AI recruitment closed-loop automatic control method based on multi-agent cooperation

Abstract

The application discloses an AI recruitment closed loop automation control method based on multi-agent cooperation, relies on multi-agent cooperation, splits recruitment execution links to form independent agent nodes, combines data marking, time sequence timeline, causal modeling and time sequence checking technology, carries out standardized reservation, causal correlation and compliance checking on the whole process data of school recruitment, constructs an interpretable and traceable AI recruitment closed loop automation process, effectively solves the defects of traditional AI recruitment decision black box, data fragmentation and logic untraceable, and improves the execution standardization, transparency and automation control capability of campus recruitment.

Description

Technical Field

[0001] This invention relates to the field of intelligent recruitment technology, specifically to an AI recruitment closed-loop automated control method based on multi-agent collaboration. Background Technology

[0002] With the rapid development of artificial intelligence technology, AI is increasingly being applied in the field of human resources, especially in large-scale campus recruitment scenarios. AI systems can automatically complete processes such as resume screening, interview scheduling, candidate evaluation, and hiring decisions, significantly improving recruitment efficiency. Multi-Agent Systems (MAS), as an important branch of distributed artificial intelligence, can effectively handle complex tasks through the collaboration and interaction of multiple autonomous intelligent agents.

[0003] Current traditional AI-driven recruitment automation methods mostly employ a pipeline model driven by a single model or rule engine. For example, in resume screening, the system extracts candidate information using natural language processing and provides screening results using a pre-trained scoring model; the interview scheduling stage relies on calendar integration and simple time-matching algorithms; and the evaluation and hiring processes often make decisions based on manually set thresholds or a single comprehensive score. Furthermore, some systems introduce workflow engines, defining the recruitment process as a fixed sequence of steps, with information exchanged between steps via data interfaces. However, these traditional methods share the following common characteristics: 1. All aspects are controlled by a single central module, lacking task division and collaboration, resulting in poor system scalability and difficulty in coping with complex scenarios.

[0004] 2. Decision-making rules and weights are usually set manually in advance, and once deployed, they are difficult to adjust dynamically and cannot adapt to changes in recruitment needs.

[0005] 3. Data is stored in a scattered manner at each stage, lacking a unified labeling and association mechanism, making it difficult to trace back the decision-making process. For example, it is impossible to clearly show why a candidate was eliminated or hired.

[0006] Traditional methods have significant shortcomings in data management and decision-making rules. Data from each stage of the recruitment process is usually stored in different systems or modules without a unified labeling and association mechanism. For example, scoring data from the resume screening stage, time records of interview arrangements, and evaluation feedback often exist in independent formats without a traceable index between them. This makes the decision-making process like an "information island," making it impossible to effectively link the final hiring result of a candidate with the decision-making basis of intermediate steps such as resume screening and interview evaluation. This makes it difficult to reconstruct the complete decision-making chain during subsequent audits or analyses, and even more difficult to verify the rationality and fairness of the decision through data backtracking. Summary of the Invention

[0007] The main objective of this invention is to provide an AI recruitment closed-loop automated control method based on multi-agent collaboration, in order to address the shortcomings of related technologies.

[0008] To achieve the above objectives, according to a first aspect of the present invention, an AI recruitment closed-loop automated control method based on multi-agent collaboration is provided, comprising preloading each agent node, wherein the agent nodes are set based on the recruitment process; each agent node is defined with output content, the decision basis corresponding to the output content, the weight of each factor in the decision basis, the data source corresponding to the node, and the corresponding triggering conditions; after obtaining recruitment needs, the first agent node is triggered, and each agent node starts and executes the recruitment process in sequence according to the triggering conditions, during which the recruitment results determined by each agent node are obtained; and the structured data generated by all agent nodes with timestamps, inputs, processing logic, and output fields are concatenated in chronological order. The recruitment decision timeline involves real-time collection of recruitment process data during the operation of each agent node. This data is categorized and labeled according to preset rules, and a unique timestamp is added to each data point to ensure it can be located to the corresponding agent node and its generation time. A decision timeline function processes the timeline and recruitment results to match each recruitment result with a corresponding decision timeline and assign a unique timeline identifier. A causal relationship function describes the causal relationships between different agent nodes within a single decision timeline. A causal inference model determines the causal chains between agent nodes, indicating how the input of a preceding node affects the output of the current node and how preceding decisions influence subsequent decisions.

[0009] Optionally, the method further includes optimizing the decision timeline based on historical data using a dynamic optimization algorithm, wherein at each decision node, the effectiveness of each decision is measured by a reward function.

