Multi-agent simulation interview method, device and equipment based on artificial intelligence and medium

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a multi-agent simulation interview method based on artificial intelligence, and the method comprises the steps: a first agent obtains the resume content and target post information of a job seeker; the first agent extracts a first interview question from a pre-constructed question bank according to the target post information and sends the first interview question to a second agent, wherein the target post information is matched with one or more dimensions of post requirements, post levels and question difficulty of the first interview question; the second agent generates a first interview answer according to the first interview question and the resume content of the job seeker and sends the first interview answer to the first agent; and the first agent adjusts the investigation knowledge point and the question difficulty of the second interview question according to the first interview answer and sends the investigation knowledge point and the question difficulty to the second agent. According to the method, a real interview process can be simulated through an interaction mode that the interviewer agent asks questions and the job seeker agent answers the questions.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based multi-agent simulated interview method, device, electronic device and storage medium. Background Art

[0002] With the rapid development of artificial intelligence (AI), its application is becoming increasingly widespread across industries. This is particularly true in the field of talent screening, where the intelligence of AI models continues to improve. AI interviews, using these models as interviewers, are becoming a crucial tool for companies to efficiently screen talent. This technology not only improves recruitment efficiency but also reduces recruitment costs for companies.

[0003] However, existing AI interview systems typically only play the role of the interviewer, lacking the ability to simulate the role of the job seeker. This prevents job seekers from fully understanding the interactive process between the two sides of the interview and from empathizing with the interviewer, thus limiting their learning and improvement. Furthermore, the questions set by existing AI interview systems often lack specificity and flexibility, and they are unable to generate personalized interview questions based on the applicant's information. This significantly reduces the effectiveness of simulated interviews and fails to truly reflect the applicant's abilities and potential. Summary of the Invention

[0004] In view of this, embodiments of the present application provide an artificial intelligence-based multi-agent simulated interview method, device, electronic device and storage medium for solving at least one technical problem.

[0005] An embodiment of the present application provides a multi-agent simulated interview method based on artificial intelligence, which is implemented based on a first agent and a second agent, wherein the first agent is used to play the role of an interviewer and the second agent is used to play the role of a job seeker. The method includes: the first agent obtains the resume content and target position information of the job seeker; the first agent extracts a first interview question from a pre-built question bank based on the target position information and sends it to the second agent, wherein the target position information is consistent with one or more dimensions of the job requirements, job level and question difficulty of the first interview question; the second agent generates a first interview answer based on the first interview question and the resume content of the job seeker and sends it to the first agent; the first agent adjusts the knowledge points and question difficulty of the second interview question based on the first interview answer and sends it to the second agent.

[0006] The method as described above further includes: the first agent calculates a simulated interview score based on the resume content and the interview answers provided by the second agent; and presents the conversation process between the first agent and the second agent and the simulated interview score to the job seeker.

[0007] According to the method described above, the first agent calculates the simulated interview score based on the resume content and the interview answers provided by the second agent, including: obtaining the background information of the job applicant from the resume content, the background information including: one or more of education, work experience and skills; calculating the job matching score based on the job requirements and background information corresponding to the target job information; calculating the interview performance score based on the interview answers provided by the second agent, the interview performance score being calculated based on one or more of the accuracy, logic and communication skills in the interview answers; and calculating the simulated interview score based on the job matching score and the interview performance score and the corresponding weight values.

[0008] As described above, the process of constructing the question bank includes: obtaining all resume contents related to the target position; constructing a job profile based on one or more of the project category, technology stack, difficulty description and corresponding solutions in each resume content, and the job profile includes: one or more of the job title, job level and job requirements; generating one or more interview questions based on the job profile of the target position and storing them in the question bank, and the interview questions are classified according to the difficulty of the questions.

[0009] According to the method described above, the first intelligent agent adjusts the knowledge points tested and the difficulty of the second interview questions based on the first interview answers, including: determining the knowledge point coverage score of the first interview answer based on the first interview answer and a predefined knowledge point library; determining the quality score of the first interview answer based on the content of the first interview answer; and determining the knowledge points tested and the difficulty of the second interview questions based on the knowledge point coverage score and the quality score of the first interview answer.

[0010] As described above, the method for optimizing the first intelligent agent and / or the second intelligent agent includes: constructing a training data set, the training data set including forward training examples and reverse training examples, the role identity in the forward training examples is consistent with the answer content, and the role identity in the reverse training examples is inconsistent with the answer content; using the training data to perform intensive training on the first intelligent agent and / or the second intelligent agent to improve the accuracy of the content output under the corresponding role.

[0011] As described above, the method for optimizing the first agent and / or the second agent includes: configuring a first identifier for the first agent and / or configuring a second identifier for the second agent; when the content output by the first agent includes the second identifier and / or the content output by the second agent includes the first identifier, commanding the first agent and / or the second agent to stop the current streaming output and reorganize the output content.

[0012] According to another aspect of the present application, an electronic device is proposed, which includes a processor and a memory, wherein a computer program instruction set is stored on the memory, and when the processor executes the computer program instruction set on the memory, the multi-agent simulation interview method based on artificial intelligence as described above is implemented.

[0013] According to another aspect of the present application, a computer-readable storage medium is proposed, wherein a computer program instruction set is stored on the computer-readable storage medium, and when the computer program instruction set is executed by a processor, the multi-agent simulation interview method based on artificial intelligence as described above is implemented.

