A target object search method and device

Abstract

The application discloses a target object search method and device, relates to the technical field of intelligent search and artificial intelligence, and triggers cross-platform target object search by dispatching sub-tasks to each platform sub-agent through a main agent, calling platform tools within the range of corresponding tool sets of each platform sub-agent, and adapting to access mechanisms of different types of platforms; when a preset suspension condition is triggered, task execution is suspended, and the conversation history and acquired target object information of each agent are retained; and the task is continued from the breakpoint by receiving a recovery request. In this way, a user only needs to submit a natural language task description to trigger cross-platform target object search, manual operation is reduced, and the task can be recovered from the breakpoint when an exception occurs, so that data loss and task re-execution are avoided.

Description

Technical Field

[0001] This application relates to the fields of intelligent search and artificial intelligence technology, and in particular to a method and apparatus for searching target objects. Background Technology

[0002] In recruitment scenarios, companies typically need to actively search for talent on multiple recruitment platforms to find suitable candidates. Currently, this process is highly reliant on manual operation: recruiters must first read and understand the job description, extracting keywords, city, experience requirements, education requirements, and other search criteria; then switch between multiple recruitment platforms, entering search criteria, browsing candidate lists, and clicking to view candidate details one by one; finally, they manually assess the match between candidates and the job requirements. Throughout this process, platform switching, repeated input of criteria, and individual candidate screening involve a large amount of repetitive manual work, resulting in low search efficiency. When multiple platforms need to be covered, these repetitive operations are further amplified, significantly increasing both labor and time costs.

[0003] Therefore, there is an urgent need for a search execution solution that can be driven by a large language model and cover everything from demand understanding to multi-platform data acquisition, so as to reduce the manual operation of recruiters in the process of searching for target objects. Summary of the Invention

[0004] To address the above problems, this application provides a method and apparatus for searching target objects, including the following: Firstly, this application provides a target object search method applied to a server, the method comprising: Obtain the natural language task description submitted by the user, create a target object search task based on the natural language task description, and start the main intelligent agent; The main intelligent agent generates a target object search subtask based on the natural language task description, and dispatches the target object search subtask to each platform sub-intelligent agent. Each platform sub-intelligent agent includes a sub-intelligent agent corresponding to an interface-type recruitment platform and a sub-intelligent agent corresponding to a page-type recruitment platform. The target object search subtask is a subtask generated by the main intelligent agent based on the target object search task and dispatched to each platform sub-intelligent agent for execution. Each platform sub-agent invokes platform tools within its respective toolset to obtain target object information corresponding to the target object search sub-task. When a preset pause condition is triggered, the execution of the target object search task is paused, and the dialogue history of the main agent and the platform sub-agents, as well as the acquired target object information, are retained. Upon receiving a resume request containing a pause indicator from the client, the target object search task is resumed from the breakpoint.

[0005] Optionally, the first toolset configured for the sub-agents of the interface-type recruitment platform is different from the second toolset configured for the sub-agents of the page-type recruitment platform; each platform sub-agent maintains its own independent dialogue history and execution state.

[0006] Optionally, when the preset pause condition is triggered, the method further includes: Generate a pause event containing the pause identifier and send the pause event to the client.

[0007] Optionally, the method further includes: The target object information acquired by each platform sub-agent is normalized to generate a target object record in a unified format. Access the company's existing talent pool and query the matching results of the target object records in the company's existing talent pool in the unified format. If a match is found, the contact information of the target individual in the company's existing talent pool will be retrieved.

[0008] Optionally, the query results for the target object records in the unified format within the enterprise's existing talent pool include: Construct a multi-dimensional search strategy based on the target object records in the unified format; The search interface of the enterprise's existing talent pool is invoked, and candidate matching results are obtained based on the multi-dimensional search strategy; The candidate matching results are deduplicated to obtain the deduplicated candidate matching results; A preliminary matching score is calculated based on the degree of overlap between the target object record in the unified format and the deduplicated candidate matching results in at least one field. Obtain detailed information about candidate matching results whose preliminary matching scores meet preset conditions; Calculate the matching confidence score between the candidate matching result and the target object record in the unified format based on the detailed information; Based on whether the matching confidence score reaches a preset threshold, it is determined whether they are the same target object.

[0009] Optionally, after the current round of activities of the main intelligent agent and each platform sub-intelligent agent has ended, the method further includes: Send an idle event to the client; Receive additional constraints submitted by the client after the idle event; The additional constraints are added to the context of the target object search task, and the saved dialogue history and the acquired target object information are reused to continue executing the target object search task.

[0010] Secondly, this application provides a target object search method applied to a client, the method comprising: Submit a natural language task description to the server; the natural language task description is used to trigger the server to create a target object search task and start the main intelligent agent, the main intelligent agent generates a target object search sub-task according to the target object search task and dispatches it to each platform sub-intelligent agent for execution; The system receives a pause event sent by the server. The pause event is generated by the server when a preset pause condition is triggered during the execution of the target object search subtask. The pause event includes a pause identifier. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. Display pause pop-up; In response to the user's confirmation of the pause pop-up, a resumption request containing the pause identifier is submitted to the server. The resumption request is used to instruct the server to continue executing the target object search subtask from the breakpoint, and to retain the dialogue history of each agent and the target object information already acquired.

[0011] Optionally, the method further includes: Display the task homepage, which includes a natural language task input box; Display a browser status bar, which is used to display the browser's running status and the login status of each recruitment platform; Display the task details page, which includes a task planning panel and a platform panel area; The task planning panel is used to display the search plan generated by the main intelligent agent and the execution status of each step; The platform panel area is used to display the operating status, search keywords, number of candidates, and number of hits of each platform sub-agent side by side. Display a candidate panel, which is used to display saved candidate information; Upon receiving the pause event sent by the server, a pause pop-up window that cannot be closed directly is displayed, showing the reason for the pause and an operation confirmation entry. Upon receiving the idle event sent by the server, the input box is switched to a follow-up question state.

[0012] Optionally, the pause pop-up displays corresponding prompt information based on the pause reason carried by the pause event. The pause reason includes at least one of the following: needing to scan a QR code to log in, needing to open a browser, needing to return to the search page, or needing manual confirmation.