[0010] Optionally, the method further includes: determining whether the execution order of the agent nodes is compliant, wherein the time interval of each node is calculated using a preset time consistency measurement rule; determining whether there is an anomaly based on the time interval; verifying the logical correlation between the preceding input and the current output; and updating the decision chain by eliminating non-compliant data or data with abnormal logical correlation.

[0011] Optionally, the method may further include quantifying the strength of the causal relationship.

[0012] Optionally, processing the decision timeline and recruitment results using a decision timeline function includes: using... Indicates the first The recruitment results correspond to the decision-making timeline as follows: Its function expression is as follows: ,in, Indicates the first Timeline of decision-making for each recruitment outcome This represents decision data related to the outcome. It is the unique identifier for that timeline. This indicates the processing logic and process information for this timeline.

[0013] Optionally, describing the causal relationships between different agent nodes within a single decision timeline based on a causal association function includes: ,in, Indicates the first The causal relationships within the decision-making timeline constitute the overall causal relationships of a recruitment process. and These represent two decision points in the timeline; causal relationship. This indicates the filtering results. How to influence interview scheduling The decision.

[0014] Optionally, determining the causal chain between agent nodes based on the decision causal inference model includes: assuming Indicates the first The input of each node, Indicates its output, Representing the decision-making process, the causal inference function is: ,in, Indicates the first The set of parent nodes of each node. Given a parent node hour, right The impact, Parent node The probability distribution.

[0015] Optionally, optimizing the decision timeline based on historical data using a dynamic optimization algorithm includes: assuming that in the first... There are nodes, and their states are... Action for reward function for ,in, , , The weighting coefficient represents a comprehensive consideration of accuracy, time efficiency, and cost-effectiveness.

[0016] Optionally, quantifying the strength of the causal relationship includes: ,in, Represents a node With nodes The strength score of the causal relationship between them For the first The decision samples from 1 to n are summed cumulatively, where n is the total number of samples. It is in the In a sample, the node The output result; It is in the In a sample, the node Its own input data; It is in the In each sample, the preceding parent node The input data, i.e., the data for the nodes Pre-existing data that leads to causal effects; For conditional probability, given a node enter With parent node enter At that time, node Output The probability reflects right The actual degree of impact, The prior probability, i.e., the previous node Input The probability of occurrence in all samples is used to weight and normalize the causal effects.

[0017] According to a second aspect of the present invention, an AI recruitment closed-loop automated control system based on multi-agent collaboration is provided, comprising: a loading unit for preloading each agent node, wherein the agent nodes are set based on the recruitment process; each agent node is defined with output content, decision basis corresponding to the output content, weight of each factor in the decision basis, data source corresponding to the node, and corresponding triggering conditions; an execution unit for triggering the first agent node after obtaining recruitment requirements, and each agent node starts and executes the recruitment process in sequence according to the triggering conditions, during which the recruitment results determined by each agent node are obtained; and a structured recruitment document with timestamps, inputs, processing logic, and output fields generated by all agent nodes are sequentially linked in time. The decision timeline involves real-time collection of recruitment process data during the operation of each agent node. This data is then categorized and labeled according to preset rules, and a unique timestamp is added to each piece of data to ensure that each piece of data can be located to the corresponding agent node and its generation time. The inference unit processes the decision timeline and recruitment results using a decision timeline function to match each recruitment result with a corresponding decision timeline and assign a unique timeline identifier. A causal relationship function describes the causal relationships between different agent nodes within a single decision timeline. A causal inference model determines the causal chains between agent nodes, where the causal chains indicate how the input of a preceding node affects the output of the current node, and how preceding decisions affect subsequent decisions.

[0018] This embodiment is based on a multi-agent collaborative AI recruitment closed-loop automated control method. Relying on multi-agent collaboration, the recruitment execution process is broken down into independent intelligent agent nodes. By combining data labeling, time-series timelines, causal modeling, and time-series verification technologies, the data of the entire campus recruitment process is standardized, retained, and verified for causal relationships and compliance. This constructs an explainable and traceable AI recruitment closed-loop automated process, effectively solving the defects of traditional AI recruitment decision-making black boxes, data fragmentation, and untraceable logic, and improving the standardization, transparency, and automated control capabilities of campus recruitment execution. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the AI ​​recruitment closed-loop automated control method based on multi-agent collaboration according to an embodiment of the present invention; Figures 2-5This is a schematic diagram illustrating the application of the AI ​​recruitment closed-loop automated control method based on multi-agent collaboration in an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0023] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0024] According to embodiments of the present invention, an AI recruitment closed-loop automated control method based on multi-agent collaboration is provided, such as... Figure 1 As shown, steps 101 to 103 are included below: Step 101: Preload each agent node, where the agent nodes are set based on the recruitment process; each agent node is defined with output content, the decision basis corresponding to the output content, the weight of each factor in the decision basis, the data source corresponding to the node, and the corresponding triggering conditions.