[0014] In the above method, a real interview process can be simulated through the interactive mode of the interviewer agent asking questions and the job seeker agent answering questions. Through multiple simulations, job seekers can promptly discover their weaknesses in the interview, improve their answering methods, enhance their self-confidence, and increase their interview success rate. In addition, when formulating questions, the interviewer agent combines the applicant's resume information and refers to multiple dimensions (such as job requirements, job level, and question difficulty) to accurately extract interview questions that meet the job requirements from the question bank. Through this method, interview questions not only fit the actual needs of the position, but also ensure the professionalism and authority of the questions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Below, the preferred embodiments of the present invention will be further described in detail with reference to the accompanying drawings, in which:

[0016] Figure 1 This is a flow chart of an artificial intelligence-based multi-agent simulation interview method according to an embodiment of the present application.

[0017] Figure 2 This is a flowchart of a method for constructing a question bank according to an embodiment of the present application.

[0018] Figure 3 yes Figure 1 Flowchart of the method for adjusting the knowledge points tested and the difficulty of the second interview question in step S140.

[0019] Figure 4 This is an interaction diagram of a dual-agent simulated interview based on artificial intelligence according to an embodiment of the present application.

[0020] Figure 5 This is a flowchart of a method for calculating a simulated interview score according to an embodiment of the present application.

[0021] Figure 6 This is a flow chart of a method for optimizing the first agent and / or the second agent according to an embodiment of the present application.

[0022] Figure 7It is a flow chart of a method for optimizing the first agent and / or the second agent according to another embodiment of the present application.

[0023] Figure 8 It is a structural diagram of a simulated interview system according to an embodiment of the present application.

[0024] Figure 9 This is a structural diagram of an artificial intelligence-based multi-agent simulation interview device according to an embodiment of the present application.

[0025] Figure 10 It is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0027] In the detailed description that follows, reference may be made to the various drawings that form part of this application and illustrate specific embodiments of the present application. In the drawings, similar reference numerals describe substantially similar components in different figures. Each specific embodiment of the present application is described below in sufficient detail to enable a person of ordinary skill in the art to implement the technical solutions of the present application. It should be understood that other embodiments may be utilized or that structural, logical, or electrical changes may be made to the embodiments of the present application.

[0028] Artificial Intelligence (AI) is a branch of computer science that studies and develops technologies and systems that can perform tasks that typically require human intelligence. Specifically, AI enables machines to mimic or implement human cognitive functions, including learning, reasoning, problem solving, language understanding, perception, and decision-making.

[0029] An intelligent agent in this application is an entity that can perceive its environment and take actions based on these perceptions to achieve a specific goal. Such an entity can be a software program, a hardware device, or a hybrid human-machine system that can perform tasks autonomously or semi-autonomously without direct human intervention.

[0030] Multi-Agent Systems (MAS) are systems composed of multiple independent and autonomous agents. These agents work together to accomplish a task or goal through collaboration, competition, or interaction. MAS is particularly suitable for solving problems in complex, distributed, and dynamic environments.

[0031] The first agent of this application is a virtual interviewer built based on a large artificial intelligence model. It simulates the real interview process by inputting a series of necessary interview information, such as job requirements, applicant resumes, interview rules and evaluation criteria. The interviewer agent can select appropriate questions to ask based on this information, and conduct real-time evaluation based on the applicant's answers during the interview. It can not only ask questions, but also analyze the applicant's answers through preset evaluation rules to evaluate whether they meet the job requirements and whether they have relevant skills. The functions of the interviewer agent include automated question setting, follow-up questions, and evaluation of the quality and accuracy of answers.

[0032] The second agent of this application is an artificial intelligence system that simulates the role of a job seeker. The candidate agent simulates the interview performance of an actual job seeker according to the set instructions and goals. It can answer questions based on the questions raised by the interviewer agent. The candidate agent not only simply answers questions, but also continuously adjusts its interview strategy, improves its expression and answer quality through interaction with the interviewer agent, thereby improving its interview performance. The self-learning and adaptability of the candidate agent enables it to gradually improve during the simulated interview process, helping job seekers practice answering skills, enhance interview experience, and make effective preparations before the formal interview.

[0033] Figure 1 This is a flow chart of an AI-based multi-agent simulated interview method according to one embodiment of the present application. The method is implemented based on a first agent and a second agent, where the first agent plays the role of the interviewer and the second agent plays the role of the job applicant. The method includes:

[0034] Step S110: The first agent obtains the resume content and target position information of the job seeker;

[0035] In step S120, the first agent extracts a first interview question from a pre-built question bank based on the target position information and sends it to the second agent, where the target position information matches one or more of the job requirements, job level, and question difficulty of the first interview question.

[0036] Step S130: The second agent generates a first interview answer based on the first interview question and the job seeker's resume and sends it to the first agent;

[0037] In step S140, the first agent adjusts the knowledge points and difficulty of the second interview questions based on the answers to the first interview.

[0038] In step S110, the first agent (i.e., the interviewer agent) obtains the applicant's resume and target position information from the applicant. A job applicant's resume typically includes information about their education, work experience, professional skills, and personal strengths, while the target position information includes the job title and level. The first agent uses this information as foundational data to inform subsequent interview questions and assessments.

[0039] In step S120, the first agent selects appropriate interview questions from a pre-built question bank based on the applicant's target position information. The interview questions in the question bank are categorized by one or more of the following: job requirements, job level, and question difficulty. Therefore, the first agent selects questions that are highly relevant to the applicant's position.

