Multi-agent voice interaction recruitment automation method and device

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses a multi-agent voice interaction recruitment automation method and device, and the method comprises the steps: obtaining and processing recruitment demand voice, and obtaining a structured demand; the JD generation agent receives and processes the structured demand to obtain compliant JD and JD key information; the resume screening agent obtains JD key information, obtains compliance JD based on the recruitment platform, and processes the JD key information and the compliance JD to obtain a candidate block; the interview scheduling agent receives and processes the candidate block to obtain interview arrangement information; obtaining and processing interview completion feedback according to the interview arrangement information to obtain an interview progress; the evaluation recommendation agent receives the interview progress and obtains evaluation data, and candidate ranking and a recommendation report are obtained; and the voice interaction module receives and processes the candidate ranking and recommendation report, and obtains and plays voice broadcast data. The recruitment efficiency can be greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a multi-agent voice interaction recruitment automation method and device. BACKGROUND

[0002] At present, the recruitment process of small and medium-sized enterprises mainly relies on traditional manual operation or basic recruitment tools. The description of JD (Job description, job description) is still manually written by enterprise HR or business department employees and then adjusted by referring to the general template of the industry. It cannot check the compliance of the JD. Secondly, the screening of the resume also depends on the subjective judgment of the HR, and there is a lack of unified quantitative scoring system. Thirdly, the coordination of the schedule with the interviewers and candidates through telephone, WeChat or email is inefficient, and it cannot automatically trigger the time arrangement of multiple rounds of interviews. Moreover, the evaluation results are easily affected by the subjective factors of the interviewers, and the integration of interview evaluation data process is time-consuming.

[0003] Therefore, how to improve the recruitment efficiency and make the recruitment process automatic through intelligent means has become a problem to be solved. SUMMARY

[0004] The present application provides a multi-agent voice interaction recruitment automation method and device, which can realize the rapid conversion of recruitment demand and the automatic generation of compliant JD by constructing a processing mechanism based on a voice interaction module and multiple agents, thereby improving the input efficiency of recruitment demand and enhancing the compliance of JD. Through the collaborative architecture of multiple agents, a recruitment whole-process data link is constructed to realize the whole automation from the input of recruitment demand to the candidate recommendation report, thereby greatly improving the recruitment efficiency and improving the degree of automation of recruitment.

[0005] In a first aspect, the embodiments of the present application provide a multi-agent voice interaction recruitment automation method, which comprises:

[0006] The voice interaction module acquires the recruitment demand voice and processes it to obtain the structured demand and send it to the JD generation agent and the data storage module;

[0007] The JD generation agent receives the structured demand and processes it to obtain the compliant JD and send it to the recruitment platform, obtains the JD key information and sends it to the resume screening agent and the data storage module;

[0008] The resume screening agent acquires the JD key information, obtains the compliant JD based on the recruitment platform, processes the JD key information and the compliant JD, and obtains the candidate profile and sends it to the interview scheduling agent and the data storage module;

[0009] The interview scheduling intelligent agent receives the candidate profile and processes it to obtain interview arrangement information; according to the interview arrangement information, it obtains interview completion feedback and processes it to obtain interview progress and sends it to the evaluation recommendation intelligent agent and the data storage module;

[0010] The evaluation recommendation intelligent agent receives the interview progress and obtains evaluation data, processes according to the interview progress and the evaluation data, obtains the candidate ranking and the recommendation report, and sends it to the voice interaction module and the data storage module;

[0011] The data storage module receives and stores the structured demand, JD key information, candidate profile, interview progress, candidate ranking, and recommendation report;

[0012] The voice interaction module receives the candidate ranking and the recommendation report and processes it to obtain voice broadcast data and play it.

[0013] Further, the method further comprises:

[0014] The data storage module obtains and stores the recruitment demand, compliance JD, JD key information, interview progress, evaluation data, candidate ranking, and recommendation report to obtain whole-process data;

[0015] The data storage module analyzes the whole-process data to obtain a recruitment efficiency report.

[0016] Further, the voice interaction module obtains the recruitment demand voice and processes it to obtain the structured demand and sends it to the JD generation intelligent agent and the data storage module, including:

[0017] The voice interaction module obtains the recruitment demand voice using WebRTC technology, recognizes the recruitment demand voice using an intelligent cloud ASR engine, and obtains text data;

[0018] The voice interaction module builds a semantic parsing model using a long short-term memory network, processes the text data according to the semantic parsing model, and obtains the structured demand;

[0019] The voice interaction module performs integrity verification on the structured demand according to the preset key information to determine whether the structured demand is complete.

[0020] If yes, the voice interaction module obtains the structured demand and sends it to the JD generation intelligent agent and the data storage module.

[0021] Further, the JD generation intelligent agent receives the structured demand and processes it to obtain the compliance JD and sends it to the recruitment platform, obtains the JD key information and sends it to the resume screening intelligent agent, including:

[0022] The JD generation intelligent agent receives the structured demand, matches an industry JD template library and the structured demand based on the job title using a GPT-4 model, and obtains initial JD content;

[0023] The JD generation agent checks the initial JD content by using preset compliance verification rules to obtain a verification pass rate, and determines whether the verification pass rate is greater than a preset threshold.

[0024] If yes, the JD generation agent obtains a compliant JD, converts the compliant JD into a preset format, and sends the compliant JD to the recruitment platform; extracts the job requirement keywords in the compliant JD, obtains JD key information, and sends the JD key information to the resume screening agent and the data storage module.

[0025] Further, the resume screening agent obtains the JD key information, obtains the compliant JD based on the recruitment platform, processes the JD key information and the compliant JD, obtains a candidate profile, and sends the candidate profile to the interview scheduling agent and the data storage module, including:

[0026] The resume screening agent obtains the JD key information, obtains the compliant JD based on the recruitment platform, uses a PDFBox tool or a POI tool to parse the compliant JD, obtains structured resume data, and sends the structured resume data to the data storage module.