[0013] Optionally, the candidate panel displays the saved candidate information, including the candidate's match results and contact information in the company's existing talent pool.

[0014] Optionally, the method further includes: In response to the user's input operation after the input box switches to the follow-up question state, an additional constraint condition is submitted to the server. The additional constraint condition is used to trigger the server to add the additional constraint condition to the context of the target object search task, and reuse the saved dialogue history and the acquired target object information to continue to execute the target object search task.

[0015] Thirdly, this application provides a target object search device applied on a server, the device comprising: The task creation module is used to obtain the natural language task description submitted by the user, create a target object search task based on the natural language task description, and start the main intelligent agent; The agent scheduling module is used to enable the main agent to generate a target object search subtask based on the natural language task description, and to dispatch the target object search subtask to each platform sub-agent. The platform sub-agents include sub-agents corresponding to the interface-type recruitment platform and sub-agents corresponding to the page-type recruitment platform. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. The tool invocation module is used to enable each platform sub-agent to invoke platform tools within its respective toolset to obtain target object information corresponding to the target object search sub-task. The pause and resume module is used to pause the execution of the target object search task when a preset pause condition is triggered, retain the dialogue history of the main agent and the platform sub-agents as well as the acquired target object information; and receive a resume request containing a pause identifier returned by the client to continue the execution of the target object search task from the breakpoint.

[0016] Fourthly, this application provides a target object search device applied to a client, the device comprising: The task submission module is used to submit a natural language task description to the server. The natural language task description is used to trigger the server to create a target object search task and start the main intelligent agent. The main intelligent agent generates a target object search sub-task according to the target object search task and dispatches it to each platform sub-intelligent agent for execution. The pause interaction module is used to receive a pause event sent by the server. The pause event is generated by the server when a preset pause condition is triggered during the execution of the target object search subtask. The pause event includes a pause identifier and displays a pause pop-up window. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. The recovery request module is used to respond to the user's confirmation operation of the pause pop-up window by submitting a recovery request containing the pause identifier to the server. The recovery request is used to instruct the server to continue executing the target object search subtask from the breakpoint and retain the dialogue history of each agent and the target object information already obtained.

[0017] Fifthly, this application provides an apparatus comprising a memory and a processor, the memory for storing instructions or code, and the processor for executing the instructions or code to cause the apparatus to perform the target object search method described in any of the implementations of the first and second aspects.

[0018] Sixthly, this application provides a computer-readable storage medium storing code, wherein when the code is executed, a device executing the code implements the target object search method described in any of the implementations of the first and second aspects above.

[0019] This application provides a method for searching for target objects. When executing the method, firstly, a natural language task description submitted by the user is obtained through the server; a target object search task is created based on this description, and the main agent is started. Then, the main agent generates a target object search subtask based on the natural language task description and dispatches the subtask to each platform sub-agent. Next, each platform sub-agent invokes platform tools within its corresponding toolset to obtain target object information corresponding to the subtask. Finally, when a preset pause condition is triggered, the execution of the target object search task is paused, the dialogue history between the main agent and each platform sub-agent, as well as the obtained target object information, are preserved. A resumption request containing a pause marker is received from the client, and the target object search task resumes execution from the breakpoint.

[0020] In this way, the main intelligent agent decomposes the search task into sub-tasks and dispatches them to various platform sub-intelligent agents. Each platform sub-intelligent agent executes cross-platform searches in parallel, avoiding users manually switching between multiple platforms and repeatedly entering search conditions, thus reducing manual operation steps in the search process. By having each platform sub-intelligent agent call platform tools within its corresponding toolset, the platform characteristics and access methods of interface-based and page-based recruitment platforms can be adaptively adapted. The heterogeneity of different platforms is encapsulated within the toolsets of each sub-intelligent agent, eliminating the need for a unified implementation method for different platforms. This ensures compatibility with multiple platforms while maintaining the stability of search execution on each platform. By pausing task execution when a preset pause condition is triggered and retaining the dialogue history of the main intelligent agent and each platform sub-intelligent agent, as well as the acquired target object information, data and execution context are not lost when exceptions occur during task execution. By receiving a recovery request to continue execution from the breakpoint, the task can resume execution after exception handling without restarting the entire search process, reducing the waste of computing resources and time caused by abnormal interruptions. This allows users to trigger cross-platform target object searches simply by submitting a natural language task description, reducing manual operations during the search process. Furthermore, it enables resumption of execution from the breakpoint when an exception occurs during the search process, avoiding the loss of acquired data and the need for task re-execution. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings used in the description of the embodiment or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the structure of a target object search system provided in an embodiment of this application; Figure 2 A flowchart illustrating a target object search method provided in an embodiment of this application; Figure 3 A flowchart illustrating another target object search method provided in this application embodiment; Figure 4 A flowchart illustrating yet another target object search method provided in this application embodiment; Figure 5 This is a schematic diagram of the structure of a target object search device provided in an embodiment of this application; Figure 6 This is a schematic diagram of another target object search device provided in an embodiment of this application. Detailed Implementation

[0023] As described in the background section of this application, in recruitment scenarios, companies typically need to actively search for talent on multiple recruitment platforms to obtain suitable candidates. In existing technologies, this process is highly reliant on manual operation: recruiters first need to read and understand the job description, extracting keywords, city, experience requirements, education requirements, and other search criteria; then switch between multiple recruitment platforms, entering search criteria, browsing candidate lists, clicking to view candidate details one by one, and finally manually judging the match between candidates and positions. Throughout this process, platform switching, repeated input of criteria, and candidate screening involve a large amount of repetitive manual operation, resulting in low search efficiency. When multiple platforms need to be covered, these repetitive operations are further amplified, significantly increasing labor and time costs.