[0040] Job requirements describe the knowledge, skills, and experience required of a candidate for the target position. They cover the specific duties and responsibilities of the position, and interview questions should assess the candidate's ability to fulfill those responsibilities. For example, for technical positions like software development and data analysis, interview questions might cover specific programming languages, algorithms, data structures, and system design.

[0041] The job level refers to the position's position in the company's organizational structure and is typically categorized as entry-level, mid-level, and senior-level positions. The job level will affect the difficulty and depth of the interview questions, and the interviewer agent will select appropriate questions based on the job level. For example, interview questions for entry-level positions typically focus on basic knowledge, basic skills, and learning ability. Interview questions for mid-level positions may not only test basic knowledge, but also problem-solving skills, certain work experience, and higher technical or management capabilities. Senior positions require applicants to possess extensive experience, deep expertise, and leadership skills. Interview questions will be more challenging and may involve architectural design, team management, decision-making capabilities, and more.

[0042] Question difficulty refers to the complexity and technical depth of interview questions, which is generally related to the job requirements and the candidate's experience level. Question difficulty can be categorized into different levels, ranging from basic questions to advanced and complex questions. Question difficulty should be designed to ensure it meets the needs of different job levels.

[0043] In step S130, after receiving the interview questions, the second agent (job seeker agent) generates corresponding answers based on the content of the job seeker's resume and the target position information. The second agent automatically generates the first interview answer based on the job seeker's background information (such as academic background, work experience, etc.), as well as the skills and knowledge required for the position. The applicant agent simulates the job seeker's interview answer process through technologies such as natural language processing, and adjusts the level of detail and complexity of the answer according to the requirements of the target position. This interactive mode simulates a real interview scenario, allowing applicants to demonstrate their abilities and experience in a more natural and dynamic environment. At the same time, when answering interview questions based on the job seeker's information, the job seeker agent demonstrates specific professional skills, communication skills, and problem-solving abilities. By watching such a simulated interview live broadcast, job seekers can not only learn from others' interview experience, but also effectively improve their own interview skills.

[0044] In step S140, after the first agent receives the first interview answer, it adjusts the next interview question or ends the mock interview based on the quality and accuracy of the first interview answer. Specifically, the first agent will evaluate the applicant's mastery of certain knowledge points or skills based on the second agent's answer. If the second agent's answer indicates that the applicant has a good grasp of a certain area or skill, the first agent may choose to increase the difficulty or ask more challenging questions; conversely, if the answer indicates that the applicant's answers in certain areas are relatively weak, the first agent may choose to adjust the difficulty or focus of the questions, or ask targeted questions for further investigation to better assess the applicant's ability.

[0045] In this method, a realistic interview process is simulated through an interactive model where the interviewer agent asks questions and the job seeker agent answers them. This approach recreates the dynamic and natural environment that job seekers experience in a real interview, allowing them to demonstrate their abilities and experience in a virtual environment. Unlike traditional static practice, this interactive nature not only helps job seekers improve their interview performance through repeated practice, but also allows them to learn from others' interview experiences and coping strategies, thereby better preparing for the actual interview. Through repeated simulations, job seekers can promptly identify their weaknesses in the interview, improve their responses, boost their confidence, and increase their chances of success in the interview.

[0046] Furthermore, when formulating questions, the Interviewer Agent combines the applicant's resume information with multiple dimensions (such as job requirements, job level, and question difficulty) to accurately extract interview questions from the question bank that meet the job requirements. This method ensures that interview questions not only meet the actual job requirements but also ensures their professionalism and authority. Whether it's in-depth questions for technical positions or decision-making issues for management positions, the Interviewer Agent ensures the quality and rationality of the questions, making the entire interview process more objective and fair, avoiding the biased or arbitrary questions common in traditional interviews.

[0047] Furthermore, unlike the "follow-up questioning mode" of traditional large models, the interviewer agent dynamically adjusts the test points and difficulty of the next question based on the interview answers provided by the job applicant. This dynamic adjustment mechanism can not only evaluate the job applicant's performance in real time, but also flexibly change the interview questions according to the applicant's level, thereby ensuring the accuracy and fairness of the interview. Through this method, the interviewer agent can deeply explore the applicant's abilities based on his or her answers, avoid the unchanging question design, and improve the depth and pertinence of the interview. In addition, dynamic adjustment can effectively control the rhythm of the interview, avoid excessive tension or excessive procrastination during the interview, optimize the efficiency of the overall interview process, and make the interview process more humane. Ultimately, this approach helps to improve the fairness of the interview and ensure the scientific nature and accuracy of the evaluation process.

[0048] Figure 2 This is a flow chart of a method for constructing a question bank according to an embodiment of the present application. Figure 2 As shown, the method includes:

[0049] Step S210, obtaining all resume contents related to the target position;

[0050] Step S220: Build a job profile based on one or more of the project category, technology stack, difficulty description, and corresponding solution in each resume. The job profile includes one or more of the job title, job level, and job requirements.

[0051] Step S230: Generate one or more interview questions based on the job profile of the target position and store them in a question bank. The interview questions are classified according to the difficulty of the questions.

[0052] In step S210, all resume content required for the target position is obtained through the recruitment platform, recruitment system, internal enterprise database, or other resume sources. This resume content may include the applicant's personal information, educational background, work experience, project experience, technical stack, work challenges involved, and solutions. Natural language processing (NLP) technology can be used to pre-process the resume content to remove noise and extract valid information relevant to the target position. For example, all resume content related to "backend development engineer" is obtained from the recruitment platform database.

