application layer of the recruitment service platform), see [link to relevant documentation]. Figure 4 , Figure 4 This is a schematic diagram of the entry page of a dialogue system according to an embodiment of the present invention. In this embodiment, the entry page of the dialogue system displays various posted positions by the recruiting user, such as "Product Manager," "Java Engineer," "Test Engineer," etc., as shown in the figure, and displays corresponding prompts, such as the text above the positions: "Select any position to start AI recruitment." When any position is clicked, the user enters the front-end application page of the dialogue system related to that position, such as... Figure 5 As shown, Figure 5 This is a schematic diagram of a front-end application page of a dialogue system according to an embodiment of the present invention. The front-end application page 10 of the dialogue system includes an information input box 101, a main text display area 102, and a side border 103. In this embodiment, the user input information is: "Help me find a product manager in Kunming, Bachelor's degree or above, 3+ years of experience, monthly salary 10k-15k". The front-end sends the user input information to the dialogue system back-end as a user request. The user request includes a user identifier, a job identifier, and the user input information. See [link to dialogue system back-end processing] for details. Figure 6 , Figure 6 This is a flowchart of a dialog method for implementing resume search according to an embodiment of the present invention.

[0054] Step S21: parse the user's dialogue request for the target job to obtain the user identifier.

[0055] Step S22: Query the FC function binding table based on the user identifier. In this embodiment, the FC function binding table records the Agent function and its interface information, with the Agent function name corresponding one-to-one with the user identifier and job title.

[0056] Step S23: Determine whether a dedicated Agent function corresponding to the user identifier is found. If so, obtain the corresponding Agent interface. If not, proceed to step S28.

[0057] Step S24: Send the user's conversation request to the Serverless platform through the Agent interface.

[0058] In step S25, the Serverless platform routes the user's conversation request to the user's dedicated Agent service instance.

[0059] Step S26: The user's dedicated Agent service instance processes the dialogue request and generates dialogue response information.

[0060] In step S27, the Serverless platform returns the dialogue response information to the recruitment service platform and then to the user, ending the dialogue round.

[0061] Step S28: Call the function creation interface of the Serverless platform to create a function for the Agent to implement the target business for the user.

[0062] Step S29: Create an interface for calling the function.

[0063] Step S291: Persist the function and its calling interface to the FC function binding table, thereby establishing the binding relationship between the function and the user, and return to step S24.

[0064] This invention maintains an FC function binding table at the AI ​​service layer, which records the proprietary FC functions created for users. The Serverless platform creates a dedicated Agent service instance for each user based on the created FC functions. When a user first uses the dialogue system, the user's identifier is not present in the FC function binding table. Therefore, in step S28, the Serverless platform's function creation interface is called to create a function for the Agent to implement the target business for the user. Specifically, when creating the Agent function for implementing the target business for the user, the Serverless platform API "CreateFunction" is called, and relevant information is configured, including the following fields: Function name: For example, when the user identification information is "hrId=123", set the function name to "ai_hr_assistant_123".

[0065] Runtime: custom-container. This configures the function execution environment as a runtime and specifies the runtime as the Agent container image.

[0066] Resources: 0.5-core CPU / 512MB RAM.

[0067] Session affinity: "x-session-hr-id" header route. This means including the session field and username in the session affinity field and adding it to the header of the user request.

[0068] Idle timeout: 1800 seconds (30 minutes). This field indicates when an instance will be reclaimed after the idle period has exceeded the specified duration.

[0069] TTL: 86400 seconds (24 hours); used to represent the maximum lifespan of an instance.

[0070] File system mount: / agent_run / prod / ai_hr_assistant_123, used to determine the storage address of the file system storing the user instance session information.

[0071] Container configuration information: including image address and port.