[0024] To address the aforementioned technical problems, this application provides a target object search method. When executing the method, firstly, the server obtains the natural language task description submitted by the user, creates a target object search task based on the natural language task description, and starts the main intelligent agent; then, the main intelligent agent generates a target object search subtask based on the natural language task description and dispatches the subtask to each platform sub-intelligent agent; then, each platform sub-intelligent agent calls platform tools within its corresponding toolset to obtain the target object information corresponding to the subtask; finally, when a preset pause condition is triggered, the execution of the target object search task is paused, the dialogue history between the main intelligent agent and each platform sub-intelligent agent, as well as the obtained target object information, are preserved, and a resumption request containing a pause indicator is received from the client, allowing the target object search task to resume execution from the breakpoint. This allows users to trigger cross-platform target object searches simply by submitting a natural language task description, reducing manual operations during the search process, and enabling resumption from the breakpoint when an exception occurs during search execution, avoiding data loss and task re-execution.

[0025] The method provided in this application is applied to a target object search system, which includes a server and a client, for searching target objects on multiple platforms.

[0026] Figure 1 This is a schematic diagram of the structure of a target object search system provided in an embodiment of this application, as shown below. Figure 1 As shown in the embodiments of this application, the target object search system includes a server and a client. The server can be a single server, a server cluster consisting of multiple servers, or a cloud computing service center. The client can be a personal computer, laptop, smartphone, tablet, or other terminal device. The client and server communicate via a wired or wireless network.

[0027] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application 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. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0028] Figure 2 A flowchart illustrating a target object search method provided in this application embodiment, applied to a server. (Combined with...) Figure 2 As shown, the target object search method provided in this application embodiment may include: S201. Obtain the natural language task description submitted by the user, create a target object search task based on the natural language task description, and start the main intelligent agent.

[0029] Natural language task descriptions refer to job target descriptions entered by users in natural language, such as "Help me find someone with 3-5 years of experience in Go language middleware in Beijing." This description directly expresses the user's search intent, eliminating the need for users to fill in form fields such as platform, city, experience, education, keywords, and page numbers. After obtaining the natural language task description, the server creates a target object search task and assigns a task identifier, while simultaneously launching the main intelligent agent. This main intelligent agent is an execution unit built based on a large language model, possessing natural language understanding, task planning, tool invocation, and reasoning decision-making capabilities. It can perform multiple rounds of reasoning based on the user-input task description and invoke appropriate tools to complete the task. In this embodiment, the main intelligent agent is responsible for overall planning, platform dispatch, result aggregation, and human-computer interaction.

[0030] S202. The main intelligent agent generates a target object search subtask based on the natural language task description, and dispatches the target object search subtask to each platform sub-intelligent agent.

[0031] The platform sub-intelligent agents include sub-intelligent agents corresponding to interface-type recruitment platforms and sub-intelligent agents corresponding to page-type recruitment platforms; the target object search sub-task is a sub-task generated by the main intelligent agent based on the target object search task and assigned to the platform sub-intelligent agents for execution.

[0032] After obtaining the natural language task description, the main agent parses the description to identify constraints such as target job title, city, experience, education, number of people, and platform scope. For example, from the task "Find me 5 people with 3-5 years of experience in Go language middleware in Beijing," the target job title is "Go language middleware," the city is "Beijing," the experience is "3-5 years," and the target number of people is "5 of each." Based on the parsing results, the main agent generates a cross-platform search plan. This plan includes the platform sub-agents to be scheduled, their corresponding recruitment platforms, keywords to be executed on each platform, and filtering conditions. Subsequently, the main agent determines the platform sub-agents to be scheduled based on the platform scope constraints and dispatches the target object search sub-tasks to each platform sub-agent at the same level. All platform sub-agents execute concurrently. If a sub-agent is already running on the same platform, the server rejects the second sub-task on the same platform and returns a busy result to avoid contention for the same browser tab or login state. Each platform sub-agent maintains its own independent dialogue history and execution state, without interfering with each other.

[0033] In this embodiment, the recruitment platform refers to a third-party platform that provides talent search services. Depending on the access mechanism, it includes at least two types: interface-based recruitment platforms and page-based recruitment platforms. Interface-based recruitment platforms provide standardized structured data interfaces and support programmatic queries. Page-based recruitment platforms provide search services through webpage interaction. Each platform sub-agent is an intelligent agent execution unit on the server corresponding to a specific recruitment platform. Each platform sub-agent corresponds to one recruitment platform, is configured with a toolset matching the characteristics of that platform, and each platform sub-agent maintains its own independent dialogue history and execution state.

[0034] The first toolset configured for the sub-agents of the interface-based recruitment platform is different from the second toolset configured for the sub-agents of the page-based recruitment platform.

[0035] Specifically, the first toolset configured for the sub-agents of the interface-based recruitment platform includes a structured search tool, a historical candidate query tool, a batch matching tool for existing enterprise talent pools, a candidate saving tool, and a pause / resume tool. The second toolset configured for the sub-agents of the page-based recruitment platform includes a page status check tool, a navigation tool, a search tool, a candidate card reading tool, a limited scrolling loading tool, a batch matching tool for existing enterprise talent pools, a candidate saving tool, and a pause / resume tool. The system does not expose tools such as page status checks, page navigation, page document model list parsing, scrolling loading, click filters, arbitrary script execution, and screenshot escape to the sub-agents of the interface-based recruitment platform. When scrolling triggers anti-crawling redirection or an abnormal login status, the sub-agents of the page-based recruitment platform do not continue scrolling or searching repeatedly; instead, they re-check the page status or request manual intervention. Each platform's sub-agents are restricted to operating within their own toolset and are not allowed to dispatch sub-tasks to each other, nor are they allowed to directly manipulate the pages of another platform or call the tools of another platform.

[0036] S203. Each platform sub-agent calls platform tools within its corresponding toolset to obtain target object information corresponding to the target object search sub-task.

[0037] After receiving the target object search subtask, each platform sub-agent calls the corresponding platform tools to perform the search based on its configured toolset.