[0053] In step S220, the project category refers to the type and field of the project in which the job seeker is involved (such as software development, system architecture design, data analysis, etc.). According to the project category, it can be judged whether the job seeker's work experience meets the basic requirements of the target position. The technology stack refers to the technical tools, programming languages, development platforms, etc. mentioned by the job seeker in the resume (such as Java, Python, Docker, Kubernetes, etc.). Based on the technology stack mastered by the job seeker, it can be inferred that the degree of match between his or her technical capabilities and the target position. The description of difficulties and the corresponding solutions are the technical difficulties encountered by the job seeker in his or her work experience, as well as the methods and strategies used to solve these problems. By analyzing the description of difficulties and solutions, the actual problem-solving ability of the job seeker and his or her innovation ability can be evaluated. Combining this information, a job profile of the target position will be automatically constructed.

[0054] For example, first, from all the resume content related to "back-end development engineer", the project categories are extracted, including: e-commerce system development, microservice architecture transformation, and data analysis platform construction; the technology stack includes: programming languages ​​​​: Java, Python, Go, databases: MySQL, MongoDB, Redis, frameworks: Spring Boot, Hibernate; difficulty descriptions include: performance bottleneck problems under high concurrency, and consistency processing of distributed transactions; solutions include: using Redis for cache design and using distributed locks to achieve transaction consistency.

[0055] Then, based on one or more of the project category, technology stack, difficulty description and corresponding solutions, as well as combined with industry standards and job information, the job requirements for "back-end development engineer" are summarized. For example, the job requirements include: in-depth understanding of Java or other back-end programming languages, familiarity with mainstream databases and optimization techniques, and mastery of microservice architecture design and tuning.

[0056] Finally, we construct a job profile. Based on the information extracted and summarized above, the job profile of "Backend Development Engineer" is summarized as shown in Table 1 below:

[0057] Table 1

[0058]

[0059] In step S230, by conducting an in-depth analysis of the job profile generated in step S220, one or more interview questions related to the target position can be generated using large model technology. Interview questions will be classified according to the difficulty of the questions to ensure that the interview questions cover different technical levels and knowledge areas. Furthermore, the automatically generated question bank can also be strictly screened, filtered and manually reviewed to ensure the quality of the questions, and ultimately form a set of interview question banks that are both comprehensive and authoritative. Through the above series of professional operations, it is intended to provide strong support for the interviewer intelligent body and improve the authenticity and authority of the interview questions.

[0060] For example, based on the job profile of "Backend Development Engineer" in Table 1, the following interview questions can be generated:

[0061] 1. Difficulty Level - Low: Please explain the common scenarios for using Redis with MySQL and their advantages and disadvantages.

[0062] 2. Difficulty Level - Advanced: Please write a program to implement a simple distributed lock using Redis.

[0063] According to another embodiment of the present application, when generating interview questions, the job requirements of the target company can also be referenced. Specifically, the recruitment platform can extract job requirements based on the job recruitment information published by the target company. Then, when simulating interviews with the target company, the extracted interview questions correspond to the job requirements, ensuring that the interview questions are closely centered on the actual needs of the target company and screening out candidates who meet the requirements. This also improves the effectiveness of the interview and optimizes the job seeker's experience in the simulated interview.

[0064] Figure 3 yes Figure 1 Flowchart of the method for adjusting the knowledge points and difficulty of the second interview question in step S140. Figure 3 As shown, the method includes:

[0065] Step S141, determining a knowledge point coverage score of the first interview answer based on the first interview answer and a predefined knowledge point library;

[0066] Step S142, determining a quality score of the first interview answer based on the content of the first interview answer;

[0067] Step S143: Determine the knowledge points and difficulty of the second interview questions based on the knowledge point coverage score and quality score of the first interview answers.

[0068] In step S141, a matching rule set is formed based on the predefined knowledge point library, including keyword matching rules and semantic association rules. The first interview answer is parsed, i.e., the keywords, syntactic structure, and semantic information in the first interview answer are extracted. The parsed answer content is matched one by one with the knowledge points in the knowledge point library, and a coverage score is calculated based on the number of matches, association strength, and coverage.

[0069] In step S142, the quality score is determined based on the following factors: answer content completeness, logic, and language expression evaluation. Answer content completeness is used to determine whether the answer fully covers the required content of the interview question; missing content will result in a penalty. Logic is evaluated based on the logical clarity of the answer, including whether the content is well-organized and clearly structured. Language expression evaluation determines the language quality score by analyzing the answer's fluency, accuracy, and professionalism.

[0070] In step S143, if the first interview answer contains knowledge points with low scores, knowledge points that have not been examined or have been examined less are preferentially selected as the focus of the second interview; when the knowledge point coverage score and quality score of the first interview answer are high, the difficulty of the question is appropriately increased; otherwise, the difficulty of the question is reduced.

[0071] This method dynamically adjusts the difficulty of follow-up questions based on interview answers, enabling a more accurate assessment of candidate capabilities while also effectively controlling the pace and depth of the interview, improving the efficiency and fairness of the entire interview process. These refined management and optimization measures not only provide the interviewer agent with a more intelligent, efficient, and user-friendly follow-up tool, but also offer users a fairer and more professional interview experience.