[0072] After receiving the configuration information from the API CreateFunction, the serverless platform creates a dedicated Agent service instance for the user upon the user's first request or upon a subsequent request after the instance has been idled and recycled. Specifically, this includes: pulling the Agent container image based on the function configuration information corresponding to the user to start the Agent container; then, the Agent container creates a session and loads the configured tool list and skill files for the target business; and finally, it establishes a call connection with the tools in the tool list. In this embodiment, the target business to be completed by the dialogue system is resume screening. To achieve resume matching, this embodiment abstracts each screening dimension into an independent YAML Skill file (technical file). When adding / modifying screening dimensions, only the corresponding YAML file needs to be edited; the Agent code does not need to be modified. The Skill file is automatically distributed to all service instances through image version management, thereby achieving hot updates. The tools in this embodiment include a search engine client, a standardized client, and a large language model.

[0073] In step S29, when creating the interface to call the function, the serverless platform's function trigger creation interface (CreateTrigger) is called to create a trigger for the function, and the trigger name, type, and trigger rules are configured. In this embodiment, the configured trigger name is "http-trigger", the type is "http", and the trigger rule configuration is, for example, the trigger method is "GET" or "POST". Then, this configuration information and the corresponding FC function name are sent to the serverless platform. Upon receiving the request, configuration information, and the name of the FC function to be triggered from the API, the serverless platform creates the trigger and returns the trigger URL, which is the interface to call the function, i.e., the Agent interface referred to in this invention. This information, including function configuration information and trigger information, is then persisted to the FC function binding table. A single piece of information persisted in the FC function binding table includes, for example, a function name field, a trigger URL field, and a session field. Therefore, in this embodiment, when the request is initiated in step S24, an HTTP request is sent to the Serverless platform using the "POST" method. The request includes the trigger URL and adds session affinity information from the function configuration information, such as "x-session-hr-id" as mentioned above, to the header.

[0074] In step S25, after receiving the HTTP request, the Serverless platform determines the corresponding function name based on the trigger URL in the request, and then routes the request to the user-specific service agent instance based on the session affinity information.

[0075] Because the cold start of an Agent instance is time-consuming, in one embodiment, during the cold start process, the Serverless platform returns a creation notification information for the user's dedicated Agent service instance to the recruitment service platform's dialogue system. Based on this notification, the recruitment service platform's dialogue system generates a loading animation and displays it to the user on the front end. In another embodiment, after the user's dedicated Agent service instance is created, it remains resident during working hours. For example, the recruitment service platform's dialogue system periodically sends heartbeat requests to the Serverless platform during working hours. The Serverless platform responds to these heartbeat requests to maintain the resident state of the user's dedicated Agent service instance. The timing interval for sending heartbeat requests is less than the idle timeout threshold configured when the function is created. For example, when the idle timeout threshold is 30 minutes, the timing interval for sending heartbeat requests is 5 minutes or 10 minutes. The dedicated Agent service instance in a resident state is called a pre-warmed instance. This pre-warmed instance avoids frequent cold starts of the Agent container, thereby improving response speed. In another embodiment, the present invention also pre-creates public functions and creates and warms up corresponding public instances on the Serverless platform. When a user request is received, if there is no warmed-up dedicated Agent service instance, the public instance is assigned to the user to complete the corresponding business, and the public function is modified to be the user's dedicated function. Then, the public function and public instance are recreated. Through the above multi-level startup optimization, the response time of user requests is effectively reduced. Therefore, in a further embodiment, when it is determined in step S23 that no dedicated Agent function corresponding to the user identifier is found in the FC function binding table, the public function is queried in the FC function binding table. Then it is determined whether the public function is found. If not, step S28 is executed. If the public function is found, the public function is assigned to the user, and step S24 is executed. When assigning the public function to the user, the configuration information of the public function is modified. For example, the user identifier is added to the function name, the user identifier is added to the session affinity field information, and the storage address of the session information in the file system is modified. When making an HTTP request to the Serverless platform, the request method is "POST". The request includes the trigger URL, adds the original information of the session affinity field to the header, and also sends the modified function configuration information to the Serverless platform.Based on the received information, the serverless platform routes the request to the corresponding Agent service instance according to the original information in the session affinity field of the header. At the same time, it modifies the function configuration information in the platform according to the received configuration information, so that when the user initiates a conversation request again, it can determine the corresponding function based on the mapping between the trigger URL and the function name, and route the corresponding Agent service instance based on the new information in the session affinity field.