[0025] In this step, taking the AI ​​recruitment scenario as an example, we clarify the output content of each intelligent agent's working node, and predefine the decision basis, parameter weights, data sources and triggering conditions to ensure that the background of the decision can be traced later.

[0026] refer to Figure 2The example illustrates how, based on different stages of the recruitment process, such as resume screening, interview scheduling, evaluation, and hiring, multiple agent work nodes are defined, with specific tasks and outputs for each node. For instance, in the resume screening stage, the agent might output a decision on "whether the candidate meets the job requirements"; in the interview scheduling stage, the agent might output decisions on "whether to schedule an interview" and "interview time arrangement." For the output of each node, the basis for the decision must be clearly defined. For example, in the resume screening stage, the decision basis might be the candidate's education, experience, and skill matching; in the interview scheduling stage, the basis might be the interviewer's availability and the candidate's available time.

[0027] Each decision-making criterion is quantified, and the weight of different factors is clearly defined. In this example, during resume screening, education accounts for 40%, experience for 30%, and skill matching for 30%. This ensures a clear and reasonable decision-making process.

[0028] Clearly define the data sources required for each decision-making step. Data can come from various channels such as recruitment platforms, HR management systems, and interviewer feedback. Determining the data sources helps ensure the reliability and verifiability of the data.

[0029] For each decision node, its triggering conditions also need to be defined, specifying under what circumstances the decision-making process for that node will be initiated. For example, resume screening will only be triggered after candidate information is entered; interview scheduling will be triggered after the candidate passes the screening and when the interviewer is available.

[0030] By completing the above process, a detailed description document for each intelligent agent node can be generated, listing the output content, decision basis, parameter weights, data sources, and triggering conditions of each node, ensuring that the subsequent decision-making process is traceable and auditable. The quantitative description includes specific values ​​and standards for the parameter weights and triggering conditions of each node's decision basis, making the entire recruitment process logically clear, verifiable, and traceable during execution and review.

[0031] After the above configuration, the node can be loaded and used during the usage phase.

[0032] Step 102: After obtaining the recruitment requirements, the first intelligent agent node is triggered. Each intelligent agent node starts and executes the recruitment process in sequence according to the trigger conditions. During this process, the recruitment results determined by each intelligent agent node are obtained. The structured recruitment decision timeline with timestamps, input, processing logic, and output fields generated by all intelligent agent nodes is connected in chronological order. During the operation of each intelligent agent node, the recruitment process generation data is collected in real time, and classified and labeled according to preset rules. A unique timestamp is added to each piece of collected data to ensure that each piece of data can be located to the corresponding intelligent agent node and the time of generation.

[0033] In this step, each agent node is pre-equipped with a data collection tagging function. The collected information is categorized and tagged according to different recruitment stages and data types, and then a unique timestamp is added to ensure that each piece of data accurately corresponds to a specific node and operation time. By equipping each agent node with a data collection tagging function, it is ensured that each piece of data accurately corresponds to a specific node and operation time. (Reference) Figure 3 The data collection and marking function diagram shows the stage type corresponding to the marked node, and a unique identifier is generated after adding a timestamp.

[0034] The core of this process lies in data identification and classification, ensuring that the data generated at each node during execution can be accurately traced, preventing information loss or confusion during subsequent decision-making checks and audits. To achieve this goal, the following aspects need to be considered: 1) For each stage of the recruitment process, such as resume screening, interview scheduling, evaluation, and hiring, the data needs to be categorized and labeled according to the tasks at each stage. For example, in the resume screening stage, the data collected may include "candidate basic information" and "resume evaluation results"; in the interview scheduling stage, the data collected may include "interview schedule" and "interviewer evaluation data". Since the tasks at each stage are different, the types and content of the data collected also differ; therefore, different labeling and classification rules should be set for different stages. 2) To ensure accurate data traceability, data collection requires not only classification but also the addition of a unique timestamp. The purpose of the timestamp is to ensure that each piece of data is accurately linked to a decision-making action at a specific point in time, avoiding data corruption caused by overlapping times or multi-threaded operations. This necessitates generating a unique timestamp during data collection to ensure that each piece of data is associated with the time of its occurrence. This goal is achieved through a data tagging function. ,in, Indicates the first The node, the first Data collected from each operation Indicates the timestamp of the operation. Indicates data type category labels, The data tagging function described above represents a unique identifier for the data, and it describes the process of generating data acquisition tags, including the timestamp. Automatically generated by the system, ensuring that each piece of data strictly corresponds to the operation time; The category labels are assigned based on node type and stage to ensure that the data can be classified according to different stages of recruitment. It uses a unique identifier generated by the system to distinguish different data under the same node. Through precise timestamps and tag classification, it ensures that each piece of data can clearly identify its source node, time point and data type, so that the data is not only traceable, but can also be accurately traced to the specific operation history of a certain node in the entire recruitment process.