[0072] Figure 4 This is an interactive diagram of a dual-agent simulated interview based on artificial intelligence according to an embodiment of the present application. Figure 4 As shown, the method includes:

[0073] Step S410, obtaining the resume content and target position information of the job seeker;

[0074] Step S420: Extract a first interview question from a pre-built question bank based on the target position information and send it to the second agent, where the target position information matches one or more of the job requirements, job level, and question difficulty of the first interview question.

[0075] Step S430: Generate a first interview answer based on the first interview question and the job seeker's resume and send it to the first agent;

[0076] Step S440: If the interview answer meets the conditions for ending the interview, execute S480;

[0077] Step S450: If the interview answer does not meet the conditions for ending the interview, adjust the knowledge points and difficulty of the next interview question based on the interview answer and send the next interview question to the second agent;

[0078] Step S460, generating an interview answer based on the next interview question and the job seeker's resume;

[0079] Step S470: If the interview answer does not meet the conditions for ending the interview, then repeat S450-S460;

[0080] Step S480: The first agent calculates a simulated interview score based on the resume content and the interview answers provided by the second agent.

[0081] Step S490: Display the conversation process between the first agent and the second agent and the simulated interview score to the job seeker.

[0082] The above steps S410-430 and steps S450-460 are the same as Figure 1 The method is the same as in , so I will not repeat it here.

[0083] In step S440, the conditions for ending the interview can be determined based on multiple factors, such as whether the applicant has answered all pre-set questions; whether the applicant's answers fully demonstrate their core competencies and meet the job requirements; whether the applicant exhibits the same deficiencies across multiple questions, precluding further evaluation; and whether the applicant has met certain scoring criteria during the interview.

[0084] In step S480, when the first agent determines that the conditions for ending the interview are met, it begins calculating the mock interview score. The mock interview score is calculated based on the applicant's performance throughout the interview process. The scoring criteria can be based on pre-set evaluation rules, including dimensions such as the completeness of the answers to questions, clarity of thinking, technical level (if it is a technical position), and communication skills (if it is a management position). The first agent integrates these scores and calculates the applicant's final mock interview score.

[0085] In step S490, the first agent displays the entire simulated interview dialogue process and the job applicant's interview score to the job applicant. The display content usually includes: Interview questions and answers: all questions in the interview process and their corresponding answers are presented to the job applicant to help him review his interview performance. Score details: a detailed display of the interview score, including the score of each question, the total score, and the specific performance of the evaluation indicators (such as knowledge mastery, communication skills, logical thinking, etc.). Interview feedback: If necessary, the first agent can also provide some interview suggestions or improvement directions based on the scoring results to help job applicants improve their interview skills. In this way, job applicants can not only clearly understand their interview performance, but also make targeted improvements based on feedback information, so as to prepare for future actual interviews.

[0086] Figure 5 This is a flow chart of a method for calculating simulated interview scores according to one embodiment of the present application. Figure 3 As shown, the method includes:

[0087] Step S510: obtaining background information of the job seeker from the resume, wherein the background information includes one or more of: education background, work experience, and skills;

[0088] Step S520, calculating a job matching score based on the job requirements and background information corresponding to the target job information;

[0089] Step S530, calculating an interview performance score based on the interview answer provided by the second agent, wherein the interview performance score is calculated based on one or more of answer accuracy, logic, and communication ability in the interview answer;

[0090] Step S540 , calculating a simulated interview score based on the job matching score, the interview performance score, and corresponding weight values.

[0091] In step S510, by obtaining this background information, the first agent can establish a basic data model related to the target position for each job seeker, thereby laying the foundation for subsequent job matching scoring.

[0092] In step S520, the calculation of the job matching score takes into account the degree of fit between the job requirements and the job seeker's background. Specifically, it includes: the matching degree between academic qualifications and job requirements, the matching degree between work experience and job requirements, and the matching degree between skills and job requirements. If the target position has clear requirements for academic qualifications, the job seeker's academic background needs to match the job requirements. For example, some positions may require a bachelor's degree or above, or a degree in a specific field. If the job seeker's academic qualifications meet the requirements, points will be added to the job matching, otherwise points may be deducted. For example, the target position requires a certain number of years of work experience and relevant industry background. If the job seeker's work experience is highly relevant to the job requirements (for example, having worked in similar positions, having experience in the same field, etc.), the job matching score is higher. For example, technical positions require skills in certain programming languages, while management positions may place more emphasis on leadership and teamwork. The higher the skill matching degree, the higher the score.

[0093] In step S530, answer accuracy refers to whether the applicant's answers to the interview questions are accurate and relevant. Answers that comprehensively and accurately address the interviewer's requirements are assigned a higher performance score. Logic evaluates the structure and clarity of the applicant's answers. Logically structured answers often demonstrate a rigorous thinking style and problem-solving abilities. Communication skills are also a key assessment criterion for many positions, particularly in management and customer service roles. During the interview, the applicant's clarity of expression and ability to convey information concisely and effectively will directly impact their score.

[0094] In step S540, the job match score and interview performance score are weighted and averaged according to preset weights. The weightings depend on the specific job requirements and the focus of the interview process. For example, for a technical position, the job match score might be weighted 70%, while the interview performance score might be weighted 30%. For a sales or management position, the interview performance score would carry a greater weight.

[0095] This method calculates a comprehensive and accurate mock interview score by comprehensively considering the candidate's background information, interview performance, and match with the job requirements. This feedback allows you to better understand your performance in different areas and better prepare for the subsequent job search process.