[0076] In another further embodiment, in step S25, it is first checked whether there is a pre-warmed dedicated agent service instance for the user. If there is, the pre-warmed dedicated agent service instance is reused and the request is routed to the user's dedicated service agent instance. If there is no pre-warmed dedicated agent service instance for the user, the user's dedicated agent service instance is created, that is, the container image is pulled, the container is started, and the agent is initialized, including loading the tool description, skill file, establishing a connection with the tool, etc., and then a session is created.

[0077] Furthermore, to reduce cold start time, in addition to the optimization methods mentioned above, this invention also optimizes the Agent image size. One method is to reduce the size of the full Python image (900MB) to a slim image (150MB), thereby reducing the fetch time for the Serverless platform. Another method is to optimize the Serverless platform's fetching of the Agent image by identifying necessary components from the full Python image and fetching only those components, thus reducing fetch time.

[0078] See Figure 7 , Figure 7 This is a flowchart illustrating a method for a user's dedicated Agent service instance to process a dialogue request and generate dialogue response information according to an embodiment of the present invention. Specifically, it includes the following steps: Step S261: The large language model is invoked to understand the user input information in the dialogue request based on the tool description and skill file. In one embodiment, through the interface provided by the large language model platform, the user input information "Help me find a product manager in Kunming, Bachelor's degree or above, 3+ years of experience, monthly salary 10k-15k", the tool description, and the skill file are added to the prompt words and sent to the large language model. The prompt words include a resume search process description, a task description from the large language model, etc. The tool description includes a list of tools and a functional description of each tool. In this embodiment, the tools include a search engine client, a standardized client, and the large language model. The search engine client is used to provide job information retrieval, resume search, and result formatting functions. The standardized client is used to implement data standardization, such as standardization of regional codes, industry classification, professional names, and tool functions. There are multiple skill files, each covering multiple filtering dimensions and multiple stages of the search process. Each skill file serves as a tool manual, defining one or more of the tool name, parameters, and description in YAML language. For example, in this embodiment, four filtering conditions are set: "Region", "Education", "Job", and "Person". The "Geographic" category includes two dimensions, the "Education" category includes seven dimensions, the "Work" category includes eight dimensions, and the "Person" category includes eight dimensions. The tool names, parameters, and descriptions required for each dimension are documented in a separate technical file. Corresponding skill files are set up for each stage of the search process, specifying the tasks, requirements, and tools needed for that stage.

[0079] In step S262, the large language model determines the task flow and the tools required for each stage of the task flow based on the descriptions in the prompts, the provided tool descriptions and skill files, and the user input information. Then, it executes from stage one to stage seven according to the following steps.

[0080] Step S263, Stage 1: Loading Target Job Information. Specifically, the large language model determines the tasks and tools required based on the skill file from Stage 1. When a user searches for talent for this job for the first time, the job information needs to be loaded from the recruitment platform's database. Therefore, the large language model sends a loading request for job information to the tool layer. After receiving this request, the MCP server in the tool layer first obtains the current target job name from the metadata, then retrieves the target job information through the job information retrieval function of the search engine client, and then passes it to the large language model. For users searching for talent for this job multiple times, since the initial session information is already stored in the file system, the job information from the recruitment platform can be obtained by loading the historical session information from the file system, and then combined with the user's input information to generate new recruitment requirements information.

