Dynamic interview test question generation method, system and device based on real-time dialogue analysis and adaptive RAG and medium

Abstract

The invention provides a dynamic interview test question generation method, system and device based on real-time dialogue analysis and adaptive RAG and a medium, and the method comprises the steps: firstly creating a competency knowledge graph according to recruitment post information, and initializing a state machine; then monitoring interview dialogue streams, analyzing answers of candidates by using an AI large model, and performing evidence intensity and logic consistency verification based on dialogues and resumes of the candidates; updating the state of the corresponding node by using a state machine based on the result of the AI large model analysis, determining the investigation intention of the next step, and generating a plurality of alternative interview questions according to the investigation intention; continuously monitoring the interview conversation stream until the interview is finished; and finally, generating a structured evaluation report based on the competency node information of the interview state machine. According to the method, the conversation content is monitored in real time, the state machine and the AI large model are utilized, the answers of candidates are dynamically analyzed in the conversation process, the questions are generated, interview is converted into accurate strike, and the coverage rate of core ability points is greatly increased.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, algorithms, and natural language processing, specifically to a method, system, device, and medium for generating dynamic interview questions based on real-time dialogue analysis and adaptive RAG. Background Technology

[0002] Current recruitment interview question matching technologies generally adopt a series architecture of "static question bank + rule template + single coarse-grained RAG (retrieval enhancement to generate query)": experts first pre-set a large number of questions and manually label them, then filter them by keyword or Boolean rules according to the overlap between the job description and the resume, and then perform a vector retrieval on the entire text before handing it over to a large model to generate questions all at once; it cannot adapt to changes in job positions and candidates, resulting in problems such as low question matching degree, poor credibility and the inability of the system to self-evolve.

[0003] Existing technologies, adhering to the outdated paradigm of "static question bank + rule filtering + single coarse-grained RAG," expose several persistent problems: First, cognitive overload: Interviewers must simultaneously handle multiple tasks such as listening, understanding, evaluating, questioning, and recording, making it impossible to guarantee the depth and quality of each step and leading to over-reliance on personal experience. Second, risk of deviating from the interview path: Interviewers are easily attracted by minor details in candidates' answers, leading to excessive follow-up questions, deviating from the core assessment objectives, unbalanced time allocation, and incomplete coverage of key competencies. Third, insufficient depth and precision of follow-up questions: Inexperienced interviewers struggle to identify logical flaws, weaknesses in experience, or "theoretical talk" in candidates' answers, failing to conduct in-depth, progressive probing like seasoned experts, thus missing crucial assessment signals. Fourth, difficulty in controlling unconscious bias: Interviewers' questions may be influenced by personal preferences or first impressions (confirmation bias), resulting in inconsistent assessment standards for different candidates and a lack of fairness in the evaluation. Fifth, existing technologies cannot address dynamic challenges: neither static question banks nor simple RAG generators can provide real-time, adaptive guidance during the interview process. They provide "ammunition," not a "real-time fire control system." The market urgently needs a technological solution that can provide interviewers with real-time, intelligent, and adaptive navigation during the interview process. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG, comprising: Based on the job posting, obtain the competency knowledge graph that has been pre-created for the job posting, and initialize the state machine based on the competency knowledge graph; Based on the intercepted interview conversations and candidate resumes, we use a large AI model to analyze the candidates' answers and verify the strength of evidence and logical consistency. Based on the results of AI large model analysis, the state machine is used to update the state of the corresponding node, determine the next assessment intention, and generate several alternative interview questions based on the assessment intention; the interview dialogue stream is monitored until the interview ends. A structured evaluation report is generated based on the competency node information of the interview state machine.

[0005] Preferably, the competency knowledge graph pre-created for the job includes: Based on the job description text, natural language processing is used to extract the various abilities required for the job, and each ability is taken as a competency node; the competency node includes: node name, ability name, weight, and related knowledge base scope; Edge relationship construction: Analyze the logical dependencies and business relationships between various capabilities, and construct edges; A competency knowledge graph is constructed based on the competency nodes and edges.

[0006] Preferably, the initialization state machine includes: A state machine is created based on a competency knowledge graph, and the state machine contains multiple competency nodes corresponding to the job position. For each competency node of the state machine, initialize the following information: coverage state, evaluation confidence, and evidence strength; And set global parameters for the state machine, including: target interview duration, core node priority, and bias intervention threshold.