[0038] For API-based recruitment platforms, the corresponding sub-agent invokes a structured search tool to retrieve candidate information. This structured search tool first connects to a locally started and logged-in browser via a browser debugging protocol, reading the login credentials and browser identity identifier under the API-based recruitment platform's domain. The login credentials are login status information stored by the browser to verify the user's identity; the browser identity identifier is information used by the platform to identify the request's origin environment. If the login credentials are missing or expired, the structured search tool returns a "Login Required" error and triggers a pause / resume process. If the login credentials are valid, the structured search tool encodes search parameters such as keywords, page number, city, experience, education, salary, and age into the request address, attaching the real login credentials, browser identity identifier, and legitimate source address in the request header, and initiates a request to the platform's candidate structured interface. This request directly calls the platform's backend interface via the HTTP protocol to obtain structured candidate data, rather than through page clicks or page document model parsing. After receiving the response from the platform, the structured search tool categorizes the results: if the response indicates login failure, it requests manual login; if the response indicates a server error, it initiates a retry; if the requests are too frequent or the platform's risk control system rejects the request, it stops searching on that platform. The structured search tool parses the candidate array returned in the response, extracting fields from each candidate record, including the platform primary key, security token, name, city, age, experience, salary, education, current company, current position, school, major, activity level, work experience, and tags, to generate a structured list of candidates. It also returns a marker indicating whether there is a next page.

[0039] For page-based recruitment platforms, the corresponding sub-agent invokes page status checking tools, search tools, and candidate card reading tools to obtain candidate information. The platform sub-agent first invokes the page status checking tool, which checks whether the browser is running, whether the user is logged in, whether they are on the enterprise version talent search page, whether the search input box is rendered, the current keywords, and the number of visible candidate cards, returning actions requiring manual intervention. If the page status checking tool returns a need to open a browser or scan a QR code to log in, the system enters a pause-and-resume process; if it returns a need to return to the search page, the page-based platform tools invoke the navigation tool to navigate to the enterprise version talent search page and check the page status again. Once the page is ready, the platform sub-agent invokes the search tool, enters keywords in the search box, triggers the search, and waits for the results to load. After the search results are loaded, the platform's sub-agent invokes a candidate card reading tool. This tool reads the candidate cards on the current search results page, automatically skipping locked cards. Locked cards are those for which the platform requires payment to unlock contact information or complete resumes. The tool returns the candidate's name, activity level, city, education, years of experience, age, gender, current company, current position, previous companies, school, skill tags, highlight text, and original summary for each card. When the number of unlocked candidates read is insufficient, the platform's sub-agent can invoke a limited scrolling loading tool for scrolling. If scrolling triggers platform anti-scraping measures or page redirection, the platform's sub-agent will not continue scrolling but will instead re-examine the page status and use the already acquired candidates.

[0040] Each platform sub-agent outputs the structured candidate information it acquires to a unified processing flow. At the same time, it sends the search keywords, hit count, tool calls, and tool results during the execution process to the client in real time for display via server push events.

[0041] S204. When a preset pause condition is triggered, the execution of the target object search task is paused, and the dialogue history of the main agent and the platform sub-agents, as well as the acquired target object information, are retained.

[0042] During the execution of a target object search task, platform tools or platform sub-agents may encounter situations requiring manual intervention to continue. Preset pause conditions may include invalid or non-existent login credentials, browser not running, requiring user login via QR code, page redirection to a non-search page, or the need for manual confirmation. When any of these preset pause conditions are triggered, the server pauses the execution of the current target object search task, retaining the dialogue history of the main agent and each platform sub-agent, as well as the acquired target object information. Specifically, the server retains the complete dialogue context of the main agent, the independent dialogue history of each platform sub-agent, the structured candidate data acquired by each platform, and the overall execution status of the current task. Simultaneously, the server creates a pause-resume channel for each task, which maintains pending pause requests.

[0043] When the preset pause condition is triggered, the method further includes: generating a pause event containing a pause identifier and sending the pause event to the client.

[0044] Specifically, when the platform tool or intelligent agent detects login failure, browser not starting, need for QR code scanning, need to return to the search page, or need for manual confirmation, it invokes the pause tool. The pause tool generates a pause reason and a pause identifier, and sends a pause event containing the pause identifier and pause reason to the client via server push event, so that the client can display the corresponding prompt information according to the pause reason. The pause identifier is used to uniquely identify a pause request so that the server can accurately locate the corresponding pause position and execution context when resuming. The server pushes main intelligent agent messages, sub-intelligent agent events, tool call results, task schedules, and final states to the client in real time through the server push event mechanism.

[0045] S205. Receive the resume request containing the pause indicator returned by the client, and continue the execution of the target object search task from the breakpoint.

[0046] After the client displays the pause pop-up, the user completes the corresponding manual operation, such as scanning a QR code to log in, opening a browser, returning to the search page, or confirming the completion of the manual operation by clicking "confirm." The client then submits a resume request containing a pause indicator to the server. Upon receiving the resume request, the server locates the request to be resumed based on the pause indicator or a first-in-first-out (FIFO) order and returns the user's response to the original tool call. After receiving the confirmation result, the original tool call resumes execution from the paused position. The main agent and each platform sub-agent retain the original dialogue history, acquired target object information, and task status, resuming execution directly from the breakpoint. If the user cancels or actively stops the task, the server terminates subsequent tool calls from the main agent and each platform sub-agent.

[0047] Furthermore, in this embodiment, after each platform sub-agent obtains the target object information, it also includes steps such as normalizing the target object information, matching it with the enterprise's existing talent pool, and enriching contact information. Specifically, the normalization module reads the candidate fields returned by each platform and maps candidate records from different sources to a unified candidate record structure. Candidates from different platforms differ in field naming, data structure, and information completeness. For example, platform A uses "company" to represent the current company, while platform B uses "cur_company". The normalization module maps the fields from different platforms to the same field names. This unified candidate record includes at least the following fields: candidate number, source platform, name, current company, current position, company experience, school, skills or position tags, age, experience, education, salary, city, and resume summary. If a platform lacks some fields, empty values ​​or the original summary are retained.

[0048] Normalized candidate records are then entered into the company's existing talent pool for matching. The company's existing talent pool is an internally maintained database of existing talent assets used to store candidate information from past interactions, including basic candidate profiles, contact information, and communication records. The server calls the company's existing talent pool to match the normalized candidate records. Specifically, the server constructs several search strategies based on the normalized candidate records, such as combinations of name plus current company, name plus previous company, name plus position, and name plus school, and calls the search interface of the company's existing talent pool to obtain candidate matching results. The server merges and deduplicates the candidate matching results returned by different search strategies. It calculates a preliminary matching score based on the degree of overlap between the normalized candidate records and the deduplicated candidate matching results in fields such as name, company, position, school, major, age, and education. It obtains detailed information about candidate matching results whose preliminary matching scores meet preset conditions, recalculates the matching confidence score between the candidate matching result and the normalized candidate records based on this detailed information, and determines whether they are the same target object based on whether the matching confidence score reaches a preset threshold. If the match is found, the system returns the candidate's talent pool ID, contact information, talent pool link, and matching confidence score from the company's existing talent pool. If the match is not found, the system retains the candidate information on the platform and continues to save it or make a manual judgment.