[0027] The data storage module saves the structured resume data.

[0028] The resume screening agent uses a BERT model to process the structured resume data and the JD key information, obtains a candidate profile, and sends the candidate profile to the interview scheduling agent and the data storage module.

[0029] Further, the interview scheduling agent receives the candidate profile and processes the candidate profile to obtain interview scheduling information; obtains interview completion feedback based on the interview scheduling information and processes the interview completion feedback to obtain interview progress, and sends the interview progress to the evaluation recommendation agent and the data storage module, including:

[0030] The interview scheduling agent receives the candidate profile, obtains schedule data, processes the candidate profile and the schedule data based on a greedy algorithm, and obtains interview scheduling information.

[0031] The interview scheduling agent obtains interview completion feedback based on the interview scheduling information and processes the interview completion feedback to obtain interview progress, and sends the interview progress to the evaluation recommendation agent and the data storage module.

[0032] Further, the evaluation recommendation agent receives the interview progress and obtains evaluation data, processes the interview progress and the evaluation data, obtains a candidate ranking and a recommendation report, and sends the candidate ranking and the recommendation report to the voice interaction module and the data storage module, including:

[0033] The evaluation recommendation agent receives interview progress and obtains evaluation data, processes the interview progress and the evaluation data according to a data integrity check algorithm and an outlier detection model, obtains evaluation data integrity data and accuracy data, and judges whether the evaluation data integrity data is greater than a preset first threshold and the accuracy data is greater than a preset second threshold.

[0034] If yes, the evaluation recommendation agent processes the evaluation data by using a hierarchical weighting algorithm, obtains candidate ranking and a recommendation report, and sends the candidate ranking and the recommendation report to the voice interaction module and the data storage module.

[0035] In a second aspect, the embodiments of the present application provide a recruitment automation device for multi-agent voice interaction, which comprises:

[0036] The voice interaction module is configured to obtain and process recruitment demand voice to obtain structured demand and send the structured demand to the JD generation agent and the data storage module, and receive candidate ranking and a recommendation report and process the candidate ranking and the recommendation report to obtain voice broadcast data and play the voice broadcast data.

[0037] The JD generation agent is configured to receive and process structured demand to obtain a compliant JD and send the compliant JD to a recruitment platform, and obtain JD key information and send the JD key information to the resume screening agent and the data storage module.

[0038] The resume screening agent is configured to obtain JD key information, obtain a compliant JD based on a recruitment platform, process the JD key information and the compliant JD, obtain candidate profiles, and send the candidate profiles to the interview scheduling agent and the data storage module.

[0039] The interview scheduling agent is configured to receive and process candidate profiles to obtain interview arrangement information, obtain interview completion feedback based on the interview arrangement information, and process the interview completion feedback to obtain interview progress and send the interview progress to the evaluation recommendation agent and the data storage module.

[0040] The evaluation recommendation agent is configured to receive interview progress and obtain evaluation data, process the interview progress and the evaluation data, obtain candidate ranking and a recommendation report, and send the candidate ranking and the recommendation report to the voice interaction module and the data storage module.

[0041] The data storage module is configured to receive and store structured demand, JD key information, candidate profiles, interview progress, candidate ranking, and a recommendation report.

[0042] Further, the data storage module is further configured to obtain and store recruitment demand, a compliant JD, JD key information, interview progress, evaluation data, candidate ranking, and a recommendation report to obtain whole-process data, and analyze the whole-process data to obtain a recruitment efficiency report.

[0043] Further, the voice interaction module is further configured to acquire the voice of the recruitment demand by using a WebRTC technology, recognize the voice of the recruitment demand by using an intelligent cloud ASR engine, and obtain text data; construct a semantic analysis model by using a long short-term memory network, process the text data according to the semantic analysis model, and obtain structured demand; perform integrity checking on the structured demand according to preset key information, and determine whether the structured demand is complete; if yes, the structured demand is obtained and sent to the JD generation agent and the data storage module.

[0044] Compared with the prior art, the technical scheme provided by the embodiment of the application has at least the following beneficial effects:

[0045] The recruitment automation method for multi-agent voice interaction provided by the embodiment of the application can realize rapid conversion of recruitment demand and automatic generation of compliant JD by constructing a processing mechanism based on the voice interaction module and the plurality of agents, thereby improving the input efficiency of the recruitment demand and enhancing the compliance of the JD; the recruitment full-process data link is constructed by the collaborative architecture of the plurality of agents, and the whole process automation from input of the recruitment demand to the candidate recommendation report is realized, thereby greatly improving the recruitment efficiency and improving the degree of automation of the recruitment. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A flowchart of a recruitment automation method for multi-agent voice interaction is provided for one exemplary embodiment of the application.

[0047] Figure 2 A structural diagram of a recruitment automation device for multi-agent voice interaction is provided for one exemplary embodiment of the application. DETAILED DESCRIPTION

[0048] The technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments of the application.

[0049] Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the application.

[0050] Please refer to Figure 1 The embodiment of the application provides a recruitment automation method for multi-agent voice interaction, which specifically includes the following steps:

[0051] In step S1, a voice interaction module acquires the voice of the recruitment demand and processes the voice to obtain structured demand and send the structured demand to a JD generation agent and a data storage module.

[0052] In some embodiments, the voice interaction module obtains the recruitment demand voice and processes to obtain the structured demand and sends to the JD generation agent and the data storage module, including:

[0053] The voice interaction module obtains the recruitment demand voice by using the WebRTC technology, and recognizes the recruitment demand voice by using the intelligent cloud ASR engine to obtain the text data.