[0096] During multi-agent communication, role confusion may occur. Agent role confusion occurs when agents in a multi-agent system are unable to correctly identify or execute their assigned roles and tasks, leading to unstable system behavior or unattainable goals. For example, an interviewer might ask a question and then answer it in the role of the job applicant. This creates a poor user experience. To address this issue of role confusion during agent role-playing, we can adopt multi-faceted strategies for optimization and adjustment. The specific optimization methods are as follows:

[0097] Figure 6 FIG. 1 is a flow chart of a method for optimizing a first agent and / or a second agent according to an embodiment of the present application. Figure 6 As shown, the method includes:

[0098] Step S610: constructing a training data set, wherein the training data set includes positive training examples and negative training examples, wherein the role identity in the positive training examples is consistent with the answer content, and the role identity in the negative training examples is inconsistent with the answer content;

[0099] Step S620: Utilize the training data to perform intensive training on the first agent and / or the second agent to improve the accuracy of outputting content under the corresponding role.

[0100] In step S610, the positive training examples refer to examples in which the role identity is consistent with the content of the answer given by the role. For example, if a role identity is a "software development engineer", then the content of the answer given by the role should be related to software development. For example, the answer may involve code optimization, system architecture design, technology stack usage, etc. The positive training examples are used to train the intelligent agent to recognize and output content that is consistent with the role identity. The negative training examples refer to examples in which the role identity is inconsistent with the content of the answer given by the role. For example, if a role identity is a "software development engineer", but the content of the answer is related to marketing (for example, discussing how to enhance brand influence), then the answer is a negative training example. The negative training examples are used to help the intelligent agent identify inappropriate answer content and correct it.

[0101] When constructing a training dataset, the system generates a large number of positive and negative training examples through manual or automatic annotation techniques. These examples can be obtained from various sources, such as historical conversation data, question-and-answer datasets, role-playing game data, and corporate documents. The training dataset should include a variety of role identities (such as technical positions, management positions, and customer support positions), as well as the correct and incorrect answers for each role identity.

[0102] In step S620, reinforcement learning (RL) is a machine learning method in which an agent continuously adjusts its strategy by interacting with the environment and based on reward signals. In the present invention, the agent adjusts its strategy for generating answers based on forward and backward training examples.

[0103] During training, the first agent and / or the second agent are trained with forward and reverse examples from the training dataset. Through reinforcement learning algorithms (such as Q-learning, Deep Q Network (DQN), etc.), each agent updates its model parameters based on a given state and action selection strategy, thereby optimizing the accuracy of its output content.

[0104] For example, if the first agent (interviewer agent) answers a question that completely matches the job requirements based on the positive example when answering the interview question, it will receive a positive reward; if the interviewer asks an irrelevant question during the interview, it will receive a negative reward and adjust its strategy through training to reduce such errors.

[0105] After multiple training cycles, the agent's output gradually converges to the criteria for its role. As training progresses, the agent becomes more precise in identifying when to output relevant information and how to generate high-quality responses based on a given role.

[0106] After the training is completed, the system will evaluate the first agent and the second agent to ensure that they can output highly accurate content in actual applications. The evaluation can be carried out through multiple rounds of simulated dialogues, user feedback, or expert scoring. If the evaluation results are not ideal, the system will further adjust the training strategy and continue to conduct intensive training until the agent can accurately output the required content in various roles. Through the above training method, this application can effectively improve the accuracy of the agent's answers in different roles, and can continuously improve itself to prevent role confusion and poor user experience.

[0107] Figure 7 FIG. 1 is a flow chart of a method for optimizing the first agent and / or the second agent according to another embodiment of the present application. Figure 7 As shown, the method includes:

[0108] Step S710, configuring a first identifier for the first agent and / or configuring a second identifier for the second agent;

[0109] Step S720: When the content output by the first agent includes the second identifier and / or the content output by the second agent includes the first identifier, command the first agent and / or the second agent to stop the current streaming output and reorganize the output content.

[0110] In step S710, the first identifier and / or the second identifier can be a string or a numerical identifier. The identifier is embedded in the content output by the first agent and / or the second agent to distinguish the content from different agents. For example, the strings are "Role A" and "Role B".

[0111] In step S720, when the first agent (e.g., the interviewer agent) begins streaming content, it will check in real time whether its output contains the second identifier. If the second identifier appears in the first agent's output, this indicates that the agent may have inadvertently referenced the second agent's content during its output, resulting in an output that does not meet expectations. Similarly, the second agent (e.g., the job seeker agent) will also be monitored in real time to see if its output contains the first identifier.

[0112] Identifier matching can be achieved through string matching, regular expression matching, or parsing of the output content using natural language processing (NLP) techniques. If the judgment condition is met (i.e., the output of the first agent contains the second identifier, or the output of the second agent contains the first identifier), a stop command is sent to the first agent and / or the second agent, pausing the current content generation. This can be achieved through control flow pause, task termination, or output terminator.

[0113] Once the streaming output is stopped, the output content will be reorganized and adjusted. The reorganization methods include:

[0114] (1) Modify the output content based on the role identity and context to ensure that the output of each agent is consistent with its predetermined role. For example, if the "interviewer" mistakenly uses the identifier of the "applicant", the interview content will be regenerated without the "applicant" identifier.

[0115] (2) Through context analysis, ensure that the agent strictly adheres to its identity when outputting content. For example, ensure that the "applicant" only discusses content related to his position, while the "interviewer" focuses on asking questions related to the interview.

[0116] (3) If the original output content is related to the other party’s character identity, the system may need to generate new content to replace it to ensure that the output of each agent is logical and consistent with the character. The regenerated content will try to avoid unnecessary character information or irrelevant dialogue content.