[0035] 3) Using time-series verification to validate the rationality of decisions ensures the consistency and accuracy of data collection and operation times throughout the process. Time-series verification ensures that the execution order of each node in the decision-making process conforms to predetermined logic, preventing data timing errors. Based on this, it allows for further checks to ensure that the decisions at each node comply with set business rules and parameter constraints, guaranteeing that the execution of each decision and data collection meets rationality and accuracy requirements.

[0036] For example, starting with inputting recruitment needs for AI graduates, the required input data for each stage is clearly defined. For instance, the input data for recruitment needs might include job information, number of openings, recruitment timeline, and required skills; in the resume screening stage, the input data might include basic candidate information and resume content. The input data at each stage forms the basis for subsequent decisions and must be accurate. Based on the defined decision criteria, parameter weights, and data sources, the processing logic of each node needs to be broken down in detail. For example, in the resume screening stage, the processing logic might use a set algorithm model to calculate a comprehensive score based on the candidate's education, work experience, and other information, and then decide whether to proceed to the next stage based on the score. In the interview scheduling stage, the processing logic might automatically match the interviewer's availability, the candidate's availability, and the interview type to schedule the interview.

[0037] The output of each node is a direct reflection of the decision. For example, the output of the resume screening stage is "whether to invite an interview"; the output of the interview scheduling stage might be "interview time arrangement" and "whether the interview was passed". These outputs will directly affect the execution of subsequent nodes, so it is essential to ensure that the output results are clear and traceable. The final recruitment decision timeline includes the inputs, processing, and outputs of each node in the entire recruitment process, arranged in chronological order.

[0038] Imagine an AI-driven campus recruitment process, comprising three stages: resume screening, interview scheduling, and hiring. Starting with the input of recruitment requirements, break down the decision-making logic of each stage and create a decision timeline. The input data for entering recruitment requirements is as follows: Job Information: Development Engineer Number of openings: 5 Required skills: Java, Python, Database Recruitment period: May 1, 2026 After data entry, recruitment requirements are generated, triggering a resume screening node. At this point, the decision logic is to "select qualified candidates from all received resumes." During the resume screening stage, input data includes candidate resume information such as education, internship experience, and skill matching. The scoring result is a comprehensive score calculated using the following scoring function: ,in, This indicates the candidate's overall score. Indicates the level of education (e.g., doctoral, master's, bachelor's). Indicates internship experience, Indicates skill matching degree. , , These are weighting coefficients. and satisfy At this point, the system will filter out candidates who meet the preset thresholds. If the candidate's score... Exceeding a certain threshold The candidate will proceed to the interview stage; Output: List of selected candidates, scores for each candidate. At the interview scheduling node, the input data mainly includes the candidate's basic information, such as name, contact information, and preferred interview time. Based on this information, the system will automatically schedule interview times according to the candidates' time preferences. Each interview typically lasts one hour, and the output is the interview schedule for each candidate. The task of this node is to schedule interviews according to the candidates' preset time preferences, ensuring that each candidate can be interviewed at a suitable time. At the hiring milestone, after the interviews are completed, the input data includes: the interviewer's evaluation, such as scores and recommendations, and the overall score, which includes the combined score of the interview and resume screening. The output is: the final list of hired candidates and the hiring status of each candidate. Following the steps outlined above, the final recruitment decision timeline is arranged chronologically as follows: In this timeline, the inputs, processing logic, and outputs of each node are clearly marked, and timestamps accurately record the chronological order of each operation, ensuring that the recruitment decision-making process is fully traceable. The decision-making basis and processing steps of each node conform to the pre-defined logic, avoiding data gaps and temporal disorder.

[0039] Step 103: Process the decision timeline and recruitment results using the decision timeline function to match each recruitment result with a corresponding decision timeline and assign a unique timeline identifier; describe the causal relationships between different agent nodes within a single decision timeline based on the causal association function; determine the causal chains between agent nodes based on the decision causal inference model, where the causal chains indicate how the input of the preceding node affects the output of the current node and how the preceding decision affects the subsequent decision. In this step, the broken decision timeline is bound to each recruitment result, and each timeline is assigned a unique identifier. This ensures that the complete decision-making process for each recruitment result can be traced, clarifying the causal relationships. Each decision node, each recruitment result, and the causal relationships between them are precisely linked, ensuring that each decision process can be fully traced back and clearly understood. To achieve this, each node and its corresponding decision timeline must be uniquely identified. A decision timeline function is defined to ensure that each recruitment result is bound to its corresponding timeline information. Indicates the first The recruitment results correspond to the decision-making timeline as follows: Its function expression is as follows: ,in, Indicates the first Timeline of decision-making for each recruitment outcome This represents decision-making data related to the outcome, such as candidate information, scores, and interview arrangements. It is the unique identifier for that timeline. This indicates the processing logic and process of the timeline, ensuring that every recruitment decision throughout the entire recruitment process is clearly traceable; A unique identifier for each timeline Generated by the system, this unique identifier ensures that decisions made across multiple nodes can be distinguished. This identifier is crucial for subsequent queries, auditing, and decision analysis.