[0007] Preferably, the step of analyzing the candidate's answers using an AI model based on the monitored interview dialogue stream and the candidate's resume, and performing evidence strength and logical consistency verification, includes: Streaming speech recognition technology is used to convert candidate speech into text streams in real time; Natural speech processing technology is used to extract keywords from the text stream. These keywords include: technical terms, project names, and quantitative indicators. The strength of evidence is determined by deep semantic analysis based on the keywords. The current answer in the interview dialogue is cross-referenced with the candidate's resume and the candidate's past statements in this interview to achieve logical consistency verification. Preferably, the step of obtaining the evidence strength based on the keyword through deep semantic analysis includes: Responses that only mention the definition of a concept will be evaluated as weak evidence. Responses containing personal opinions but lacking supporting examples are assessed as moderate evidence; Responses that provide specific projects, roles, or quantifiable results are evaluated as strong evidence.

[0008] Preferably, the results of the AI-based large model analysis are used to update the state of the corresponding node using a state machine and to determine the next step of the investigation, including: Using a state machine, the state of the corresponding competency node is updated based on the results of evidence strength and logical consistency verification. The state machine is then used to adaptively reconstruct the retrieval and enhance query generation. Preferably, updating the state of the corresponding competence node based on the evidence strength and logical consistency verification results includes: When a candidate's answer is identified as providing strong evidence and being logically clear, the assessment confidence level of the corresponding competency node is raised to "strong". When a candidate's answer is identified as merely a theoretical explanation, the evaluation confidence of the corresponding node is marked as "weak" and a "needs further investigation" flag is added. When a candidate's current answer is found to contradict their resume or past answers, the corresponding node will be marked as "questionable".

[0009] Preferably, the adaptive reconstruction of the retrieval enhancement query generation includes: The current dialogue context, the node information to be examined in the state machine, and the keywords in the candidate's answer are dynamically combined to generate contextualized instructions for a specific examination target. The instructions include: query instructions for searching the enterprise knowledge base and instructions for generating follow-up question options; Preferably, the step of generating several candidate interview questions based on the assessment intent includes: The instructions generated from the aforementioned follow-up questions and multi-source evidence retrieved from the enterprise knowledge base are used as inputs; Utilize AI multimodal large models to generate follow-up questions and accompanying information for different tactical paths; The accompanying information includes: each follow-up question is accompanied by the following information: tactical description, recommendation rating, and estimated response time; The tactical approach includes: in-depth analysis, horizontal connections, and challenge reflection; The multi-source evidence includes: knowledge base fragments, historical excellent follow-up questions, and the original text of the candidates' answers.

[0010] Preferably, after generating several candidate interview questions based on the assessment intent, the method further includes: The system provides a human-computer coordination interface, allowing interviewers to view follow-up question options and accompanying information displayed by the large model during the interview process, as well as the status of competency nodes in the competency graph, historical question-and-answer summaries, and confidence change curves. The interface also provides real-time reminders to interviewers to correct biases, including: bias detection, warnings of discriminatory topics, and suggestions for balanced time allocation. Furthermore, the interface enables star rating, and based on the star rating, it adjusts the weights of corresponding competency nodes in the competency graph and enhances search preferences. The big data model uses each choice made by the interviewer as implicit feedback to fine-tune the system's recommendation strategy. Preferably, the generation of a structured evaluation report based on the competency node information of the interview state machine includes: Based on the competency assessment map, the human-computer coordination interface is used to display the final confidence level, key evidence fragments, and contradiction points of each node. Use a human-computer interaction interface to review the interview process; It also showcases the candidate's highlights and risks, generating recommendations.

[0011] Based on the same inventive concept, this invention also provides a dynamic interview question generation system based on real-time dialogue analysis and adaptive RAG, comprising: An initialization module is used to obtain a competency knowledge graph pre-created for the job posting and to initialize a state machine based on the competency knowledge graph. The analysis module is used to analyze candidates' answers using a large AI model based on the monitored interview dialogue stream and candidate resumes, and to verify the strength of evidence and logical consistency. The intelligent decision-making module is used to update the state of the corresponding node using a state machine based on the results of AI large model analysis, determine the next assessment intention, and generate several alternative interview questions based on the assessment intention; it continues to monitor the interview dialogue flow until the interview ends. The assessment report module is used to generate a structured assessment report based on the competency node information of the interview state machine.