[0117] These optimizations work together to significantly reduce the agent's confusion during role-playing and improve the quality of its responses and user experience during interactions. Through this meticulous tuning, we ensure that the agent maintains character consistency and stability in a variety of complex dialogue scenarios, providing users with a more fluid and natural interaction experience.

[0118] Figure 8 This is a schematic diagram of the structure of a simulated interview system according to an embodiment of the present application. Figure 8 As shown, the simulated interview system includes an interviewer agent 810 and a job seeker agent 820. The simulated interview process of the interviewer agent 810 and the job seeker agent 820 can be presented to the user (job seeker) through text communication or through live interactive communication in the form of virtual people.

[0119] The core task of the interviewer agent 810 is to generate interview questions related to job requirements in real time based on the company's recruitment needs and interview conversation history, and provide real-time feedback and evaluation based on the applicant's answers. It can complete the task through auxiliary tools, including: resume key information extraction and reading API, company recruitment information RAG API and person-job matching score API. Among them, the resume key information extraction and reading API is used to extract core information from the applicant's resume, such as work experience, skills and educational background. The company recruitment information RAG API is used to match the company's recruitment needs with the applicant's resume, and to refine the core inspection points and key issues. The person-job matching score API is used to evaluate the degree of match between the applicant's background and the target position, and provide a reference for the interviewer.

[0120] The Job Seeker Agent 820's primary task is to simulate a job seeker, automatically generating interview responses based on the company's recruitment information and job requirements, and responding to questions posed by the Interviewer Agent. Its responses are based on the company's recruitment JD information, the applicant's resume, and the principles of answering questions, simulating the logic of a real job search scenario to ensure that the answers are consistent with the job requirements.

[0121] During the simulated interview, the interviewer agent 810 first generates interview questions related to the job requirements based on the company's recruitment information, and asks preliminary questions based on the resume information of the job seeker agent.

[0122] After receiving the question, the job seeker agent 820 combines the input company recruitment JD and its own resume information to generate an answer that meets the job requirements.

[0123] Interviewer Agent 810 evaluates responses in real time, analyzing them using knowledge coverage and quality scores. Based on these scores, Interviewer Agent 810 determines the next question to ask. Based on the candidate agent's responses, Interviewer Agent 810 dynamically adjusts the focus and difficulty of questions to achieve a targeted assessment.

[0124] The two agents continuously engage in question-and-answer sessions and provide feedback, simulating a real-world interview scenario until the interview process is complete. The interviewer agent will comprehensively evaluate the interview process and provide a comprehensive evaluation. This evaluation consists of two parts: an objective job-matching score, derived by calling our job-matching API, based on the candidate's background and the fit between the job requirements; and a comprehensive evaluation of the candidate's actual performance during the interview, including their understanding of the questions, the accuracy and logic of their responses, and their communication skills. This evaluation mechanism not only ensures comprehensiveness and scientificity, but also provides strong support for companies' hiring decisions.

[0125] In the above method, by providing information guidance in various aspects to the interviewer agent and the applicant agent, they can simulate the entire interview process more realistically, allowing the applicant to improve their interview skills and competitiveness by observing such an interview.

[0126] Corresponding to the method embodiment of the present application, the present application also provides a multi-agent simulation interview device based on artificial intelligence, such as Figure 9 As shown, the multi-agent simulated interview device 1000 includes: a first agent 100 and a second agent 200.

[0127] The acquisition module 110 of the first agent is used to obtain the resume content and target position information of the job seeker;

[0128] The question extraction module 120 of the first agent is used to extract a first interview question from a pre-built question bank based on the target position information and send it to the second agent, wherein the target position information matches one or more dimensions of the job requirements, job level, and question difficulty of the first interview question;

[0129] The receiving module 210 of the second agent receives the first interview question, and the answer generating module 220 generates a first interview answer based on the first interview question and the resume of the job seeker and sends it to the first agent;

[0130] The question adjustment module 130 of the first agent adjusts the knowledge points and difficulty of the second interview questions based on the answers to the first interview.

[0131] Figure 10This is a schematic diagram of the hardware structure principle of an electronic device according to an embodiment of the present application. The electronic device can be implemented as a server or various other terminal devices, such as a desktop personal computer, a tablet computer, a laptop computer, a mobile phone, etc., which includes a processor 601 and a memory 602. A program instruction set is stored on the memory 602. When the processor 601 executes the program instruction set on the memory 602, any of the aforementioned message distribution methods based on message attributes is implemented.

[0132] Specifically, the processor 601 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiment of the present invention.

[0133] The memory 602 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 602 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 602 may include removable or non-removable (or fixed) media. Where appropriate, the memory 602 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, the memory 602 is a non-volatile solid-state memory.

[0134] The memory may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the artificial intelligence-based multi-agent simulation interview method provided by the present invention.

[0135] In one example, the electronic device may further include a communication interface 603 and a bus 604. The processor 601, memory 602, and communication interface 603 are connected via bus 604 and communicate with each other. The communication interface 603 is primarily used to implement communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The bus 604 includes hardware, software, or both, and couples the components of the online data traffic metering device to each other. By way of example, and not limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industrial Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industrial Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, the bus 604 may include one or more buses. Although embodiments of the present invention describe and illustrate a particular bus, the present invention contemplates any suitable bus or interconnect.