[0081] Step S264, Stage Two: Extracting Filtering Dimension Information. Specifically, during the initial session, the large language model extracts multiple default filtering dimensions from the target job information. For subsequent sessions, filtering dimensions are extracted from newly generated job requirements information.

[0082] Step S265, Stage 3: Standardize the information across multiple filtering dimensions sequentially. Specifically, the large language model sends standardization requests for the corresponding dimensions to the MCP server in the tool layer based on the skill files. The MCP server performs the standardization processing sequentially and returns the results.

[0083] Step S266, Stage Four: Prioritization. The large language model treats the returned standardized filtering dimension information as a single filtering item. Based on the priority requirements defined in the skill file for this stage, it labels the returned multiple filtering items as essential or bonus items, and adjusts the labeling information based on its understanding of user input. For example, the skill file may define some filtering items as essential, some as bonus items, and others as either essential or bonus items. The technical file also requires prioritizing filtering items based on user intent. For instance, based on the tone of the input, some essential filtering items may be adjusted to bonus items.

[0084] Step S267, Stage Five: Generate Search Criteria. The large language model generates search criteria based on the extracted filter items, according to the description in the technical document for this stage. In one embodiment, based on the filter items extracted from the target job information, filter items are added or removed according to user input, and then the filter items are combined according to semantics, category, and Boolean operators such as AND and OR to form search criteria. The search criteria include at least two categories of filter items with different priority levels: required and bonus.

[0085] Step S268, Stage Six: Perform a search to obtain multiple resumes. Specifically, the large language model sends a standard request for tool invocation to the MCP server in the tool layer, specifying the search engine client and search criteria in the request. The MCP server in the tool layer invokes the resume search function of the search engine client, searches for resumes that meet the search criteria, and returns them to the large language model.

[0086] Step S269, Stage Seven: Analyze resumes and generate a pre-formatted resume list. The large language model analyzes the returned resumes, calculating and scoring them according to the scoring requirements and rules provided in the skills document for this stage. For example, the overall score for each resume is calculated using the following formula.

[0087] Overall Score = Responsibilities × 0.4 + Skills × 0.3 + Industry Score × 0.2 + Basic Score × 0.1 Then, we analyze the highlights of each resume compared to other resumes, such as outstanding work experience and outstanding skills.

[0088] The large language model then sends an output information formatting request to the MCP server in the tool layer. The MCP server calls the result formatting function of the search engine client to organize the resume, rating, and highlights into a pre-formatted information and returns it to the large language model.

[0089] Then, in step S27, the system returns to the front end of the dialogue system and displays it to the user.

[0090] Alternatively, in another embodiment, after generating search criteria in stage five, the generated search criteria can be sent to the user. Upon receiving confirmation from the user, stages six and seven can then proceed. If the user modifies the generated search criteria, the search criteria are regenerated based on the user's modifications until the user confirms.

[0091] Furthermore, in some embodiments, during the aforementioned task flow processing, at each stage, the large language model generates corresponding response information and returns it to the front end of the dialogue system for display to the user. The response information may be, for example, a response to user input, or a description of the task being processed at the current stage and a description of the processing result. Figure 8 As shown, Figure 8 This is a schematic diagram of a dialog system front-end application page during the execution of phase two of a task flow according to an embodiment of the present invention, showing the processing results of phase one and the task description of phase two. Figure 9 As shown, Figure 9 This is a schematic diagram of the front-end application page of a dialogue system during stage six of a task flow execution, according to an embodiment of the present invention. It displays the search criteria generated in stage five, with required and priority items marked, along with explanatory information requesting user confirmation. After user confirmation, stage six is ​​executed to perform a resume search. Figure 10 This is a schematic diagram of the front-end application page of a dialogue system after the task flow has been completed, according to an embodiment of the present invention. It displays a formatted resume list for illustrative purposes. Figure 10 Only one resume search result is displayed. The number of search results in the resume list can be described in the prompts or skill files of the large language model. Resume search results include resume information and resume analysis information. Resume information includes, for example, the job seeker's job objective, basic personal information, and key work experience. Resume analysis information mainly describes the match between the current resume information and the job posting. Of course, other information may also be included, such as using icons to indicate the degree of match between the resume and the job, providing access to detailed reports, and communication channels.