[0040] As an optional implementation of this embodiment, processing the decision timeline and recruitment results through a decision timeline function includes: using... Indicates the first The recruitment results correspond to the decision-making timeline as follows: Its function expression is as follows: ,in, Indicates the first Timeline of decision-making for each recruitment outcome This represents decision data related to the outcome. It is the unique identifier for that timeline. This indicates the processing logic and process information for this timeline.

[0041] It is necessary to ensure that each node and its corresponding decision timeline can be uniquely identified. This is achieved by defining a decision timeline function to ensure that each recruitment result is bound to the corresponding timeline information. Indicates the first The recruitment results correspond to the decision-making timeline as follows: Its function expression is as follows: ,in, Indicates the first Timeline of decision-making for each recruitment outcome This represents decision-making data related to the outcome, such as candidate information, scores, and interview arrangements. It is the unique identifier for that timeline. This indicates the processing logic and procedures of the timeline, ensuring that every recruitment decision throughout the entire recruitment process is clearly traceable; each timeline is represented by a unique identifier. Generated by the system, this unique identifier ensures that decisions made across multiple nodes can be distinguished. This identifier is crucial for subsequent queries, auditing, and decision analysis.

[0042] Furthermore, in order to trace the causal relationships in the decision-making process, it is also necessary to define a causal association function. Used to describe each decision timeline The causal relationships between the nodes in the function. This function is expressed by the following formula: ,in, Indicates the first The causal relationships within the decision-making timeline constitute the overall causal relationships of a recruitment process. and These represent two decision-making nodes in the timeline. For example, between the resume screening node and the interview scheduling node, The selection results represent the candidates. The decision to arrange interviews. Cause and effect relationship. This indicates the filtering results. How to influence interview scheduling The decision.

[0043] As an optional implementation of this embodiment, determining the causal chain between agent nodes based on the decision causal inference model includes: setting Indicates the first The input of each node, Indicates its output, Representing the decision-making process, the causal inference function is: ,in, Indicates the first The set of parent nodes of each node. Given a parent node hour, right The impact, Parent node The probability distribution.

[0044] To better capture the causal relationships between recruitment decision nodes, a causal inference model is introduced. By observing the input-output relationship, the causal chain between each decision node is automatically derived. At the implementation level, a structured causal model (SCM) is used to define the causal relationships between nodes. Furthermore, a Bayesian network or causal graph model is combined to accurately model "how the input of the preceding node affects the output of the current node, and how the preceding decision affects the subsequent decision". set up Indicates the first The input of each node, Indicates its output, To represent the decision-making process, define a causal inference function: ,in, Indicates the first The set of parent nodes of each node. Given a parent node hour, right The impact, Parent node The probability distribution is used to infer the contribution of each node to the final recruitment result based on information from the previous node. This causal link allows for precise tracing of the decision-making process and logic behind each recruitment outcome, clearly identifying the inputs and outputs of each decision, and clarifying the relationships between nodes, thereby enhancing the transparency and credibility of the entire recruitment process.

[0045] The above content is for reference only. Figure 4 The diagram illustrating causal relationship binding and tracing demonstrates the logic of the above scheme.

[0046] As an optional implementation of this embodiment, the decision timeline is optimized based on historical data using a dynamic optimization algorithm. In each decision node, the effectiveness of each decision is measured by a reward function to dynamically optimize the node weights and decision paths.

[0047] In this optional implementation, the broken decision timeline is bound to each recruitment result, and each timeline is assigned a unique identifier so that the complete decision-making process can be found for each recruitment result, clarifying the causal relationship. This accurately links each decision node, each recruitment result, and the causal relationship between them in the recruitment process, ensuring that each decision-making process can be fully traced back and clearly understood.

[0048] As an optional implementation of this embodiment, the decision timeline is optimized based on historical data using a dynamic optimization algorithm. In each decision node, the effectiveness of each decision is measured by a reward function to dynamically optimize the node weights and decision paths.