[0054] The voice interaction module constructs a semantic analysis model by using the long short-term memory network, and processes the text data according to the semantic analysis model to obtain the structured demand.

[0055] The voice interaction module performs integrity verification on the structured demand according to the preset key information to determine whether the structured demand is complete.

[0056] If yes, the voice interaction module obtains the structured demand and sends to the JD generation agent and the data storage module.

[0057] Among them, the HR or business person in charge of small and medium-sized enterprises can input the natural language recruitment demand voice through the voice interaction entrance of APP or web page, such as "I need to recruit 1 e-commerce operation officer, belonging to the operation department, mainly responsible for the operation of Taobao and Jingdong platform, requires 2 years of e-commerce operation experience, bachelor's degree, salary range 6000-8000 yuan / month". The voice interaction module obtains the recruitment demand voice by using the WebRTC technology, and the sampling rate can be set to 16 kHz and the bit depth to 16 bit to ensure clear voice signal and avoid affecting the subsequent text conversion accuracy due to signal distortion.

[0058] In some embodiments, the intelligent cloud ASR engine is used to recognize the recruitment demand voice, the recognition language is set to Chinese, and the recognition accuracy threshold is greater than or equal to 95%, the recruitment demand voice is converted into text data; if the recognition accuracy is less than 95%, the user is prompted to re-input through voice, the voice for prompting can be "your demand is not clearly recognized, please describe the recruitment position requirements again in a quiet environment", until the text data with qualified accuracy is obtained.

[0059] In some embodiments, a semantic analysis model is constructed based on a long short-term memory network to extract keywords and structure the processed text. First, the text is segmented by using a segmentation tool, and the way of Jieba segmentation can be used to remove meaningless stop words such as "of" and "need"; then the entity recognition algorithm is used to label the key information category, and the BiLSTM-CRF can be used to extract the key information such as job title, department, number of recruitment, core responsibilities, qualifications and salary range, to obtain the structured demand.

[0060] In some embodiments, the structured demand can be as shown in Table 1:

[0061] Table 1

[0062]

[0063]

[0064] In some embodiments, the voice interaction module performs integrity checking on the generated structured data, and the preset key information includes the job title, the number of recruitment, the core responsibilities and the job requirements; if any key information is missing, the user can be asked to supplement through voice; after the user supplements the complete information, the final confirmed structured data is sent to the JD generation agent and the data storage module, ensuring that the subsequent JD generation has complete and accurate requirements.

[0065] Step S2, the JD generation agent receives the structured requirements and processes them to obtain a compliant JD and sends it to the recruitment platform, and obtains the JD key information and sends it to the resume screening agent and the data storage module.

[0066] In some embodiments, the JD generation agent receives the structured requirements and processes them to obtain a compliant JD and sends it to the recruitment platform, and obtains the JD key information and sends it to the resume screening agent, including:

[0067] The JD generation agent receives the structured requirements, matches the industry JD template library and the structured requirements based on the job title using the GPT-4 model, and obtains the initial JD content.

[0068] The JD generation agent uses preset compliance checking rules to check the initial JD content, obtains a pass rate, and determines whether the pass rate is greater than a preset threshold.

[0069] If yes, the JD generation agent obtains a compliant JD, converts the compliant JD into a preset format and sends it to the recruitment platform; extracts the job requirements keywords in the compliant JD, obtains the JD key information and sends it to the resume screening agent and the data storage module.

[0070] In some embodiments, GPT-4 is used as the basic large model, LoRA technology is used to fine-tune based on the industry JD corpus (which can include 10 industries such as e-commerce, finance, manufacturing, etc., a total of 500,000 compliant JD data), the learning rate can be set to 2e-4 and trained for 10 rounds, and the method of gradient descent is used to optimize the parameters of the GPT-4 model, ensuring that the GPT-4 model can capture the expression specifications and core points of JDs in different industries, and generate initial JD content that meets the characteristics of the industry.

[0071] In some embodiments, the JD generation agent reads the structured recruitment requirements generated in step 1 from the data storage module, matches the industry JD template library based on the job title, for example, "E-commerce Operation Specialist" matches "E-commerce Industry Operation Position Template"; the industry JD template library includes multiple fixed fields (job basic information, job responsibilities, job requirements, salary and benefits, company introduction, and recruitment methods) and variable fields (dynamically filled according to requirements); for example, the "salary and benefits" in the fixed field contains general content such as paid annual leave, and the "job responsibilities" in the variable field is dynamically generated according to the core responsibilities in the structured requirements, which can avoid the disconnection between templated content and actual requirements.

[0072] In some embodiments, the GPT-4 model generates initial JD content based on structured requirements and industry JD template library. For example, based on the core responsibilities, combined with the "Taobao / Jingdong platform store operation" requirement, generate "responsible for daily operation of Taobao and Jingdong stores, including product listing, price adjustment and inventory management; plan and execute store promotion activities, optimize activity strategies through data analysis, and improve store traffic and sales"; for example, based on job requirements, combined with "2 years of e-commerce operation experience, bachelor's degree" requirement, generate "bachelor's degree or above, 2 years or above of Taobao and Jingdong platform operation experience; familiar with e-commerce platform rules (such as Taobao direct bus, Jingdong fast car promotion mechanism), have independent activity planning and execution ability, and can use Excel, SQL for basic data analysis", to ensure that the content is specific and can be implemented, and to avoid vague expressions.