[0049] Furthermore, in this embodiment, after the current round of activity of the main agent and each platform sub-agent ends, the server sends an idle event to the client. This idle event notifies the client that the current round of agent activity has ended, and the client can switch the input box to a follow-up question state. The server receives the additional constraints submitted by the client after the idle event, adds the additional constraints to the context of the target object search task, reuses the saved dialogue history and the acquired target object information, and continues to execute the target object search task. The additional constraints refer to the filtering conditions added by the user during task execution, such as "only those from Beijing" or "filter out those from 985 universities". After the server adds these additional conditions to the task context, the main agent or platform sub-agent continues to perform the search and filtering according to the updated constraints.

[0050] It should be noted that the data transmission, status updates, and event push involved in this application embodiment are all implemented through a server push event mechanism. The server pushes main agent messages, sub-agent events, tool call results, task plans, progress of sub-tasks on each platform, number of candidates, hit count, save results, and final status to the client in real time via server push events. The client receives and displays the above events through the server push event connection. This server push event mechanism enables users to view the system execution process and candidate save results in real time.

[0051] Figure 3 A flowchart illustrating another target object search method provided in this application embodiment, applied to a client. (Combined with...) Figure 3 As shown, the target object search method provided in this application embodiment may include: S301. Submit the natural language task description to the server.

[0052] In this embodiment, the client submits a natural language task description to the server, which triggers the server to execute the corresponding search processing flow. Specifically, the client receives the natural language task description entered by the user through the natural language task input box provided on the task homepage. After the user enters the recruitment requirement description in the input box and clicks the submit button, the client calls the server's task creation interface to submit the natural language task description to the server, records the task history, and redirects to the task details page.

[0053] S302, Receive the pause event sent by the server.

[0054] During the execution of the target object search subtask on the server side, when a preset pause condition is triggered, the server generates a pause event containing a pause identifier and sends it to the client, which receives the pause event. The pause event is a notification triggered by the server when manual user intervention is required. The pause identifier uniquely identifies a pause request, allowing the server to accurately locate the corresponding pause position and execution context upon subsequent resumption. The pause event also carries a pause reason, which may include at least one of the following: requiring QR code login, requiring the browser to be opened, requiring a return to the search page, or requiring manual confirmation.

[0055] S303, Display pause pop-up.

[0056] After receiving a pause event from the server, the client identifies the event and converts it to a pending pause state. The client then displays a pause pop-up on the interface. This pop-up cannot be closed directly; users cannot close it by clicking outside the pop-up or by clicking the close button. Confirmation or cancellation must be performed within the pop-up itself before it can be closed. The pause pop-up displays corresponding prompts based on the pause reason carried by the pause event: if the reason is requiring QR code login, the user is prompted to switch to the corresponding platform to scan the code; if it's requiring opening a browser, the user is prompted to launch the browser; if it's requiring returning to the search page, the user is prompted to manually return to the search page; if it's requiring manual confirmation, the user is prompted to confirm the relevant information. The pause pop-up also displays platform information and operation confirmation entry points to guide the user to complete the required manual operation.

[0057] S304. In response to the user's confirmation of the pause pop-up, submit a resumption request containing the pause identifier to the server.

[0058] After the user completes the manual operation prompted by the pause pop-up and clicks "Confirm," the client submits a resume request containing a pause indicator to the server. The resume request instructs the server to resume execution of the target object search subtask from the breakpoint, while retaining the dialogue history of each agent and the acquired target object information. If the user chooses to cancel, the client submits a cancellation request to the server, and the server suspends subsequent tool calls from the main agent and each platform sub-agent.

[0059] In this embodiment, the client also receives and displays the execution status pushed by the server in real time during task execution via server push event connection. Specifically, the client receives events such as the search plan generated by the main agent, the running progress of each platform sub-agent, tool call results, and candidate information via server push events, and displays the received events. On the task details page, the client displays the step plan generated by the main agent and the execution status of each step through the task plan panel, including pending, in progress, completed, and skipped. The platform panel area displays the independent running status of each platform sub-agent, recent search keywords, number of captured candidates, number of locked cards, and number of saved candidates side-by-side. The candidate panel displays saved candidate information, including the candidate's name, current company, current position, rating, score, contact information, talent pool profile link, and resume summary, as well as the candidate's matching result and matching tag in the company's existing talent pool. This server push event mechanism allows users to observe the system execution process and candidate saving results in real time on the task details page.

[0060] In this embodiment, the client also periodically checks the browser's running status and the login status of each recruitment platform. The client displays the browser's running status and the login status of each recruitment platform in the browser's status bar. If the browser is not running, an entry point to launch the browser is displayed for the user to click and start. If the user is not logged into any recruitment platform, a corresponding login prompt is displayed to guide the user to switch to the corresponding platform and complete the QR code login. The above checks are performed continuously before the user submits the natural language task description to ensure that the browser and platform login status are available during task execution. During task execution, if the server detects an invalid login status, it will notify the client via a pause event. The client will then prompt the user to re-scan the QR code to log in via a pause pop-up window.

[0061] In this embodiment, the client also supports a task follow-up function. After the current round of activity of the server's main agent and each platform sub-agent ends, the server sends an idle event to the client. Upon receiving this idle event, the client switches the input box to a follow-up state, allowing the user to add constraints, such as "only Beijing" or "screen out 985 universities." In response to the user's input in the follow-up state, the client submits the added constraints to the server. These constraints trigger the server to add them to the context of the target object search task, reusing the saved dialogue history and acquired target object information to continue executing the task. The client also supports a task stop function. The user can trigger a stop operation at any time via a stop button. The client submits a stop request to the server, which triggers a termination signal to stop subsequent tool calls by the main agent and each platform sub-agent.