[0012] Based on the same inventive concept, the present invention also provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG provided by the present invention is implemented.

[0013] Based on the same inventive concept, the present invention also provides a readable storage medium having an executable program stored thereon, wherein when the executable program is executed, it implements a dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG provided by the present invention.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG, comprising: obtaining a competency knowledge graph pre-created for the job posting, and initializing a state machine based on the competency knowledge graph; analyzing the candidate's answers using an AI big data model based on the monitored interview dialogue stream and the candidate's resume, and verifying the strength of evidence and logical consistency; updating the state of the corresponding nodes using the state machine based on the results of the AI ​​big data model analysis, determining the next assessment intention, and generating several candidate interview questions based on the assessment intention; continuing to monitor the interview dialogue stream until the interview ends; generating a structured evaluation report based on the competency node information of the interview state machine; this invention, by monitoring the dialogue content in real time and utilizing a state machine and an AI big data model, dynamically analyzes the candidate's answers and generates follow-up questions during the dialogue, transforming the interview into a "precision strike," significantly improving the coverage of core competencies. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG according to the present invention. Figure 2 This is a diagram of the overall system architecture of Example 1; Figure 3 This is a block diagram of the adaptive RAG system in Example 1; Figure 4 This is a schematic diagram of the structure of a dynamic interview question generation system based on real-time dialogue analysis and adaptive RAG according to the present invention. Figure 5 This is a schematic diagram of an electronic device structure according to the present invention. Detailed Implementation

[0016] This invention systematically eradicates the aforementioned defects through event-driven spatiotemporal dynamic knowledge graphs, cross-granularity gated retrieval, interpretable triple tracing, and explicit star rating closed loops. It achieves minute-level self-adaptation, precise recall, full traceability, and continuous self-evolution, thereby completing the leap from a "static tool" to a "living system" in terms of technological generation.

[0017] This invention discloses a dynamic interview question method based on real-time dialogue analysis and adaptive RAG (Real-Time Analysis and Evaluation). By monitoring the dialogue flow between the interviewer and candidate in real time, it utilizes AI multimodal large-scale model discourse analysis technology to deeply analyze the semantic structure, strength of evidence, and logical consistency of the candidate's answers. It constructs an interview state machine to dynamically track the coverage status and assessment confidence of the competency knowledge graph. Based on the current dialogue context rather than static input, it adaptively reconstructs RAG queries to accurately retrieve relevant evidence from the company's private knowledge base. The large-scale model generates multi-path tactical follow-up question options (deep mining / lateral association / challenge reflection), presented to the interviewer through a non-intrusive human-computer interaction interface, and proactively intervenes in potential biases. After the interview, the system generates a structured evaluation report and supports closed-loop self-evolution based on explicit feedback from the interviewer.

[0018] To better understand this invention, the following description, in conjunction with the accompanying drawings and examples, will further illustrate the invention.

[0019] Example 1: like Figure 1 As shown, this invention discloses a dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG, comprising: S1. Obtain a competency knowledge graph pre-created for the job posting based on the job posting, and initialize the state machine based on the competency knowledge graph; S2. Based on the intercepted interview dialogue stream and candidate resumes, use AI big data models to analyze the candidate's answers and verify the strength of evidence and logical consistency. S3. Based on the results of the AI ​​large model analysis, update the state of the corresponding node using a state machine, determine the next assessment intention, and generate several alternative interview questions based on the assessment intention; continue to monitor the interview dialogue flow until the interview ends; S4. Generate a structured evaluation report based on the competency node information of the interview state machine.

[0020] The following is combined with Figure 2 The present invention will be further described.

[0021] The specific step S1 requires first determining the job posting information and obtaining candidate resumes, and then performing initialization work.

[0022] S1.1 Input Processing Job Analysis: Obtain the job description text and identify the hard skills, soft skills, and corporate culture and values ​​requirements.

[0023] Candidate resume analysis: Extract structured information such as educational background, project experience, skill set, and past achievements.