[0073] In some embodiments, the JD generation agent calls the built-in compliance verification rule library, which is based on laws and regulations and contains more than 100 verification rules, to check the initial JD one by one. In some embodiments, rule 1: prohibit the inclusion of discriminatory provisions such as age, gender, and region, such as detecting "35 years old" and "male only" expressions, which are automatically deleted; rule 2: must mention the salary range or salary composition, such as not mentioning it, which is supplemented based on industry average (by calling third-party industry salary database, such as e-commerce operation specialist industry average of 5500-7500 yuan / month), for example, "salary range 6000-8000 yuan / month, including basic salary 4500 yuan + performance bonus 1500-3500 yuan"; rule 3: job responsibility description needs to be clear, avoiding vague expressions such as "responsible for related work", such as detecting it, which is optimized to specific responsibilities, for example, "responsible for store operation related work" is optimized to "responsible for store product management and activity operation related work". The verification pass rate must be greater than or equal to the preset threshold, which can be 98%; if it does not pass, it will be regenerated until it meets the compliance requirements, reducing the legal risk of enterprises.

[0074] In some embodiments, the JD generation agent converts compliant JDs into a preset HTML format. On one hand, it sends the JD to the recruitment platform via an API interface, automatically filling in the platform's posting form and submitting it without requiring manual login. On the other hand, it extracts job requirement keywords (such as "2 years of e-commerce operation experience", "Bachelor's degree", "event planning", "SQL data analysis") and job responsibility categories (such as "daily store operation", "event planning", "data analysis") from the JD and synchronizes them to the resume screening agent and data storage module. This provides clear matching criteria for subsequent resume screening, ensuring that the screening dimensions are highly consistent with the JD requirements.

[0075] Step S3: The resume screening agent obtains key JD information, acquires compliant JDs based on the recruitment platform, processes the key JD information and compliant JDs, obtains candidate profiles, and sends them to the interview scheduling agent and data storage module.

[0076] In some embodiments, the resume screening agent obtains key JD information, acquires compliant JDs based on the recruitment platform, processes the key JD information and compliant JDs, obtains candidate profiles, and sends them to the interview scheduling agent and data storage module, including:

[0077] The resume screening agent obtains key information from job descriptions, acquires compliant job descriptions from recruitment platforms, parses compliant job descriptions using PDFBox or POI tools, obtains structured resume data, and sends it to the data storage module.

[0078] The data storage module stores structured resume data.

[0079] The resume screening agent uses the BERT model to process structured resume data and key JD information, obtains candidate profiles, and sends them to the interview scheduling agent and data storage module.

[0080] In some embodiments, after the recruitment platform publishes the job description, candidates begin to submit their resumes, at which point the resume screening agent starts working.

[0081] In some embodiments, the recruitment platform pushes resume data (in PDF or Word format) to the resume screening agent in real time via API. The resume screening agent uses PDFBox to parse PDF resumes and POI to parse Word resumes, extracting basic candidate information (name, age, education, work experience), skills, work experience, etc., converting it into structured resume data, and storing it in the data storage module to avoid information omissions due to inconsistent resume formats.

[0082] Specifically, first, a BERT model fine-tuned for vertical recruitment is used to parse the resume text. The BERT model is trained based on a large amount of text data in the recruitment field and can understand the semantic information in the resume. After that, a word segmentation tool is combined to perform word segmentation on the resume text, splitting the text into multiple meaningful lexical units while removing meaningless stop words such as "de" (of), "shi" (is), etc. Then, the BiLSTM-CRF entity recognition algorithm is used to label the key information categories and extract key information such as job titles, work experience, skills, etc. For example, for the content in the resume "From March 2021 to May 2023, an e-commerce operation specialist at a certain e-commerce company, responsible for the operation of the JD store, increasing the monthly sales of the store from 500,000 yuan to 1,200,000 yuan", after the above processing, the structured resume data is presented in the following format: presented in a table form, including the field "Work Experience" with the content "From March 2021 to May 2023, an e-commerce operation specialist at a certain e-commerce company"; the field "Responsibility Description" with the content "Responsible for the operation of the JD store"; the field "Achievements" with the content "Increasing the monthly sales of the store from 500,000 yuan to 1,200,000 yuan", etc. The structured format makes the resume information clearer, easier to retrieve and analyze.

[0083] In some embodiments, the resume screening agent uses a BERT model fine-tuned based on a recruitment corpus, which contains more than 100,000 resumes and corresponding screening result data, such as "Passed the screening" and "Failed to pass the screening" labels. The training objective is "inputting resume data and JD key information and outputting a matching score". The training process can use the Adam optimizer, the learning rate can be set to 1e-4, and train for 20 rounds. Adjust the parameters of the BERT model through cross-validation to ensure that the accuracy of the BERT model on the test set is greater than or equal to 85%, so as to accurately identify the matching degree between the resume and the job requirements.

[0084] Specifically, first, collect a large amount of resume data as the training dataset and preprocess it, such as cleaning the data, annotating key information, etc., to ensure the data quality; second, select the BERT model as the base model and fine-tune it for the vertical recruitment field; finally, use the test set to comprehensively evaluate the model and calculate indicators such as accuracy, recall rate, and F1 value to ensure that the BERT model can be well applied.

[0085] In some embodiments, the resume screening agent calculates the matching score between the candidate and the job from 5 dimensions, with a full score of 100 points. Among them, the weights and calculation logics of each dimension can be as shown in Table 2 below:

[0086] Table 2

[0087]

[0088] Finally, the resume screening agent pushes the candidate profiles to the interview scheduling agent, and stores all candidates' rating data and structured resume data in the data storage module to provide data support for subsequent interview scheduling and evaluation. The entire screening process for each resume takes less than 1 second, which is more than 100 times more efficient than manual screening.

[0089] In step S4, the interview scheduling agent receives and processes candidate blocking information to obtain interview arrangement information; based on the interview arrangement information, it obtains and processes interview completion feedback to obtain interview progress and sends it to the evaluation and recommendation agent and data storage module.