[0092] In a further embodiment, to enable the large language model to provide accurate response information when processing user input, in step S261, when understanding the user input information in the dialogue request, historical session information is further read from the mounted file system storing the user's session information, and the historical session information is sent to the large language model, allowing the large language model to understand the user input information in the dialogue request based on the historical session information, tool descriptions, and skill files. Therefore, the user does not need to repeatedly describe their needs and can continue the same round of dialogue across devices and time, improving the user experience. Furthermore, when loading the historical session information of the user's target business from the file system, the most recent preset round (e.g., 10 rounds) of complete session information is loaded; if the number of rounds in the historical session information is greater than the preset rounds, a summary of the complete session information greater than the preset rounds is obtained and loaded, thus effectively saving AI computing power and reducing AI costs.

[0093] Furthermore, during the dialogue, the dialogue information is stored in the user's session information set in the file system, and the session information is isolated by the target business identifier. The dialogue information includes user input information in the user request and the result information returned to the user. Therefore, even if the Serverless platform reclaims the user-specific Agent service instance or the dialogue connection is accidentally interrupted, the session information will not be lost and can be loaded from the file system when a new instance of the user-specific Agent service starts.

[0094] Although the above example illustrates the dialogue system used by recruiters, it is clear that a similar dialogue system can be provided to job seekers to offer job search services, which will not be elaborated upon here.

[0095] For recruitment service platforms, dialogue systems require continuous iteration and optimization, including adding interactive skills, optimizing dialogue response logic, fixing system bugs, and adapting to new business rules, resulting in a high frequency of iterations and updates. In existing technologies, traditional dialogue systems typically require full-system maintenance or phased shutdowns during new version upgrades, making it impossible to achieve second-level gray-scale switching and rapid rollback. If a new version has issues, the impact is significant. To address this problem, this invention, after creating the function of the Agent that implements the target business, also calls the Serverless platform's function alias creation interface (such as CreateAliasRequest) to create a function alias for the Agent that implements the target business, sets the alias name (such as last), the new version number of the Agent container image (such as v2.0), and configures the Agent container image version information and its traffic weight. For example, during normal operation, for 100% of requests, the function alias points to the old version of the Agent container image (e.g., v1.0). During the canary release phase, x1% of requests point to the new version of the Agent container image, and x2% of requests point to the old version of the Agent container image, where x1 + x2 = 100. During the full release of the new version, for 100% of requests, the function alias points to the Agent container image. Then, the function alias and configuration information of the Agent implementing the target business are persisted to the FC function binding table, and a binding relationship is established with the corresponding user. When the Serverless platform routes a user's dialogue request to the user's dedicated Agent service instance, it determines the corresponding version of the dedicated Agent service instance based on the function alias and traffic weight carried in the Agent interface. As can be seen from the solution provided by the present invention, business upgrades can be achieved without being noticed through canary releases, and version switching or rollback can be achieved in seconds by modifying traffic weights. This can effectively prevent the normal use of all or a large number of users from being directly affected by defects such as logical vulnerabilities, adaptation failures, and compatibility issues in the new version, thus effectively improving the stability of system operation and the reliability of services.

[0096] See Figure 11 , Figure 11 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. The electronic device can be implemented as a server or terminal device in the aforementioned recruitment service platform architecture, including a processor 601 and a memory 602. The memory 602 stores a program instruction set. When the processor 601 executes the program instruction set in the memory 602, the aforementioned dialogue method for realizing the target business is implemented.

[0097] Specifically, the processor 601 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0098] Memory 602 may include mass storage for data or instructions. For example, and not limitingly, memory 602 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 602 may include removable or non-removable (or fixed) media. Where appropriate, memory 602 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 602 is non-volatile solid-state memory.