[0024] Enterprise knowledge base integration: Loading private data such as technical documents, project review reports, failure case libraries, and product strategy documents.

[0025] S1.2 Competency Knowledge Graph Construction Node Extraction: Core assessment points (such as "distributed system design ability," "user growth mindset," and "stress resistance") are extracted from job descriptions using natural language processing technology. Each node includes: node name, ability name, weight, and related knowledge base scope. Edge relationship construction: Analyze the logical dependencies and business connections between capabilities, construct directed / undirected edges, and form a computable knowledge graph.

[0026] S1.3 Initialization of the Interview State Machine Create a state machine instance and initialize the state for each competency node: coverage state (uncovered / covered / overcovered), confidence level assessment (unknown / weak / medium / strong), and evidence strength (theoretical / opinion / quantitative example). Set global parameters: target interview duration, core node priority, and bias intervention threshold.

[0027] Further, in step S2, based on the monitored interview conversation stream and the candidate's resume, an AI large-scale model is used to analyze the candidate's answers and perform evidence strength and logical consistency verification; specifically including: S2.1 Real-time speech transcription / input from other interview devices Streaming speech recognition technology is used to convert candidate speech into a text stream in real time.

[0028] Supports speech feature analysis: speech rate, pauses, and frequency of hesitant words (to help assess confidence).

[0029] It can also collect input information from other interview devices, such as video information from WeChat, Tencent, etc., audio and video information collected by audio and video capture devices in remote conference rooms, interview systems set up on smart devices such as computers or mobile phones at the interview site, and text information, as well as any information related to the interview.

[0030] S2.2 Deep Semantic and Structural Analysis (Core Innovation) Entity and Methodology Recognition: Utilizing natural language processing techniques to extract technical terms, project names, and quantitative indicators, and to detect structured expressions.

[0031] Assessment of the strength of evidence: Theoretical explanation: Only the conceptual definition is mentioned (weak evidence) Opinion expressed: Personal viewpoint but no concrete examples (moderate evidence) Quantitative examples: providing specific projects, roles, and measurable results (strong evidence). Logical consistency detection: Cross-compare the current answer with the resume and the history of the current conversation to detect contradictions. Knowledge tracing and association: Associate keywords in the answer with the enterprise knowledge base to identify knowledge gaps.

[0032] Further step S3 involves updating the state of the corresponding node using a state machine based on the results of AI large-scale model analysis, determining the next assessment intent, and generating several candidate interview questions based on the assessment intent; continuing to monitor the interview dialogue flow until the interview ends; specifically involving interview state updates and adaptive RAG query generation, multi-path tactical follow-up question generation, human-machine collaborative decision-making during the interview process, and closed-loop self-evolution, which will be discussed below in conjunction with... Figure 3 The adaptive RAG system block diagram provides a detailed description of the present invention.

[0033] S3.1 Interview Status Update and Adaptive RAG Query Generation S3.1.1 Dynamic Update of State Machine Based on the S2 parsing results, update the node status of the competency knowledge graph: If a candidate provides strong evidence and the logic is clear, set the confidence level of the corresponding node to "strong"; If there is only theoretical explanation without examples, set the confidence level to "weak" and mark it as "needs further investigation"; If a contradiction is detected, the node is marked as "questionable" and a challenging follow-up questioning strategy is triggered.

[0034] The state machine provided by this invention can track the coverage and evaluation confidence of the competency knowledge graph in real time, transforming discrete question-and-answer dialogues into a computable, goal-oriented state transition process, thereby achieving "navigation-level" control of the interview process.

[0035] S3.1.2 Adaptive RAG Query Reconstruction Query commands dynamically generate contextualized, task-oriented complex queries for the RAG knowledge base.

[0036] Adaptive RAG query reconstruction can dynamically generate task-oriented and contextualized complex queries based on the current dialogue context, evidence strength, and confidence gap, achieving precise linkage between "asking what - retrieving what - generating what," breaking through the static retrieval paradigm of traditional RAG.