[0090] In some embodiments, the interview scheduling agent receives and processes candidate blocking to obtain interview arrangement information; it obtains and processes interview completion feedback based on the interview arrangement information to obtain interview progress and sends it to the evaluation recommendation agent and data storage module, including:

[0091] The interview scheduling agent receives candidate profiles, obtains schedule data, processes the candidate profiles and schedule data using a greedy algorithm, and obtains interview arrangement information.

[0092] The interview scheduling agent obtains and processes the interview completion feedback based on the interview arrangement information, obtains the interview progress, and sends it to the evaluation and recommendation agent and the data storage module.

[0093] In some embodiments, the interview scheduling agent can be configured to obtain candidate authorization via a pop-up window, synchronize their calendar data, extract the candidate's available interview time slots (default 9:00-18:00 daily, excluding already occupied time slots); simultaneously, it connects to the enterprise's OA system, reads the interviewer's schedule data, extracts available interview time slots, and stores all schedule data in real time to the data storage module to ensure that scheduling is based on the latest schedule information.

[0094] In some embodiments, an enterprise's OA system requires enterprise authorization and can connect using methods such as API interface keys, usernames and passwords, and OAuth authentication. During the connection process, the system interfaces according to the protocol provided by the OA system, calling the OA system's interfaces to obtain information such as the interviewer's schedule data; it can also parse and process the received schedule data, converting it into an internal data format.

[0095] In some embodiments, the interview scheduling agent can use a greedy algorithm to coordinate the interview times for each round in the order of initial interview, second interview, and final interview. Taking the initial interview coordination as an example, the algorithm logic is as follows: Extract the available interview time slots for all candidates and the available interview time slots for the initial interviewers, and calculate the intersection time slots; sort the intersection time slots in descending order of time slot length, prioritizing longer time slots, such as two-hour time slots; allocate a 30-minute interview slot within the intersection time slot for each candidate, ensuring that the time slots do not overlap; if a candidate has no overlapping time slot with the initial interviewer, automatically generate three alternative time slots and push them to the candidate via SMS / email, and update the schedule after the candidate confirms. After the initial interview time is determined, the agent automatically associates the schedules of the interviewers for the second and final interviews, and pre-allocates the interview time for subsequent rounds according to the same logic (e.g., conducting the second interview within two working days after passing the initial interview), forming a complete multi-round interview time plan and avoiding repeated coordination in subsequent scheduling. Subsequently, interview notifications are pushed out: the agent generates standardized interview notifications, including the interview round, time, method (voice interview link, generated based on the system's built-in WebRTC voice interaction function), interviewer information, and interview preparation items (such as resume and portfolio display requirements). These notifications are simultaneously pushed to candidates and interviewers via SMS, email, and system messages, ensuring an information reach rate of ≥99%. At the same time, the interview arrangement information (which may include candidate ID, interviewer ID, interview time, and link) is stored in the data storage module and synchronized to the evaluation and recommendation agent for subsequent interview progress tracking.

[0096] In some embodiments, during the interview process, the interview scheduling agent tracks the interview progress in real time: if a candidate does not enter the interview link on time, an SMS reminder is automatically sent; if an interviewer is temporarily unable to participate, the interview scheduling agent immediately selects a replacement from the backup interviewer pool (users can also set 2-3 backup interviewers in advance), reschedules the interview, and pushes an update notification to ensure that the interview delay rate is less than 5%; after each round of interviews, the interviewer submits feedback on the completion of the interview through the system, and the interview scheduling agent updates the progress to "initial interview completed and awaiting re-interview" or "interview failed," and synchronizes it to the data storage module and the evaluation and recommendation agent.

[0097] In some embodiments, after the interview scheduling agent determines the interview time, it automatically activates the built-in voice interview function. First, it prepares the interview environment. After the candidate clicks the interview link to enter the voice room, it automatically detects the microphone and speaker status, prompting the candidate to choose a quiet environment to ensure clear audio signals. Simultaneously, it provides the interviewer with an interview outline generation function, automatically generating interview questions based on key information in the job description, preventing the interviewer from asking questions that deviate from the job requirements.

[0098] In some embodiments, voice recording and real-time transcription are enabled during the interview, and a complete voice file is recorded simultaneously. The transcribed text is displayed in real time on the interviewer's interface, making it convenient for the interviewer to quickly review the conversation without manual recording. After the interview, the voice file and transcribed text are automatically associated with the candidate ID and stored in the data storage module as a reference for subsequent evaluation.

[0099] In some embodiments, quantitative assessment data collection is then initiated to generate a standardized quantitative assessment form for the interviewer, containing 5 core assessment dimensions, each scored from 1 to 5, with 1 point being poor and 5 points being excellent, as shown in Table 3 below:

[0100] Table 3

[0101]

[0102] In some embodiments, the interviewer completes and submits the evaluation form within one hour after the interview. The evaluation data is automatically synchronized to the evaluation recommendation agent and data storage module. If the interviewer fails to submit on time, an email reminder is sent to ensure the timely collection of evaluation data.

[0103] Step S5: The evaluation and recommendation agent receives the interview progress and obtains the evaluation data. Based on the interview progress and evaluation data, it processes the data to obtain the candidate ranking and recommendation report, and sends them to the voice interaction module and data storage module.

[0104] In some embodiments, the evaluation and recommendation agent receives interview progress and obtains evaluation data, processes the interview progress and evaluation data, obtains candidate rankings and recommendation reports, and sends them to the voice interaction module and data storage module, including:

[0105] The evaluation and recommendation agent receives the interview progress and obtains the evaluation data. It processes the interview progress and evaluation data according to the data integrity verification algorithm and the outlier detection model to obtain the evaluation data integrity data and accuracy data. It then determines whether the evaluation data integrity data is greater than a preset first threshold and the accuracy data is greater than a preset second threshold.