[0049] In this optional implementation, as the recruitment process progresses, new data and decision-making information may be continuously added. Therefore, the recruitment decision-making process needs to be adaptive, capable of automatically adjusting the decision path based on real-time data. Through dynamic optimization algorithms (such as reinforcement learning), the decision timeline can be continuously optimized based on historical data. Define a reinforcement learning model at each decision node. In the middle, through the reward function This is used to measure the effectiveness of each decision. The model continuously adjusts the decision paths and weights to maximize the final recruitment result. Assume that in the... There are nodes, and their states are... Action for Then the reward function It can be defined as: ,in, , , The weighting coefficients represent a comprehensive consideration of accuracy, time efficiency, and cost-effectiveness. By weighting these factors, the reinforcement learning algorithm dynamically adjusts and optimizes the recruitment decision path to obtain the optimal recruitment decision.

[0050] As an optional implementation of this embodiment, the method further includes: determining whether the execution order of the agent nodes is compliant, wherein the time interval of each node is calculated using a preset time consistency measurement rule; determining whether there is an anomaly based on the time interval; verifying the logical correlation between the preceding input and the current output; and updating the decision chain by eliminating non-compliant data and data with abnormal logical correlation.

[0051] In this optional implementation, time-series verification is used to validate the rationality of decisions, check the decision timeline and the actual generation order, examine logical connections, and eliminate logical gaps and time-series inconsistencies to ensure the accuracy of traceability results. The first four steps sequentially complete the definition of intelligent agent decision node rules, data collection and marking with unique timestamps, construction of the entire recruitment process decision timeline, and unique binding and association of recruitment results with the decision timeline and node causal relationships. This outputs complete, traceable, uniquely identified, and causal logic-containing full-link data and timeline of recruitment decisions. Based on this output, time-series verification and decision rationality checks are conducted. By verifying the decision timeline, data marking, node execution order, and causal relationships formed in the aforementioned steps, and eliminating abnormal content such as logical gaps and time-series inconsistencies, this verifies the accuracy and ensures compliance of the output results of the first four steps, providing a reliable data foundation for the final generation of a traceable report. For example, based on the decision timeline and data markers, the system can ensure that the timestamps of each node conform to the predetermined execution order, avoiding any disruption in the node execution order. The core here is verifying whether the timestamps and the node execution order match. It also verifies whether the output of each node conforms to the expected input-output relationship. For instance, the execution of interview arrangements should be based on the results of resume screening, and hiring decisions should rely on interview feedback. For nodes with disordered timelines or illogical logic, the system should be able to mark and remove them, ensuring that the final recruitment decision chain is reasonable and smooth.

[0052] To achieve automatic marking and removal of nodes with timing errors and logical gaps, a timing consistency metric is introduced to quantify the execution order of nodes. This is done by calculating the time interval between each node to quantify the sequential relationship of node execution. For example, suppose node... and nodes The timestamps are respectively and Then the time consistency measure Represented as: ,in, hour, Represents a node Relative to node The time interval. If If the value is less than the predetermined tolerance value, it indicates that the timing is correct; otherwise, it indicates that there is a timing error or anomaly.

[0053] As an optional implementation of this embodiment, the method further includes quantifying the strength of the causal relationship.

[0054] In this optional implementation, to quantify the strength of causality, it is assumed that the nodes... and nodes The causal relationship between them is , is represented as: , Represents a node With nodes The causal relationship strength / causal consistency score between two nodes represents the point-to-point causal strength and is the final quantified result. For the first Candidate 1 / 1 The decision samples from 1 to n are summed cumulatively, where n is the total number of samples. It is in the In a sample, the node The output results, such as resume screening approval, successful interview scheduling, and hiring decision, etc. It is in the In a sample, the node Its own input data, such as nodes The decision-making basis, scoring, parameters, etc. It is in the In each sample, the preceding parent node The input data, i.e., the data for the nodes Pre-existing data that leads to causal effects For conditional probability, given a node enter With parent node enter At that time, node Output The probability reflects right The actual degree of impact, Prior probability: Preorder node Input The probability of occurrence in all samples is used to weighted normalize the causal effect. This is achieved by quantifying the nodes. Output Is it affected by the node? enter The impact is calculated and weighted according to the causal relationship between each node.

[0055] Furthermore, after the aforementioned processing, a traceable report sorted by time can be generated, clearly displaying information throughout the entire recruitment decision-making process and meeting the various query needs of schools, students, and recruiters.

[0056] In this embodiment, in order to classify and label data according to recruitment stage and data type, and to assign a unique identifier to each piece of data, exclusive classification labels are set for different campus recruitment stages such as resume screening, interview arrangement, evaluation, and hiring, for information such as candidate basic information, resume evaluation results, interview time arrangement, and interviewer evaluation data. At the same time, a system-unique identifier is generated for each piece of data to distinguish different data records under the same node.