[0037] S3.2: Multi-path tactical questioning generation module S3.2.1 AI Multimodal Large Model Generation Strategy The queries generated by S3 and the multi-source evidence retrieved (knowledge base fragments, historical excellent follow-up questions, and candidate's original answers) are injected into the AI ​​multimodal big model. The instructions require the simultaneous generation of a large multimodal AI model, offering follow-up question options through different tactical paths such as in-depth exploration, lateral association, and challenge-reflection. Therefore, this invention employs a multi-path tactical question generation approach, rather than outputting a single question. Instead, it generates differentiated follow-up question options based on three tactical strategies: in-depth exploration, lateral association, and challenge-reflection, taking into account the assessment of depth, breadth, and critical thinking.

[0038] S3.2.2 Output Formatting The large model outputs a formatted record of each follow-up question, including tactical descriptions (the objective being examined), a recommendation score (based on evidence matching), and an estimated response time.

[0039] S3.3: Human-Machine Collaborative Decision-Making in the Interview Process S3.3.1 Non-intrusive display During the interview process, the interviewer can view the follow-up questions and options of the interview questions displayed in the large model, see the real-time display of the competency knowledge graph visualization (uncovered / partially covered / fully covered) status, and view the human-computer coordination interface such as historical question and answer summaries and confidence change curves.

[0040] S3.3.2 Integration of Human Decision-Making Interviewers may accept suggestions from the system and ask questions directly, or edit and follow up with questions to adjust the wording, or ignore the suggestions and ask questions independently.

[0041] The system uses each choice made by the interviewer as an implicit feedback system to fine-tune and optimize the system's recommendation strategy.

[0042] The AI ​​tactical collaboration mechanism of this invention provides interviewers with expert-level follow-up questions, enabling even inexperienced interviewers to conduct in-depth investigations and improving the consistency of assessments among interviewers.

[0043] S3.3.3 Real-time bias intervention Confirmation bias detection: If the same node is asked more than 3 times after obtaining a "strong" confidence score, the interface will prompt "This capability has been fully evaluated, and it is recommended to explore other dimensions".

[0044] Warning against discriminatory topics: If the question text contains sensitive words related to gender / region / age, the system will highlight the warning and suggest modifications.

[0045] Time allocation balance: If there is insufficient remaining time and core nodes are not covered, the system will push a "fast coverage" strategy suggestion.

[0046] This invention uses a real-time proactive bias intervention mechanism to detect and confirm biases and discriminatory topics during the interview process, thereby reducing biases at the source and lowering the fairness index (standard deviation of evaluation).

[0047] S3.4: Closed-loop self-evolution S3.4.1 Explicit Rating Collection After each round of follow-up questions, the interviewer can give a star rating for the effectiveness of the system device.

[0048] After the interview, the interviewer calibrates and scores the overall competence assessment to ensure accuracy.

[0049] S3.4.2 Dynamic Weight Adjustment Based on the interviewer's scores, the system adjusts the graph node weights and RAG retrieval preferences: Node weight update formula: w ← w × (1 + α × s_avg) Where α=0.15 is the learning rate, and s_avg is the problem effectiveness score.

[0050] High-frequency effective nodes receive weight boosts, while low-weight nodes trigger half-life decay (T). =30 days), automatically "forgetting" outdated knowledge. This invention introduces a node half-life decay mechanism to automatically eliminate outdated knowledge, keeping the knowledge graph lightweight and efficient, thereby improving retrieval speed and reducing storage costs, solving the problem of bloated knowledge graphs and degraded retrieval caused by the continuous growth of traditional knowledge bases.

[0051] This invention constructs an event-driven, minute-level updated dynamic knowledge graph that can perceive the dialogue status in real time and adaptively adjust to ensure that the generated questions are accurately aligned with the current interview context, thereby improving the question matching accuracy.

[0052] This invention innovatively adopts a cross-granularity four-level retrieval architecture, which progresses from entities, fragments, chapters to scenarios in a progressive manner. It combines hierarchical gating attention mechanism for intelligent fusion, which significantly improves the accuracy of evidence recall, increases the recall accuracy rate, and reduces the illusion rate of large models.

[0053] S3.4.3 Fine-tuning trigger system real-time evolution When the accumulated explicit scores reach a pre-set threshold, the model fine-tuning device is automatically triggered, updating the parameters of the AI ​​multimodal large model and enabling the system to evolve in real time. This invention combines explicit scoring with the DPO fine-tuning algorithm to form an auditable and quantifiable closed-loop learning system, thereby improving the effectiveness of problem solving.