[0106] If so, the evaluation and recommendation agent uses a hierarchical weighted algorithm to process the evaluation data, obtain candidate rankings and recommendation reports, and send them to the voice interaction module and data storage module.

[0107] In some embodiments, after receiving the evaluation data from each round of interviews, the evaluation recommendation agent first performs evaluation data verification: checking for omissions or outliers in the evaluation form (all dimensions are scored 1 or 5 points). If omissions are found, a reminder to the interviewer to fill in the missing information is automatically sent. If outliers are found, a second verification is performed using the transcript of the interview speech. If the data is confirmed to be correct, it is retained; otherwise, the interviewer is prompted to re-evaluate, ensuring that the completeness of the evaluation data is greater than a preset first threshold of 99% and the accuracy is greater than a preset second threshold of 95%. Subsequently, the evaluation data from multiple rounds is weighted and integrated: a hierarchical weighting algorithm is used, setting different weights according to the importance of the interview rounds (initial interview weight 0.3, second interview weight 0.4, final interview weight 0.3), to calculate the candidate's comprehensive evaluation score, as shown in the following formula:

[0108]

[0109] Where S is the candidate's overall evaluation score, n is the number of interview rounds (e.g., 3 rounds), Wi is the weight of the i-th round of interviews, and Si is the average score of the i-th round of interviews (the average score of each evaluation dimension in the i-th round of interviews).

[0110] Furthermore, the evaluation and recommendation agent sorts all candidates who have passed multiple rounds of interviews in descending order of their comprehensive evaluation scores to obtain a candidate ranking. If the scores are the same, the final interview scores are compared. If the final interview scores are still the same, the scores of the work experience matching dimension are compared to ensure that the ranking is unique. In addition, a recommendation report for the candidate is generated, which includes the ranking, comprehensive score, details of each round of evaluation, interviewer comments, and key information from the resume. The recommendation report is in PDF format and can be downloaded and viewed online.

[0111] In step S6, the data storage module receives and stores the structured requirements, key JD information, candidate profiles, interview progress, candidate rankings, and recommendation reports.

[0112] In step S7, the voice interaction module receives and processes the candidate ranking and recommendation report to obtain and play the voice broadcast data.

[0113] In some embodiments, the evaluation and recommendation agent sends candidate rankings and recommendation reports to the voice interaction module and the data storage module. The voice interaction module informs the user through voice broadcast, and the user can also view the detailed report. The data storage module saves the candidate rankings and recommendation reports, thereby enabling subsequent hiring decisions and historical data tracing.

[0114] The multi-agent voice interaction-based recruitment automation method provided in the above embodiments can achieve rapid conversion of recruitment needs and automatic generation of compliant job descriptions by constructing a processing mechanism based on a voice interaction module and multiple agents, thereby improving the input efficiency of recruitment needs and enhancing the compliance of job descriptions. Through the collaborative architecture of multiple agents, a full-process data link for recruitment is constructed, realizing full automation from the input of recruitment needs to the candidate recommendation report, thereby significantly improving recruitment efficiency and increasing the degree of automation in recruitment.

[0115] In some embodiments, the method further includes:

[0116] The data storage module acquires and stores recruitment requirements, compliant job descriptions, key job description information, interview progress, evaluation data, candidate rankings, and recommendation reports, thus obtaining full-process data.

[0117] The data storage module analyzes the data from the entire process to generate a recruitment efficiency report.

[0118] In some embodiments, after the evaluation and recommendation agent generates candidate rankings, the system enters the recruitment decision support stage: a candidate recommendation report is pushed to the user, and recruitment suggestions are generated based on historical recruitment data stored in the data storage module; the user can also directly mark the recruitment status, and the system automatically synchronizes the results to the data storage module after marking.

[0119] In some embodiments, if a user chooses to hire, a personalized offer notification is generated based on the company's hiring template and the candidate's information. This notification is sent to the candidate via SMS and email, and also sent to the HR's email address for record-keeping. If a user chooses to decline, a polite rejection notification is automatically generated, informing the candidate that they are grateful for the interview, that no suitable position has been found for them yet, and that they will be contacted first when a suitable position becomes available, thus maintaining the company's image.

[0120] In some embodiments, a process data review can also be initiated: extracting the entire recruitment process data from the data storage module to generate a recruitment efficiency report for the company to optimize subsequent recruitment strategies; at the same time, adding the entire process data to the historical database to provide training data for subsequent model optimization (such as JD generation model, resume screening model), thereby forming a closed loop.

[0121] This application significantly improves recruitment efficiency, achieving an automation rate of over 90% across the entire process. Specifically, job description (JD) generation time is reduced from 2-3 hours manually to less than 5 minutes; resume screening efficiency increases from 10-15 resumes per hour to over 100 per second; interview scheduling time is shortened from 1-2 days manually to less than 10 minutes; and the recruitment cycle is compressed from an average of 20-30 days to 5-7 days, resulting in an overall recruitment efficiency improvement of over 70%. Simultaneously, this invention enhances the accuracy and compliance of recruitment. Resume screening accuracy increases from less than 60% manually to over 85%, preventing the overlooking of high-quality candidates; the JD compliance verification pass rate reaches 98%, automatically removing discriminatory clauses and supplementing compliant content, reducing legal risks for enterprises, and improving compliance by 90% compared to traditional manual methods. Furthermore, this invention reduces labor costs, decreasing repetitive work for HR in various stages, reducing the annual recruitment labor cost ratio for SMEs from 8%-12% to 3%-5%. Based on a 50-person company recruiting an average of 10 people annually, this translates to annual labor cost savings of 30,000-50,000 yuan. In terms of candidate experience, the automatic coordination of interview scheduling reduces the waiting time for candidate feedback from 3-5 days to less than 24 hours. Interview notifications are delivered through multiple channels, delay reminders are timely, and rejection notices are standardized, improving candidate satisfaction and helping companies establish a good employer brand image.