[0057] All collected data is appended with a precise and unique timestamp. The timestamp is then bound to category labels and unique identifiers through a data tagging function to generate standardized collected data. This ensures that each piece of recruitment-related data strictly corresponds to the specific operation time, effectively avoiding data time sequence disorder in scenarios such as multi-threaded parallel operations and time overlap. At the same time, it provides a precise time dimension basis for subsequent time sequence verification and decision backtracking.

[0058] Starting with the input of recruitment needs, the entire chain of input-processing-output for each decision node is structurally broken down, forming a complete timeline of decision-making. Based on the defined node rules, the input data, quantitative processing logic, and output results of each node are clarified. A continuous and uninterrupted recruitment decision-making timeline is formed according to the actual execution order, transforming the abstract recruitment process into an executable, verifiable, and traceable structured process, providing a standardized process framework for causal binding and timeline verification.

[0059] By binding recruitment results one-to-one with the decision timeline and assigning a unique identifier IDtime to each timeline, the k-th recruitment result, associated decision data, unique identifier, and processing logic are precisely linked through the decision timeline function. This ensures that every hiring, shortlisting, or elimination conclusion corresponds to a complete decision chain, giving recruitment results the technical characteristics of being traceable, verifiable, and locatable, fundamentally solving the problem of AI recruitment results being unsubstantiated and unverifiable.

[0060] A decision causal inference model is introduced and a structured causal model (SCM) is used to define the causal relationships between nodes. By combining Bayesian networks and causal graphs, the input of the preceding node, the output of the current node, and the influence of the parent node are quantitatively modeled. The transmission logic and influence strength between nodes are clarified through causal association functions, so that the recruitment decision is transformed from a "black box output" into a transparent logic that is "explainable and deducible".

[0061] Based on the decision timeline, we conduct time sequence verification, logical correlation checks, and anomaly removal. We verify the order of node timestamps using a time sequence consistency measurement formula, verify the dependency relationship between preceding and subsequent nodes, and automatically identify and remove abnormal records with time sequence errors, logical gaps, and causal gaps. While ensuring compliance with the process execution order, we ensure that the decision logic of the entire chain is self-consistent and the data is accurate, providing a high-quality and highly reliable data foundation for the final generation of traceable reports.

[0062] Architecture reference applicable to the above methods Figure 5 The framework diagram of the AI ​​recruitment closed-loop automated control system based on the multi-agent collaborative architecture is shown. The core of the method in this embodiment can be implemented in the multi-agent collaborative layer, the decision engine layer, and the timing verification layer.

[0063] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0064] According to an embodiment of the present invention, an AI recruitment closed-loop automated control system based on a multi-agent collaborative architecture is also provided, comprising: a loading unit for preloading each agent node, wherein the agent nodes are set based on the recruitment process; each agent node is defined with output content, decision basis corresponding to the output content, weight of each factor in the decision basis, data source corresponding to the node, and corresponding triggering conditions; an execution unit for triggering the first agent node after obtaining recruitment requirements, and each agent node starts and executes the recruitment process in sequence according to the triggering conditions, during which the recruitment results determined by each agent node are obtained; and a structured recruitment document with timestamps, inputs, processing logic, and output fields generated by all agent nodes are concatenated in chronological order. The recruitment decision timeline involves real-time collection of recruitment process data during the operation of each agent node. This data is then categorized and labeled according to preset rules, and a unique timestamp is added to each piece of data to ensure that each piece of data can be located to the corresponding agent node and its generation time. The inference unit processes the decision timeline and recruitment results using a decision timeline function to match each recruitment result with a corresponding decision timeline and assign a unique timeline identifier. A causal relationship function describes the causal relationships between different agent nodes within a single decision timeline. A causal inference model determines the causal chains between agent nodes, where the causal chains indicate how the input of a preceding node affects the output of the current node and how preceding decisions affect subsequent decisions.

[0065] According to embodiments of the present invention, the present invention also provides a readable storage medium storing computer instructions that enable a computer to perform the methods described in any of the above embodiments when executed.

[0066] According to embodiments of the present invention, the present invention also provides a computer program product that, when executed by a processor, can implement the methods described in any of the above embodiments.