[0054] Furthermore, step S4 involves generating a structured evaluation report based on the competency node information of the interview state machine, including the output of a structured interview report, specifically including: Competency assessment map: Visualizes the final confidence level of each node, key evidence fragments, and contradiction markers.

[0055] Interview path review: The time sequence of the "question-answer-follow-up question" decision tree is displayed, supporting replay and strategy optimization analysis.

[0056] Candidate profile summary: Automatically extract highlights and risk points to generate recommendations.

[0057] By using the interviewer's choice as implicit feedback and the star rating as explicit feedback, and through the DPO algorithm and dynamic weight adjustment of the AI ​​multimodal large model device, the system can achieve continuous self-evolution and become more intelligent with use. Example 2: like Figure 4 As shown, in order to implement the dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG in the above embodiment, the present invention also provides a dynamic interview question generation system based on real-time dialogue analysis and adaptive RAG, including: initialization module 1, analysis module 2, intelligent decision-making module 3 and evaluation report module 4.

[0058] The system comprises the following modules: Initialization Module 1, which acquires a competency knowledge graph pre-created for the job posting and initializes a state machine based on the competency knowledge graph; Analysis Module 2, which analyzes the candidate's answers using an AI model based on the monitored interview dialogue and the candidate's resume, and performs evidence strength and logical consistency verification; Intelligent Decision Module 3, which updates the state of the corresponding node using the state machine based on the results of the AI ​​model analysis, determines the next assessment intention, and generates several alternative interview questions based on the assessment intention; and continues to monitor the interview dialogue until the interview ends; and Evaluation Report Module 4, which generates a structured evaluation report based on the competency node information of the interview state machine.

[0059] Furthermore, initialization module 1 includes a competency knowledge graph creation unit, a state machine initialization unit, and... The competency knowledge graph creation unit is used to extract various capabilities required for a job based on the job description text using natural language processing, with each capability serving as a competency node; the competency node includes: node name, capability name, weight, and associated knowledge base scope. Edge relationship construction: Analyze the logical dependencies and business relationships between various capabilities, and construct edges; A competency knowledge graph is constructed based on the competency nodes and edges.

[0060] A state machine initialization unit is used to create a state machine based on a competency knowledge graph, wherein the state machine contains multiple competency nodes corresponding to the job position. For each competency node of the state machine, initialize the following information: coverage state, evaluation confidence, and evidence strength; And set global parameters for the state machine, including: target interview duration, core node priority, and bias intervention threshold.

[0061] Further analysis module 2 includes: a streaming speech recognition unit, a keyword extraction unit, a semantic parsing unit, and a verification unit. Specifically... The streaming speech recognition unit is used to convert candidate speech into a text stream in real time using streaming speech recognition technology; The keyword extraction unit is used to extract keywords from the text stream using natural speech processing technology. The keywords include: technical terms, project names, and quantitative indicators. A semantic parsing unit is used to perform deep semantic parsing based on the keywords to obtain the strength of evidence. Specifically, answers that only mention conceptual definitions will be assessed as weak evidence; answers that contain personal insights but lack supporting examples will be assessed as moderate evidence; and answers that provide specific projects, roles, or quantifiable results will be assessed as strong evidence.

[0062] The verification unit is used to cross-compare the current answer in the current interview dialogue with the candidate's resume and the historical expressions in this interview to achieve logical consistency verification.

[0063] Furthermore, the intelligent decision-making module 3 includes: an update unit and a generation unit. The update unit is used to update the state of the corresponding competency node based on the evidence strength and logical consistency verification results using the state machine. A generation unit is used to adaptively reconstruct the retrieval enhancement and generate queries using the state machine.

[0064] Preferably, the method of updating the state of the corresponding node using a state machine based on the results of AI large model analysis includes: When a candidate's answer is identified as providing strong evidence and being logically clear, the assessment confidence level of the corresponding competency node is raised to "strong". When a candidate's answer is identified as merely a theoretical explanation, the evaluation confidence of the corresponding node is marked as "weak" and a "needs further investigation" flag is added. When a candidate's current answer is found to contradict their resume or past answers, the corresponding node will be marked as "questionable".