[0099] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the dialogue method for implementing the target service provided by this invention.

[0100] In one example, the electronic device may also include a communication interface 603 and a bus 604. The processor 601, memory 602, and communication interface 603 are connected via the bus 604 and communicate with each other.

[0101] The communication interface 603 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of the present invention.

[0102] Bus 604 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 604 may include one or more buses. While specific buses are described and illustrated in embodiments of the invention, the invention contemplates any suitable bus or interconnect.

[0103] The present invention also provides a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement the dialogue method for achieving the target service as described in any of the foregoing embodiments. The computer-readable storage medium can be any medium that can tangibly contain or store computer-executable instructions for use by or in connection with an instruction execution system, apparatus, or device. The storage medium can be a transient computer-readable storage medium or a non-transitory computer-readable storage medium. Non-transitory computer-readable storage media may include, but are not limited to, magnetic storage devices, optical storage devices, and / or semiconductor storage devices. Examples of such storage devices include, for example, magnetic disks, optical discs based on CD, DVD, or Blu-ray technology, and persistent solid-state storage such as flash memory and solid-state drives.

[0104] This invention also provides a computer program product, comprising a set of computer program instructions, which, when executed by a processor, implements any of the dialogue methods for achieving the target business described in the foregoing embodiments. The computer program product includes, but is not limited to, application installation packages published on websites and in app stores, application plugins, and mini-programs that can run within certain applications.

[0105] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0106] The above embodiments are for illustrative purposes only and are not intended to limit the invention. Those skilled in the art can make various changes and modifications without departing from the scope of the invention. Therefore, all equivalent technical solutions should also fall within the scope of the invention.

Claims

1. A dialogue method for achieving target business, applied to a recruitment service platform, characterized in that, include: Parse the dialogue requests initiated by users at the front-end application layer of the recruitment service platform to obtain user identifiers; The Agent interface is obtained based on the user identifier; wherein, the Agent is configured with a tool description for completing the target business and one or more independent skill files; The user's conversation request is sent to the Serverless platform via the Agent interface; The serverless platform routes the user's conversation request to the user's dedicated Agent service instance; The user's dedicated Agent service instance processes the dialogue request and generates dialogue response information; and The serverless platform returns the dialogue response information to the recruitment service platform and then back to the application layer to display it to the user. The steps for the user's dedicated Agent service instance to process the dialogue request and generate dialogue response information include: The large language model is invoked to understand user input information in dialogue requests based on tool descriptions and skill files, determine the task flow to complete the target task, and identify the tools required for each stage of the task flow; and The task flow is executed, and response information for the dialogue request is generated based on the execution result of the task flow.

2. The method according to claim 1, characterized in that, When obtaining the Agent interface based on the user identifier, the system queries the pre-set FC function binding table for the dedicated Agent function and its interface that implements the target business for the user. The FC function binding table records the dedicated Agent function and its interface information that are bound to the user identifier to implement the target business.

3. The method according to claim 2, characterized in that, In response to the absence of a function and its interface for the Agent that implements the target service for the user from the pre-configured FC function binding table, the method further includes: The function is called through the Serverless platform's function creation interface to create an Agent for implementing the target business for the user. Create an interface to call the function; and The function and its calling interface are persisted to the FC function binding table, and a binding relationship with the user is established.

4. The method according to claim 3, characterized in that, When calling the function creation interface of the serverless platform to create an Agent for implementing the target business for the user, at least the following information should be configured: The function name should be set with at least the target business information and user identifier; Configure the function's runtime environment as a runtime environment, and specify the runtime as the Agent container image; Set session affinity and include session fields and user identifiers in the session affinity information; and Configure the file system information to be mounted.