[0122] This application also features data-driven recruitment optimization, with automatic storage and review of data throughout the entire process, generating recruitment efficiency reports to provide decision-making support for enterprises. Historical data is used for iterative model optimization, ensuring the system's recruitment capabilities continuously improve over time, creating a virtuous cycle. Furthermore, this application boasts high adaptability to various scenarios, supporting campus recruitment, social recruitment, internal selection, and more. By configuring JD templates and adjusting scoring weights, it can be adapted to different industries and job types, enabling SMEs to quickly deploy without custom development, and offering high coverage of applicable scenarios.

[0123] Please see Figure 2 Another embodiment of this application provides a multi-agent voice interaction-based automated recruitment device, the device comprising:

[0124] The voice interaction module is used to acquire and process recruitment requirements in voice, obtain structured requirements, and send them to the JD generation agent and data storage module; it also receives and processes candidate rankings and recommendation reports, obtains voice broadcast data, and plays it.

[0125] The JD generation agent is used to receive and process structured requirements, obtain compliant JDs and send them to the recruitment platform, obtain key JD information and send it to the resume screening agent and data storage module.

[0126] The resume screening agent is used to obtain key information about job descriptions (JDs), acquire compliant JDs based on recruitment platforms, process key JD information and compliant JDs, obtain candidate profiles, and send them to the interview scheduling agent and data storage module.

[0127] The interview scheduling agent receives and processes candidate blocking requests to obtain interview arrangement information; it then obtains and processes interview completion feedback based on the interview arrangement information, gets the interview progress, and sends it to the evaluation and recommendation agent and data storage module.

[0128] The evaluation and recommendation agent is used to receive interview progress and obtain evaluation data. Based on the interview progress and evaluation data, it processes the data to obtain candidate rankings and recommendation reports, which are then sent to the voice interaction module and data storage module.

[0129] The data storage module is used to receive and store structured requirements, key JD information, candidate profiles, interview progress, candidate rankings, and recommendation reports.

[0130] In some embodiments, the data storage module is also used to acquire and store recruitment requirements, compliant job descriptions, key job description information, interview progress, evaluation data, candidate rankings, and recommendation reports to obtain full-process data; and to analyze the full-process data to obtain a recruitment efficiency report.

[0131] In some embodiments, the voice interaction module is further configured to acquire recruitment requirement voice using WebRTC technology, recognize the recruitment requirement voice using an intelligent cloud ASR engine to obtain text data; construct a semantic parsing model using a long short-term memory network, process the text data according to the semantic parsing model to obtain structured requirements; perform integrity verification on the structured requirements according to preset key information to determine whether the structured requirements are complete; if so, obtain the structured requirements and send them to the JD generation agent and data storage module.

[0132] The specific limitations of the multi-agent voice interaction-based automated recruitment device provided in this embodiment can be found in the above embodiment of the multi-agent voice interaction-based automated recruitment method, and will not be repeated here. Each module in the above-described multi-agent voice interaction-based automated recruitment device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0133] This application provides a computer device that may include a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it causes the processor to perform the steps of a multi-agent voice interaction-based automated recruitment method as described in any of the above embodiments.

[0134] The working process, working details, and technical effects of the computer device provided in this embodiment can be found in the embodiment of a multi-agent voice interaction-based recruitment automation method described above, and will not be repeated here.

[0135] This application provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of a multi-agent voice interaction-based recruitment automation method as described in any of the above embodiments. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0136] The working process, working details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiment of a multi-agent voice interaction-based recruitment automation method described above, and will not be repeated here.

[0137] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0138] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0139] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A multi-agent voice interaction-based automated recruitment method, characterized in that, The method includes: The voice interaction module acquires and processes the voice of recruitment requirements, obtains structured requirements, and sends them to the JD generation agent and data storage module. The JD generation agent receives and processes the structured requirements, obtains compliant JDs and sends them to the recruitment platform, obtains key JD information and sends it to the resume screening agent and data storage module; The resume screening agent obtains the key information of the JD, obtains the compliant JD based on the recruitment platform, processes the key information of the JD and the compliant JD, obtains the candidate profile, and sends it to the interview scheduling agent and data storage module. The interview scheduling agent receives and processes the candidate's block to obtain interview arrangement information; it obtains and processes the interview completion feedback based on the interview arrangement information to obtain the interview progress and sends it to the evaluation and recommendation agent and the data storage module. The evaluation and recommendation agent receives the interview progress and obtains the evaluation data. Based on the interview progress and the evaluation data, it processes the data to obtain the candidate ranking and recommendation report, and sends them to the voice interaction module and the data storage module. The data storage module receives and stores the structured requirements, key JD information, candidate profiles, interview progress, candidate rankings, and recommendation reports. The voice interaction module receives and processes the candidate ranking and the recommendation report to obtain voice broadcast data and play it.

2. The automated recruitment method using multi-agent voice interaction according to claim 1, characterized in that, The method further includes: The data storage module acquires and stores the recruitment requirements, the compliant job description, the key information of the job description, the interview progress, the evaluation data, the candidate ranking, and the recommendation report, thus obtaining full-process data. The data storage module analyzes the data from the entire process to obtain a recruitment efficiency report.