[0065] Preferably, the adaptive reconstruction of the retrieval enhancement query generation includes: The current dialogue context, the node information to be examined in the state machine, and the keywords in the candidate's answer are dynamically combined to generate contextualized instructions for a specific examination target. The instructions include: query instructions for searching the enterprise knowledge base and instructions for generating follow-up question options; Preferably, the step of generating several candidate interview questions based on the assessment intent includes: The instructions generated from the aforementioned follow-up questions and multi-source evidence retrieved from the enterprise knowledge base are used as inputs; Utilize AI multimodal large models to generate follow-up questions and accompanying information for different tactical paths; The accompanying information includes: each follow-up question is accompanied by the following information: tactical description, recommendation rating, and estimated response time; The tactical approach includes: in-depth analysis, horizontal connections, and challenge reflection; The multi-source evidence includes: knowledge base fragments, historical excellent follow-up questions, and the original text of the candidates' answers.

[0066] Preferably, after generating several candidate interview questions based on the assessment intent, the method further includes: The system provides a human-computer coordination interface, allowing interviewers to view follow-up question options and accompanying information displayed by the large model during the interview process, as well as the status of competency nodes in the competency graph, historical question-and-answer summaries, and confidence change curves. The interface also provides real-time reminders to interviewers to correct biases, including: bias detection, warnings of discriminatory topics, and suggestions for balanced time allocation. Furthermore, the interface enables star rating, and based on the star rating, it adjusts the weights of corresponding competency nodes in the competency graph and enhances search preferences. The big data model uses each choice made by the interviewer as implicit feedback to fine-tune the system's recommendation strategy. Preferably, the generation of a structured evaluation report based on the competency node information of the interview state machine includes: Based on the competency assessment map, the human-computer coordination interface is used to display the final confidence level, key evidence fragments, and contradiction points of each node. Use a human-computer interaction interface to review the interview process; It also showcases the candidate's highlights and risks, generating recommendations.

[0067] In addition, this system can be set up on smart terminals such as computers or mobile phones. It can remotely connect to the candidate's audio and video terminals via the network, such as Tencent Meeting, audio and video access devices in remote conference rooms, etc. It can also use the audio and video devices of the device integrated with this system to receive the candidate's real-time voice and video information.

[0068] Example 3 like Figure 5As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0069] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to realize the steps of the dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG in the above embodiment.

[0070] Example 4 Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both built-in storage media within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of the dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG in the above embodiments.

[0071] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0072] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0074] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0075] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0076] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG, characterized in that, include: Based on the job posting, obtain the competency knowledge graph that has been pre-created for the job posting, and initialize the state machine based on the competency knowledge graph; Based on the intercepted interview conversations and candidate resumes, we use a large AI model to analyze the candidates' answers and verify the strength of evidence and logical consistency. Based on the results of AI large model analysis, the state machine is used to update the state of the corresponding node, determine the next assessment intention, and generate several candidate interview questions based on the assessment intention. Continue listening to the interview conversation until the interview ends; A structured evaluation report is generated based on the competency node information of the interview state machine.

2. The method as described in claim 1, characterized in that, The competency knowledge graph pre-created for the job includes: Based on the job description text, natural language processing is used to extract the various abilities required for the job, and each ability is taken as a competency node; the competency node includes: node name, ability name, weight, and related knowledge base scope; Edge relationship construction: Analyze the logical dependencies and business relationships between various capabilities, and construct edges; A competency knowledge graph is constructed based on the competency nodes and edges.

3. The method as described in claim 1, characterized in that, The initialization state machine includes: A state machine is created based on a competency knowledge graph, and the state machine contains multiple competency nodes corresponding to the job position. For each competency node of the state machine, initialize the following information: coverage state, evaluation confidence, and evidence strength; And set global parameters for the state machine, including: target interview duration, core node priority, and bias intervention threshold.