[0062] Figure 4 A flowchart of another target object search method provided in the embodiments of this application is shown below. Figure 4 As shown, the target object search method provided in this application embodiment may include the following interactive steps: S401. The client submits a natural language task description to the server.

[0063] The client receives the natural language task description input by the user through the natural language task input box provided on the task homepage. After the user enters the recruitment requirement description and clicks the submit button, the client calls the task creation interface of the server and submits the natural language task description to the server.

[0064] S402. The server obtains the natural language task description submitted by the user, creates a target object search task based on the natural language task description, and starts the main intelligent agent.

[0065] After receiving the natural language task description submitted by the client, the server assigns a task identifier to the task, creates a target object search task, and starts the main intelligent agent. The main intelligent agent is an execution unit based on a large language model, possessing the capabilities of natural language understanding, task planning, tool invocation, and reasoning decision-making.

[0066] S403. The main intelligent agent generates a target object search subtask based on the natural language task description and dispatches the subtask to each platform sub-intelligent agent.

[0067] The main agent parses the natural language task description, identifying constraints such as target job title, city, experience, education, number of applicants, and platform scope. Based on the parsing results, it generates a cross-platform search plan and dispatches the target object search sub-tasks to the sub-agents corresponding to the interface-based recruitment platform and the page-based recruitment platform, respectively. Each platform's sub-agent maintains its own independent dialogue history and execution state, without interfering with each other. If a sub-agent is already running on the same platform, the server rejects a second sub-task on the same platform and returns a busy result.

[0068] S404. Each platform sub-agent calls platform tools within its corresponding toolset to obtain target object information.

[0069] The sub-agents of interface-based recruitment platforms invoke structured search tools, read the platform's login credentials and browser identity from the real browser using browser debugging protocols, and construct structured interface requests to retrieve candidate data. The sub-agents of page-based recruitment platforms invoke page state checking tools to confirm page readiness, then invoke search tools to perform searches and read candidate cards using candidate card reading tools, skipping locked cards. Each platform's sub-agents send the retrieved candidate information to the client in real-time via server push events for display.

[0070] S405, The server pushes the execution status to the client.

[0071] The server pushes messages from the main agent, events from sub-agents, tool call results, task plans, progress of sub-tasks on each platform, number of candidates, number of hits, saved results, and final status to the client in real time via a server-pushed event mechanism. The client receives and displays these events via a server-pushed event connection. On the task details page, the client displays the step plan generated by the main agent and the execution status of each step through the task plan panel, and displays the running status, search keywords, number of candidates, and number of hits of sub-agents on each platform side-by-side through the platform panel area.

[0072] S406. When the preset pause condition is triggered, the server pauses the execution of the target object search task, generates a pause event containing a pause identifier, and sends it to the client.

[0073] During the execution of a target object search task, when preset pause conditions are triggered, including invalid or non-existent login credentials, browser not running, user login requiring QR code scanning, page redirection to a non-search page, or manual confirmation required, the server pauses the execution of the current target object search task, retaining the dialogue history of the main agent and various platform sub-agents, as well as the acquired target object information. The pause tool generates a pause reason and a pause identifier, and sends a pause event containing the pause identifier and pause reason to the client via server push event. The pause identifier is used to uniquely identify a pause request so that the server can accurately locate the corresponding pause position and execution context when resuming.

[0074] S407, Client displays a pause pop-up.

[0075] After receiving a pause event from the server, the client identifies the event and converts it to a pending pause state, then displays a pause pop-up on the interface. This pause pop-up cannot be closed directly; users cannot close it by clicking outside the pop-up or by clicking the close button. The pause pop-up displays corresponding prompts based on the pause reason carried by the pause event: if it requires scanning a QR code to log in, it prompts the user to switch to the corresponding platform to scan the QR code; if it requires opening a browser, it prompts the user to launch the browser; if it requires returning to the search page, it prompts the user to manually return to the search page; if it requires manual confirmation, it prompts the user to confirm the relevant matters.

[0076] S408. After the user completes the manual operation and confirms, the client submits a resumption request containing a pause indicator to the server.

[0077] After the user completes the manual operation prompted by the pause pop-up and clicks "Confirm," the client submits a resumption request containing a pause indicator to the server. If the user chooses to cancel, the client submits a cancellation request to the server. Before submitting the resumption request, the client periodically checks the actual browser running status and the login status of each recruitment platform. If the browser is not running, an entry point to start the browser is displayed; if the user is not logged into a platform, a corresponding QR code login prompt is displayed to guide the user to complete the necessary manual operation.

[0078] S409. The server receives the recovery request and continues the target object search task from the breakpoint.

[0079] After receiving a resume request containing a pause flag from the client, the server locates the request to be resumed based on the pause flag and returns the user's response to the original tool call. Upon receiving confirmation, the original tool call resumes execution from the paused point. The main agent and each platform sub-agent retain the original dialogue history, acquired target object information, and task status, resuming execution directly from the breakpoint. If the user cancels, the server terminates subsequent tool calls from the main agent and each platform sub-agent.

[0080] After the current round of activities of S410, the main intelligent agent, and the sub-intelligent agents of each platform are completed, the server sends an idle event to the client.

[0081] After the current round of agent activity ends, the server sends an idle event to the client, notifying the client that the current round of agent activity has ended. Upon receiving the idle event, the client switches the input box to a follow-up-to-question state. The client displays the saved candidate information in the candidate panel, including the candidate's name, current company, current position, rating, score, contact information, talent pool profile link and resume summary, as well as the candidate's matching result and matching tag in the company's existing talent pool.

[0082] S411. The client submits additional constraints to the server when it is in a state where it can be further questioned.

[0083] Users can add constraints to input boxes in a follow-up-to-question state, such as "only those from Beijing" or "filter out those from 985 universities." The client responds to the user's input by submitting the added constraints to the server. Upon receiving the constraints, the server adds them to the context of the target object search task, reuses the saved dialogue history and acquired target object information, and continues executing the target object search task. The client also supports a task stop function; users can trigger a stop operation at any time using a stop button. The client submits a stop request to the server, which triggers a termination signal to stop subsequent tool calls by the main agent and various platform sub-agents.