[0067] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0068] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

Claims

1. A closed-loop automated control method for AI recruitment based on multi-agent collaboration, characterized in that, include: Load each agent node, which is set based on the recruitment process; each agent node is defined with output content, the decision basis corresponding to the output content, the weight of each factor in the decision basis, the data source corresponding to the node, and the corresponding triggering conditions. After obtaining recruitment needs, the first agent node is triggered. Each agent node starts and executes the recruitment process in sequence according to the trigger conditions. During this process, the recruitment results determined by each agent node are output to the outside world. The structured recruitment decision timeline generated by all agent nodes is connected in chronological order, with timestamps, input, processing logic, and output fields. In particular, during the operation of each agent node, the recruitment process data is collected in real time, and classified and labeled according to preset rules. A unique timestamp is added to each piece of collected data to ensure that each piece of data can be located to the corresponding agent node and the time of generation. The decision timeline and recruitment results are processed by a decision timeline function to match each recruitment result with a corresponding decision timeline and assign a unique timeline identifier; the causal relationship between different agent nodes within a single decision timeline is described by a causal association function; and the causal chain between agent nodes is determined by a decision causal inference model, wherein the causal chain indicates how the input of the preceding node affects the output of the current node and how the preceding decision affects the subsequent decision.

2. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration as described in claim 1, characterized in that, The method also includes: The decision timeline is optimized based on historical data using a dynamic optimization algorithm, wherein at each decision node, the effectiveness of each decision is measured by a reward function.

3. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration as described in claim 1, characterized in that, The method also includes: To determine whether the execution order of the intelligent agent nodes is compliant, the time interval of each node is calculated using a preset time sequence consistency measurement rule; and anomalies are determined based on the time interval. Verify the logical relationship between the previous input and the current output; The decision-making chain is updated by removing non-compliant data and data with abnormal logical relationships.

4. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration according to claim 1, characterized in that, The method also includes quantifying the strength of the causal relationship.

5. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration according to claim 1, characterized in that, Processing the decision timeline and recruitment results using the decision timeline function includes: use Indicates the first The recruitment results correspond to the decision-making timeline as follows: Its function expression is as follows: ,in, Indicates the first Timeline of decision-making for each recruitment outcome This represents decision data related to the outcome. It is the unique identifier for that timeline. This indicates the processing logic and process information for this timeline.

6. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration according to claim 1, characterized in that, The causal relationships between different agent nodes within a single decision timeline, described using causal association functions, include: ,in, Indicates the first The causal relationships within the decision-making timeline constitute the overall causal relationships of a recruitment process. and These represent two decision points in the timeline; causal relationship. This indicates the filtering results. How to influence interview scheduling The decision.

7. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration according to claim 1, characterized in that, The causal chain between agent nodes is determined based on the decision causal inference model, including: set up Indicates the first The input of each node, Indicates its output, Representing the decision-making process, the causal inference function is: ,in, Indicates the first The set of parent nodes of each node. Given a parent node hour, right The impact, Parent node The probability distribution.

8. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration according to claim 2, characterized in that, Optimizing the decision timeline based on historical data using a dynamic optimization algorithm includes: assuming that at the... There are nodes, and their states are... Action for reward function for ,in, , , The weighting coefficient represents a comprehensive consideration of accuracy, time efficiency, and cost-effectiveness.

9. The AI ​​recruitment closed-loop automated control method based on multi-agent collaboration according to claim 4, characterized in that, Quantifying the strength of the causal relationship includes: ,in, Represents a node With nodes The strength score of the causal relationship between them For the first The decision samples from 1 to n are summed cumulatively, where n is the total number of samples. It is in the In a sample, the node The output result; It is in the In a sample, the node Its own input data; It is in the In each sample, the preceding parent node The input data, i.e., the data for the nodes Pre-existing data that leads to causal effects; For conditional probability, given a node enter With parent node enter At that time, node Output The probability reflects right The actual degree of impact, The prior probability, i.e., the previous node Input The probability of occurrence in all samples is used to weight and normalize the causal effects.

10. A closed-loop automated control system for AI recruitment based on multi-agent collaboration, characterized in that, include: The loading unit preloads each agent node, which is set based on the recruitment process. Each agent node is defined with output content, decision basis for the output content, weight of each factor in the decision basis, data source for the node, and triggering conditions. The execution unit triggers the first agent node after obtaining recruitment requirements. Each agent node starts and executes the recruitment process in sequence according to the trigger conditions. During this process, the recruitment results determined by each agent node are obtained. The unit also connects all the structured recruitment decision timelines generated by the agent nodes in chronological order, which include timestamps, inputs, processing logic, and output fields. In the process of each agent node, recruitment process data is collected in real time and classified and labeled according to preset rules. A unique timestamp is added to each piece of collected data to ensure that each piece of data can be located to the corresponding agent node and the time of generation. The inference unit processes the decision timeline and recruitment results through a decision timeline function to match each recruitment result with a corresponding decision timeline and assign a unique timeline identifier; it describes the causal relationship between different agent nodes within a single decision timeline based on a causal association function; and it determines the causal chain between agent nodes based on a decision causal inference model, wherein the causal chain indicates how the input of the preceding node affects the output of the current node and how the preceding decision affects the subsequent decision.

Images