5. The method according to claim 3, characterized in that, The step of creating an interface for calling the function further includes: The serverless platform's function trigger creation interface is used to create a trigger for the function, and the trigger name, type, corresponding function name, and triggering rules are configured; and Receive the trigger URL returned by the Serverless platform, where the trigger URL is the Agent interface that calls the function.

6. The method according to claim 1, characterized in that, The serverless platform routes the user's conversation requests to the user's dedicated Agent service instance based on the session affinity information carried by the Agent interface.

7. The method according to claim 6, characterized in that, After receiving the dialogue request information transmitted through the Agent interface, the Serverless platform further includes: Check if a pre-warmed dedicated Agent service instance for the user exists in the Serverless platform; and In response to the existence of a pre-warmed dedicated Agent service instance for the user in the Serverless platform, the pre-warmed dedicated Agent service instance is reused.

8. The method according to claim 7, characterized in that, In response to the absence of a pre-warmed dedicated agent service instance for the user in the serverless platform, the method further includes the step of creating a dedicated agent service instance for the user, including: Based on the function configuration information corresponding to the user, pull the Agent container image to start the Agent container; The agent container creates a session and loads the configured list of tools and skill files for the target business; and Establish a connection with the tools in the tool list.

9. The method according to claim 8, characterized in that, The steps of creating a dedicated Agent service instance for the user further include: The serverless platform returns a notification message about the creation of the user's dedicated Agent service instance to the recruitment service platform; and The recruitment service platform generates a loading animation based on the creation notification information of the user's exclusive Agent service instance and displays it to the user on the front end.

10. The method according to claim 8, characterized in that, When pulling the Agent container image based on the function configuration information corresponding to the user, the version of the Agent container image to be pulled is determined based on the Agent container image alias, and / or the Agent container image is optimized when pulling the Agent container image based on the optimization configuration information.

11. The method according to claim 4, characterized in that, When calling the function creation interface of the serverless platform to create an Agent for implementing the target business for the user, the idle timeout threshold information of the instance is also configured. Correspondingly, after the user's dedicated Agent service instance is created, the Serverless platform monitors the idle time of the user's dedicated Agent service instance, and when the idle timeout threshold is reached, the user's dedicated Agent service instance is reclaimed.

12. The method according to claim 11, characterized in that, After the user's dedicated Agent service instance is created, the user's dedicated Agent service instance is kept in a persistent state during working hours.

13. The method according to claim 12, characterized in that, The recruitment service platform sends heartbeat requests to the serverless platform at regular intervals during working hours. The serverless platform responds to the heartbeat requests by maintaining the persistent state of the user's dedicated Agent service instance. The timing period for sending heartbeat requests is less than the idle timeout threshold.

14. The method according to claim 4, characterized in that, Further includes: Call the function alias creation interface of the Serverless platform to create a function alias that implements the target business, and configure the Agent container image version information and its traffic weight. as well as The function aliases and configuration information of the Agent that implements the target business are persisted to the FC function binding table, and a binding relationship with the corresponding user is established.

15. The method according to claim 14, characterized in that, When routing a user's conversation request to the user's dedicated Agent service instance, the Serverless platform determines the corresponding version of the dedicated Agent service instance based on the function alias and traffic weight carried in the Agent interface.

16. The method according to claim 2, characterized in that, Further includes: Create public functions and interfaces for the Agent that implement the target business, and persist them to the FC function binding table; Correspondingly, the Serverless platform creates instances of the aforementioned public functions as reserved public instances during working hours.

17. The method according to claim 16, characterized in that, In response to the fact that no function and its interface for the Agent to implement the target business for the user are found in the preset FC function binding table, the public function interface is obtained as the Agent interface; Correspondingly, the Serverless platform routes the user's conversation requests to a reserved public instance.