[0136] The present invention also provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement any of the multi-agent simulation interview methods based on artificial intelligence in the aforementioned embodiments. The computer-readable storage medium can be any medium that can tangibly contain or store computer-executable instructions for use by or in combination with an instruction execution system, device, and apparatus. The storage medium can be a transient computer-readable storage medium or a non-transient computer-readable storage medium. Non-transient computer-readable storage media may include, but are not limited to, magnetic storage devices, optical storage devices, and / or semiconductor storage devices. Examples of such storage devices include, for example, magnetic disks, optical disks based on CD, DVD, or Blu-ray technology, and persistent solid-state memories such as flash memory, solid-state drives, and the like.

[0137] The present invention also provides a computer program product comprising a set of computer program instructions that, when executed by a processor, implement any of the artificial intelligence-based multi-agent simulated interview methods described in the aforementioned embodiments. The computer program product may include, but is not limited to, an application installation package published on a website or in an app store, an application plug-in, or a mini-program that can be run within certain applications.

[0138] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0139] The above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. Ordinary technicians in the relevant technical field can make various changes and modifications without departing from the scope of the present invention. Therefore, all equivalent technical solutions should also fall within the scope of the present invention.

Claims

1. A multi-agent simulation interview method based on artificial intelligence, characterized in that: The method is implemented based on a first agent and a second agent, wherein the first agent is used to play the role of an interviewer and the second agent is used to play the role of a job seeker. The method includes: The first agent obtains the resume content and target position information of the job seeker; The first agent extracts a first interview question from a pre-built question bank according to the target position information and sends the first interview question to the second agent, wherein the target position information matches one or more dimensions of the position requirement, position level, and question difficulty of the first interview question; The second agent generates a first interview answer based on the first interview question and the resume content of the job seeker and sends it to the first agent; The first agent adjusts the knowledge points and difficulty of the second interview questions based on the answers to the first interview and sends them to the second agent.

2. The method according to claim 1, characterized in that: Further including: The first agent calculates the mock interview score based on the resume content and the interview answers provided by the second agent; The conversation process between the first agent and the second agent and the mock interview score are presented to the job applicant.

3. The method according to claim 2, characterized in that The first agent calculates the simulated interview score based on the resume content and the interview answers provided by the second agent, including: Obtaining background information of the job seeker from the resume, the background information including: one or more of education, work experience and skills; Calculate the job matching score based on the job requirements and background information corresponding to the target job information; Calculating an interview performance score based on the interview answer provided by the second agent, wherein the interview performance score is calculated based on one or more of answer accuracy, logic, and communication ability in the interview answer; The mock interview score is calculated based on the job match score, interview performance score and corresponding weight values.

4. The method according to claim 1, characterized in that: The process of constructing the question bank includes: Obtain all resume contents related to the target position; Constructing a job profile based on one or more of the project category, technology stack, difficulty description, and corresponding solution in each resume, wherein the job profile includes one or more of: job title, job level, and job requirements; One or more interview questions are generated according to the job profile of the target position and stored in a question bank, and the interview questions are classified according to the difficulty of the questions.

5. The method according to claim 1, characterized in that The first agent adjusts the knowledge points and difficulty of the second interview questions based on the answers to the first interview, including: Determine the knowledge point coverage score of the first interview answer based on the first interview answer and a predefined knowledge point library; Determine the quality score of the first interview answers based on the content of the first interview answers; The knowledge points and difficulty of the second interview questions will be determined based on the knowledge point coverage score and quality score of the first interview answers.

6. The method according to claim 1, characterized in that The method for optimizing the first agent and / or the second agent includes: Constructing a training data set, wherein the training data set includes positive training examples and negative training examples, wherein the role identity in the positive training examples is consistent with the answer content, and the role identity in the negative training examples is inconsistent with the answer content; The training data is used to perform reinforcement training on the first agent and / or the second agent to improve the accuracy of output content under the corresponding role.

7. The method according to claim 1, characterized in that The method for optimizing the first agent and / or the second agent includes: Configuring a first identifier for the first agent and / or configuring a second identifier for the second agent; When the content output by the first agent includes the second identifier and / or the content output by the second agent includes the first identifier, the first agent and / or the second agent is commanded to stop the current streaming output and reorganize the output content.

8. A multi-agent simulated interview device based on artificial intelligence, characterized in that: The device is implemented based on a first agent and a second agent, wherein the first agent is used to play the role of an interviewer and the second agent is used to play the role of a job seeker. The device includes: The acquisition module of the first agent is used to obtain the resume content and target position information of the job seeker; The question extraction module of the first agent is used to extract the first interview question from the pre-built question bank according to the target position information and send it to the second agent, wherein the target position information is consistent with one or more dimensions of the position requirements, position level and question difficulty of the first interview question; The answer generation module of the second agent is used to generate a first interview answer based on the first interview question and the resume content of the job seeker and send it to the first agent; The question adjustment module of the first intelligent agent adjusts the knowledge points and question difficulty of the second interview question according to the first interview answer.

9. An electronic device, characterized in that: It includes a processor and a memory, wherein a computer program instruction set is stored in the memory, and when the processor executes the computer program instruction set in the memory, the artificial intelligence-based multi-agent simulation interview method described in any one of claims 1-7 is implemented.

10. A computer-readable storage medium, characterized in that: in, The computer-readable storage medium stores a computer program instruction set, and when the computer program instruction set is executed by the processor, it implements the artificial intelligence-based multi-agent simulation interview method described in any one of claims 1-7.

Images