[0084] The above are some specific implementations of a target object search method provided in the embodiments of this application. Based on this, this application also provides a corresponding apparatus. The apparatus provided in the embodiments of this application will be described below from the perspective of functional modularity.

[0085] Figure 5 This is a schematic diagram of a target object search device provided in an embodiment of this application, applied to a server. (Combined with...) Figure 5 As shown, the target object search device provided in this application embodiment includes: The task creation module 510 is used to obtain the natural language task description submitted by the user, create a target object search task based on the natural language task description, and start the main intelligent agent. The agent scheduling module 520 is used to enable the main agent to generate a target object search subtask based on the natural language task description, and to dispatch the target object search subtask to each platform sub-agent. The platform sub-agents include sub-agents corresponding to the interface-type recruitment platform and sub-agents corresponding to the page-type recruitment platform. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. The tool invocation module 530 is used to enable each platform sub-agent to invoke platform tools within its respective tool set to obtain target object information corresponding to the target object search sub-task. The pause / resume module 540 is used to pause the execution of the target object search task when a preset pause condition is triggered, retain the dialogue history of the main agent and the platform sub-agents as well as the acquired target object information; and receive a resume request containing a pause identifier returned by the client to continue the execution of the target object search task from the breakpoint.

[0086] In one implementation of this application, the first toolset configured for the sub-agent corresponding to the interface-type recruitment platform is different from the second toolset configured for the sub-agent corresponding to the page-type recruitment platform; each platform sub-agent maintains its own independent dialogue history and execution state.

[0087] In one implementation of this application, the apparatus further includes: The pause event sending module is used to generate a pause event containing the pause identifier when the preset pause condition is triggered, and send the pause event to the client.

[0088] In one implementation of this application, the apparatus further includes: The data matching module is used to normalize the target object information obtained by each platform sub-agent and generate a target object record in a unified format; call the enterprise's existing talent pool to query the matching results of the unified format target object record in the enterprise's existing talent pool; if a match is found, obtain the contact information of the target object in the enterprise's existing talent pool.

[0089] In one implementation of this application, when the data matching module queries the matching results of the target object records in the enterprise's existing talent pool in the unified format, it is specifically used for: Construct a multi-dimensional search strategy based on the target object records in the unified format; The search interface of the enterprise's existing talent pool is invoked, and candidate matching results are obtained based on the multi-dimensional search strategy; The candidate matching results are deduplicated to obtain the deduplicated candidate matching results; A preliminary matching score is calculated based on the degree of overlap between the target object record in the unified format and the deduplicated candidate matching results in at least one field. Obtain detailed information about candidate matching results whose preliminary matching scores meet preset conditions; Calculate the matching confidence score between the candidate matching result and the target object record in the unified format based on the detailed information; Based on whether the matching confidence score reaches a preset threshold, it is determined whether they are the same target object.

[0090] In one implementation of this application, the apparatus further includes: The task continuation module is used to send an idle event to the client after the current round of activities of the main agent and each platform sub-agent has ended; receive additional constraints submitted by the client after the idle event; add the additional constraints to the context of the target object search task; and reuse the saved dialogue history and the acquired target object information to continue executing the target object search task.

[0091] Figure 6 This is a schematic diagram of another target object search device provided in an embodiment of this application. (Combined with...) Figure 6 As shown in the embodiment of this application, another target object search device includes: The task submission module 610 is used to submit a natural language task description to the server; the natural language task description is used to trigger the server to create a target object search task and start the main intelligent agent, and the main intelligent agent generates a target object search sub-task according to the target object search task and dispatches it to each platform sub-intelligent agent for execution. The pause interaction module 620 is used to receive a pause event sent by the server. The pause event is generated by the server when a preset pause condition is triggered during the execution of the target object search subtask. The pause event includes a pause identifier and displays a pause pop-up window. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. The recovery request module 630 is used to submit a recovery request containing the pause identifier to the server in response to the user's confirmation operation of the pause pop-up. The recovery request is used to instruct the server to continue executing the target object search subtask from the breakpoint, and to retain the dialogue history of each agent and the target object information already obtained.

[0092] In one implementation of this application, the apparatus further includes: The front-end display module is used to display the task homepage, which includes a natural language task input box; a browser status bar, which displays the browser's running status and the login status of each recruitment platform; a task details page, which includes a task plan panel and a platform panel area; the task plan panel displays the search plan generated by the main agent and the execution status of each step; the platform panel area displays the running status, search keywords, number of candidates, and hit count of each platform sub-agent side-by-side; and a candidate panel displays saved candidate information.

[0093] In one implementation of this application, the pause interaction module, when displaying the pause pop-up, is specifically used for: Upon receiving the pause event sent by the server, a pause pop-up window that cannot be closed directly is displayed. The pause pop-up window displays the reason for the pause and an operation confirmation entry. The pause pop-up window displays corresponding prompt information based on the pause reason carried by the pause event. The pause reason includes at least one of the following: requiring QR code login, requiring the browser to be opened, requiring the user to return to the search page, or requiring manual confirmation.

[0094] In one implementation of this application, the candidate information displayed on the candidate panel includes: the candidate's match results and contact information in the company's existing talent pool.

[0095] In one implementation of this application, the apparatus further includes: The follow-up module is used to switch the input box to a follow-up state after receiving an idle event sent by the server; in response to the user's input operation after the input box is switched to the follow-up state, it submits additional constraints to the server. The additional constraints are used to trigger the server to add the additional constraints to the context of the target object search task, and reuse the saved dialogue history and the acquired target object information to continue to execute the target object search task.

[0096] This application also provides corresponding devices and computer storage media for implementing the solutions provided in this application.

[0097] The device includes a memory and a processor. The memory stores instructions or code, and the processor executes the instructions or code to cause the device to perform the method described in any embodiment of this application.

[0098] The computer storage medium stores code, and when the code is run, the device running the code implements the method described in any embodiment of this application.

[0099] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus a general-purpose hardware platform. Based on this understanding, the technical solution of this application can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as a read-only memory (ROM) / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a router) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0100] It is understood that in the specific embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved need to obtain user permission or consent when the above embodiments of this application are applied to specific products or technologies, and the collection, use and processing of related data need to comply with the relevant laws, regulations and standards of relevant countries and regions.