4. The method as described in claim 1, characterized in that, The process involves analyzing candidate responses using a large AI model based on the monitored interview dialogue stream and candidate resumes, and verifying the strength of evidence and logical consistency, including: Streaming speech recognition technology is used to convert candidate speech into text streams in real time; Natural speech processing technology is used to extract keywords from the text stream. These keywords include: technical terms, project names, and quantitative indicators. The strength of evidence is determined by deep semantic analysis based on the keywords. The current answer in the interview dialogue is cross-referenced with the candidate's resume and the candidate's past statements in this interview to achieve logical consistency verification. Preferably, the step of obtaining the evidence strength based on the keywords through deep semantic analysis includes: Responses that only mention the definition of a concept will be evaluated as weak evidence. Responses containing personal opinions but lacking supporting examples are assessed as moderate evidence; Responses that provide specific projects, roles, or quantifiable results are evaluated as strong evidence.

5. The method as described in claim 1, characterized in that, The results of the AI-based large model analysis are used to update the state of the corresponding nodes using a state machine and to determine the next step of the investigation, including: Using a state machine, the state of the corresponding competency node is updated based on the results of evidence strength and logical consistency verification. The state machine is then used to adaptively reconstruct the retrieval and enhance query generation. Preferably, updating the state of the corresponding competence node based on the evidence strength and logical consistency verification results includes: When a candidate's answer is identified as providing strong evidence and being logically clear, the confidence level of the corresponding competency node is raised to "strong". When a candidate's answer is identified as merely a theoretical explanation, the evaluation confidence of the corresponding node is marked as "weak" and a "needs further investigation" flag is added. When a candidate's current answer is found to contradict their resume or past answers, the corresponding node will be marked as "questionable".

6. The method as described in claim 5, characterized in that, The adaptive retrieval enhancement query generation includes: The current dialogue context, the node information to be examined in the state machine, and the keywords in the candidate's answer are dynamically combined to generate contextualized instructions for a specific examination target. The instructions include: query instructions for searching the enterprise knowledge base and instructions for generating follow-up question options; Preferably, the step of generating several candidate interview questions based on the assessment intent includes: The instructions generated from the aforementioned follow-up questions and multi-source evidence retrieved from the enterprise knowledge base are used as inputs; Utilize AI multimodal large models to generate follow-up questions and accompanying information for different tactical paths; The accompanying information includes: each follow-up question is accompanied by the following information: tactical description, recommendation rating, and estimated response time; The tactical approach includes: in-depth analysis, horizontal connections, and challenge reflection; The multi-source evidence includes: knowledge base fragments, historical excellent follow-up questions, and the original text of the candidates' answers.

7. The method as described in claim 1, characterized in that, After generating several candidate interview questions based on the stated examination intent, the process also includes: The system provides a human-computer coordination interface, allowing interviewers to view follow-up question options and accompanying information displayed by the large model during the interview process, as well as the status of competency nodes in the competency graph, historical question-and-answer summaries, and confidence change curves. The interface also provides real-time reminders to interviewers to correct biases, including: bias detection, warnings of discriminatory topics, and suggestions for balanced time allocation. Furthermore, the interface enables star rating, and based on the star rating, it adjusts the weights of corresponding competency nodes in the competency graph and enhances search preferences. The big data model uses each choice made by the interviewer as implicit feedback to fine-tune the system's recommendation strategy. Preferably, the generation of a structured evaluation report based on the competency node information of the interview state machine includes: Based on the competency assessment map, the human-computer coordination interface is used to display the final confidence level, key evidence fragments, and contradiction points of each node. Use a human-computer interaction interface to review the interview process; It also showcases the candidate's highlights and risks, generating recommendations.

8. A dynamic interview question generation system based on real-time dialogue analysis and adaptive RAG, characterized in that, include: An initialization module is used to obtain a competency knowledge graph pre-created for the job posting and to initialize a state machine based on the competency knowledge graph. The analysis module is used to analyze candidates' answers using a large AI model based on the monitored interview dialogue stream and candidate resumes, and to verify the strength of evidence and logical consistency. The intelligent decision-making module is used to update the state of the corresponding node using a state machine based on the results of AI large model analysis, determine the next examination intention, and generate several alternative interview questions based on the examination intention. Continue listening to the interview conversation until the interview ends; The assessment report module is used to generate a structured assessment report based on the competency node information of the interview state machine.

9. An electronic device, characterized in that, include: At least one processor and memory; The memory and processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG as described in any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that, It contains an execution program, which, when executed, implements a dynamic interview question generation method based on real-time dialogue analysis and adaptive RAG as described in any one of claims 1 to 7.

Images