18. The method according to claim 1, characterized in that, The steps for the user's dedicated Agent service instance to process the dialogue request and generate dialogue response information also include: Load the user's target business historical session information from the file system and send the historical session information to the large language model, enabling the large language model to understand the user input information in the dialogue request based on the historical session information, tool descriptions, and skill files; and The current session information is stored in the user's session information set in the file system, and the session information is isolated by the target business identifier. The current session information includes user input information and corresponding response information in the user request.

19. The method according to claim 18, characterized in that, When loading the user's target service's historical session information from the file system, load the complete session information of the most recent preset round; if the round of the historical session information is greater than the preset round, obtain a summary of the complete session information of the round greater than the preset round, and load the summary.

20. The method according to claim 1, characterized in that, The target business is the resume screening business corresponding to the target position. There are multiple skill files to complete the target business, and each skill file corresponds to a screening dimension. The corresponding tools include search engine clients, standardized clients, and large language models.

21. The method according to claim 20, characterized in that, The steps for executing the task flow to obtain the response information of the dialogue request include: Use a search engine client to obtain job information for the target position; The large language model is used to extract the corresponding screening criteria from the job information of the target job based on the skill files corresponding to the screening dimensions. The extracted filtering criteria are standardized using a standardized client to obtain standardized filtering criteria; A large language model is used to generate search criteria based on filtering conditions extracted from user input and job information. The search engine client is used to search a pre-set number of resumes based on search criteria. A large language model is used to analyze the preset number of resumes to obtain the analysis results for each resume; and The system uses a search engine client or a standardized client to output resume search results in a pre-formatted format. The resume search results include resume information and resume analysis results; wherein, the resume search results serve as response information to the dialogue request.

22. The method according to claim 21, characterized in that, The analysis results of the resumes include resume scoring instructions and / or descriptions of highlights.

23. The method according to claim 21, characterized in that, After generating search criteria using a large language model based on the filtering conditions extracted from user input information and job information, the search criteria are returned to the user. After receiving confirmation from the user regarding the search criteria, the search engine client is invoked, enabling the search engine client to search for a preset number of resumes based on the search criteria.

24. The method according to claim 21, characterized in that, The search criteria include multiple search options, which correspond to at least two or more different priorities.

25. A dialogue system for achieving target business, applied to a recruitment service platform, characterized in that, include: The front-end interface layer is configured to receive and parse the dialogue request initiated by the user in the front-end application layer of the recruitment service platform to the target business in order to obtain the user identifier, and return the dialogue response information to the front-end application layer of the recruitment service platform to be displayed to the user. The AI ​​service layer is configured to obtain the Agent interface based on the user identifier; The user's dialogue request is sent to the function computation layer hosted by the serverless platform via the Agent interface. The Agent is configured with tool descriptions and one or more skill files for completing the target business task. The function computation layer is configured to route the user's dialogue request to the user's dedicated Agent service instance; the user's dedicated Agent service instance processes the dialogue request and generates dialogue response information, which is then returned to the AI ​​service layer. The large language model layer, configured to provide a large language model, responds to calls from Agent service instances, understands user input in dialogue requests based on tool descriptions and skill files, determines the task flow to complete the target task and the tools required for each stage of the task flow, and generates dialogue response information based on the execution results of the task flow; and The tool layer is configured to respond to tool calls made by the Agent service instance when performing phase tasks, and provides the corresponding tools to complete the phase tasks.

26. An electronic device comprising a processor and a memory, wherein the memory stores a set of computer program instructions, characterized in that, The processor executes the dialogue method for implementing the target service as described in any one of claims 1-24 when executing a set of computer program instructions in memory.

27. A computer-readable storage medium, wherein, The computer-readable storage medium stores a set of computer program instructions, characterized in that, when the set of computer program instructions is executed by a processor, it executes the dialogue method for implementing the target service as described in any one of claims 1-24.

28. A computer program product comprising a computer program instruction set, characterized in that, When the computer program instruction set is executed by the processor, it performs the dialogue method for implementing the target service as described in any one of claims 1-24.

Images