3. The automated recruitment method using multi-agent voice interaction according to claim 1, characterized in that, The voice interaction module acquires and processes the voice of recruitment requirements, obtains structured requirements, and sends them to the JD generation agent and data storage module, including: The voice interaction module uses WebRTC technology to acquire the voice of the recruitment requirements, and uses the intelligent cloud ASR engine to recognize the voice of the recruitment requirements and obtain text data. The voice interaction module uses a long short-term memory network to construct a semantic parsing model, and processes the text data according to the semantic parsing model to obtain structured requirements; The voice interaction module performs a completeness check on the structured requirements based on preset key information to determine whether the structured requirements are complete. If so, the voice interaction module obtains the structured requirements and sends them to the JD-generated agent and data storage module.

4. The automated recruitment method using multi-agent voice interaction according to claim 1, characterized in that, The JD generation agent receives and processes the structured requirements, obtains compliant JDs and sends them to the recruitment platform, and obtains key JD information and sends it to the resume screening agent, including: The JD generation agent receives the structured requirements and uses the GPT-4 model to match the job title with the industry JD template library and the structured requirements to obtain the initial JD content. The JD generation agent uses preset compliance verification rules to verify the initial JD content, obtains the verification pass rate, and determines whether the verification pass rate is greater than a preset threshold. If so, the JD generation agent obtains the compliant JD, converts the compliant JD into a preset format and sends it to the recruitment platform; it extracts the job requirement keywords from the compliant JD, obtains the key JD information and sends it to the resume screening agent and data storage module.

5. The automated recruitment method using multi-agent voice interaction according to claim 1, characterized in that, The resume screening agent obtains the key information of the job description (JD), acquires the compliant JD based on the recruitment platform, processes the key information of the JD and the compliant JD, obtains the candidate profile, and sends it to the interview scheduling agent and data storage module, including: The resume screening agent obtains the key information of the JD, acquires the compliant JD based on the recruitment platform, parses the compliant JD using PDFBox or POI tools, obtains structured resume data, and sends it to the data storage module. The data storage module stores the structured resume data; The resume screening agent uses the BERT model to process the structured resume data and the key information of the JD, obtains the candidate profile, and sends it to the interview scheduling agent and the data storage module.

6. The automated recruitment method using multi-agent voice interaction according to claim 1, characterized in that, The interview scheduling agent receives and processes the candidate's blocking information to obtain interview arrangement information; it then obtains and processes interview completion feedback based on the interview arrangement information to obtain the interview progress and sends it to the evaluation and recommendation agent and data storage module, including: The interview scheduling agent receives the candidate's profile, obtains the schedule data, processes the candidate's profile and the schedule data according to a greedy algorithm, and obtains the interview arrangement information. The interview scheduling agent obtains and processes the interview completion feedback based on the interview arrangement information, obtains the interview progress, and sends it to the evaluation and recommendation agent and the data storage module.

7. The automated recruitment method using multi-agent voice interaction according to claim 1, characterized in that, The evaluation and recommendation agent receives the interview progress and obtains evaluation data. Based on the interview progress and the evaluation data, it processes the data to obtain candidate rankings and recommendation reports, which are then sent to the voice interaction module and the data storage module. This includes: The evaluation and recommendation agent receives the interview progress and obtains the evaluation data. It processes the interview progress and the evaluation data according to the data integrity verification algorithm and the outlier detection model to obtain evaluation data integrity data and accuracy data. It then determines whether the evaluation data integrity data is greater than a preset first threshold and whether the accuracy data is greater than a preset second threshold. If so, the evaluation and recommendation agent processes the evaluation data using a hierarchical weighted algorithm to obtain candidate rankings and recommendation reports, which are then sent to the voice interaction module and the data storage module.

8. A multi-agent voice interaction automated recruitment device, characterized in that, The device includes: The voice interaction module is used to acquire and process recruitment requirements in voice, obtain structured requirements, and send them to the JD generation agent and data storage module; it also receives and processes the candidate rankings and recommendation reports to obtain and play the voice broadcast data. JD generation intelligent agent is used to receive and process the structured requirements, obtain compliant JDs and send them to the recruitment platform, obtain key JD information and send it to the resume screening intelligent agent and data storage module; A resume screening intelligent agent is used to obtain the key information of the JD, obtain the compliant JD based on the recruitment platform, process the key information of the JD and the compliant JD, obtain the candidate profile and send it to the interview scheduling intelligent agent and data storage module. An interview scheduling agent is used to receive and process the candidate's blocking information to obtain interview arrangement information; based on the interview arrangement information, it obtains and processes the interview completion feedback to obtain the interview progress and sends it to the evaluation and recommendation agent and the data storage module. An evaluation and recommendation agent is used to receive the interview progress and obtain evaluation data, process the interview progress and the evaluation data to obtain candidate ranking and recommendation report, and send them to the voice interaction module and data storage module; The data storage module is used to receive and store the structured requirements, key information of the JD, candidate profiles, interview progress, candidate rankings, and recommendation reports.

9. The automated recruitment device with multi-agent voice interaction according to claim 8, characterized in that, The data storage module is also used to acquire and store the recruitment requirements, the compliant job descriptions, the key information of the job descriptions, the interview progress, the evaluation data, the candidate rankings, and the recommendation reports to obtain full-process data; and to analyze the full-process data to obtain a recruitment efficiency report.

10. The automated recruitment device with multi-agent voice interaction according to claim 8, characterized in that, The voice interaction module is also used to acquire recruitment requirement voice using WebRTC technology, recognize the recruitment requirement voice using an intelligent cloud ASR engine to obtain text data; construct a semantic parsing model using a long short-term memory network, process the text data according to the semantic parsing model to obtain structured requirements; perform integrity verification on the structured requirements according to preset key information to determine whether the structured requirements are complete; if so, obtain the structured requirements and send them to the JD generation agent and data storage module.

Images