[0101] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the device and apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and apparatus embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components indicated 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 modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0102] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for searching a target object, characterized in that, Applied to the server side, the method includes: Obtain the natural language task description submitted by the user, create a target object search task based on the natural language task description, and start the main intelligent agent; The main intelligent agent generates a target object search subtask based on the natural language task description, and dispatches the target object search subtask to each platform sub-intelligent agent. Each platform sub-intelligent agent includes a sub-intelligent agent corresponding to an interface-type recruitment platform and a sub-intelligent agent corresponding to a page-type recruitment platform. The target object search subtask is a subtask generated by the main intelligent agent based on the target object search task and dispatched to each platform sub-intelligent agent for execution. Each platform sub-agent invokes platform tools within its respective toolset to obtain target object information corresponding to the target object search sub-task. When a preset pause condition is triggered, the execution of the target object search task is paused, and the dialogue history of the main agent and the platform sub-agents, as well as the acquired target object information, are retained. Receive a resume request containing a pause indicator from the client and continue executing the target object search task from the breakpoint.

2. The method according to claim 1, characterized in that, The first toolset configured for the sub-agents of the interface-based recruitment platform is different from the second toolset configured for the sub-agents of the page-based recruitment platform; each platform sub-agent maintains its own independent dialogue history and execution state.

3. The method according to claim 1, characterized in that, When the preset pause condition is triggered, the method further includes: Generate a pause event containing the pause identifier and send the pause event to the client.

4. The method according to claim 1, characterized in that, The method further includes: The target object information acquired by each platform sub-agent is normalized to generate a target object record in a unified format. Access the company's existing talent pool and query the matching results of the target object records in the company's existing talent pool in the unified format. If a match is found, the contact information of the target individual in the company's existing talent pool will be retrieved.

5. The method according to claim 4, characterized in that, The query results for the target object records in the unified format within the company's existing talent pool include: Construct a multi-dimensional search strategy based on the target object records in the unified format; The search interface of the enterprise's existing talent pool is invoked, and candidate matching results are obtained based on the multi-dimensional search strategy; The candidate matching results are deduplicated to obtain the deduplicated candidate matching results; A preliminary matching score is calculated based on the degree of overlap between the target object record in the unified format and the deduplicated candidate matching results in at least one field. Obtain detailed information about candidate matching results whose preliminary matching scores meet preset conditions; Calculate the matching confidence score between the candidate matching result and the target object record in the unified format based on the detailed information; Based on whether the matching confidence score reaches a preset threshold, it is determined whether they are the same target object.

6. The method according to claim 1, characterized in that, After the current round of activities of the main intelligent agent and each platform sub-intelligent agent has ended, the method further includes: Send an idle event to the client; Receive additional constraints submitted by the client after the idle event; The additional constraints are added to the context of the target object search task, and the saved dialogue history and the acquired target object information are reused to continue executing the target object search task.

7. A method for searching a target object, characterized in that, Applied to a client, the method includes: Submit a natural language task description to the server; the natural language task description is used to trigger the server to create a target object search task and start the main intelligent agent, the main intelligent agent generates a target object search sub-task according to the target object search task and dispatches it to each platform sub-intelligent agent for execution; The system receives a pause event sent by the server. The pause event is generated by the server when a preset pause condition is triggered during the execution of the target object search subtask. The pause event includes a pause identifier. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. Display pause pop-up; In response to the user's confirmation of the pause pop-up, a resumption request containing the pause identifier is submitted to the server. The resumption request is used to instruct the server to continue executing the target object search subtask from the breakpoint, and to retain the dialogue history of each agent and the target object information already acquired.

8. The method according to claim 7, characterized in that, The method further includes: Display the task homepage, which includes a natural language task input box; Display a browser status bar, which is used to display the browser's running status and the login status of each recruitment platform; Display the task details page, which includes a task planning panel and a platform panel area; The task planning panel is used to display the search plan generated by the main intelligent agent and the execution status of each step; The platform panel area is used to display the operating status, search keywords, number of candidates, and number of hits of each platform sub-agent side by side. Display a candidate panel, which is used to display saved candidate information; Upon receiving the pause event sent by the server, a pause pop-up window that cannot be closed directly is displayed, showing the reason for the pause and an operation confirmation entry. Upon receiving the idle event sent by the server, the input box is switched to a follow-up question state.

9. A target object search device, characterized in that, Applied to the server side, the device includes: The task creation module is used to obtain the natural language task description submitted by the user, create a target object search task based on the natural language task description, and start the main intelligent agent; The agent scheduling module is used to enable the main agent to generate a target object search subtask based on the natural language task description, and to dispatch the target object search subtask to each platform sub-agent. The platform sub-agents include sub-agents corresponding to the interface-type recruitment platform and sub-agents corresponding to the page-type recruitment platform. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. The tool invocation module is used to enable each platform sub-agent to invoke platform tools within its respective toolset to obtain target object information corresponding to the target object search sub-task. The pause and resume module is used to pause the execution of the target object search task when a preset pause condition is triggered, retain the dialogue history of the main agent and the platform sub-agents as well as the acquired target object information; and receive a resume request containing a pause identifier returned by the client to continue the execution of the target object search task from the breakpoint.

10. A target object search device, characterized in that, Applied to a client, the device includes: The task submission module is used to submit a natural language task description to the server. The natural language task description is used to trigger the server to create a target object search task and start the main intelligent agent. The main intelligent agent generates a target object search sub-task according to the target object search task and dispatches it to each platform sub-intelligent agent for execution. The pause interaction module is used to receive a pause event sent by the server. The pause event is generated by the server when a preset pause condition is triggered during the execution of the target object search subtask. The pause event includes a pause identifier and displays a pause pop-up window. The target object search subtask is a subtask generated by the main agent based on the target object search task and dispatched to each platform sub-agent for execution. The recovery request module is used to respond to the user's confirmation operation of the pause pop-up window by submitting a recovery request containing the pause identifier to the server. The recovery request is used to instruct the server to continue executing the target object search subtask from the breakpoint and retain the dialogue history of each agent and the target object information already obtained.

Images