An interview question divergent generation method and system

Abstract

The application provides an interview question divergent generation method, comprising the following steps: S1, obtaining current question data and candidate's current answer data in the interview interaction process, and simultaneously loading interview investigation target data, post image, candidate background auxiliary reference data; S2, performing dimension analysis and path coverage analysis on the current question data and the current answer data, identifying the opened investigation direction of the current question, the covered investigation direction of the current answer, and marking the to-be-investigated direction which is not expanded; the application realizes systematization and multi-level divergent generation of the interview question by obtaining the interview interaction data, performing path coverage analysis, constructing a dynamic update investigation space expansion diagram, generating and evaluating candidate questions, outputting optimal questions and correlation update, has systematization and multi-level divergent capacity, and significantly improves the coverage, pertinence, logic, dynamic adaptability and traceability of the interview evaluation.

Description

Technical Field

[0001] This invention relates to the field of intelligent recruitment technology, and in particular to a method and system for generating interview questions. Background Technology

[0002] With the deep integration of artificial intelligence technology into human resource management, the intelligent generation of interview questions has become a key aspect of improving the efficiency of talent assessment and strengthening the systematic nature of interview evaluation. Existing technologies have shifted from purely manual design of interview questions to semi-automated question generation based on job descriptions and resume information. Mainstream interview question generation methods include: building a general question library based on job competency models and directly retrieving corresponding questions according to ability dimensions; matching pre-set question templates with explicit keywords in the resume text to complete simple personalized content replacement; and generating a small number of related questions through limited scenario-based expansion for a single assessment point.

[0003] While the above methods have reduced the workload of manually designing questions to some extent and achieved preliminary standardization and lightweight generation of interview questions, they are limited by technical defects such as the lack of divergent logic in question generation, an incomplete dimensional system, a crude core anchoring mechanism, and insufficient optimization capabilities. As a result, they have been unable to form a systematic and multi-level divergent generation capability for interview questions, and it is difficult to adapt to the in-depth and personalized assessment needs of enterprises for candidates of different positions and levels.

[0004] The core shortcomings of existing interview question generation technologies are specifically reflected in:

[0005] The generation of questions lacks a systematic and divergent framework, resulting in a lack of scalability and hierarchy: Existing technologies do not construct a multi-level and multi-dimensional question divergence system based on the interview assessment logic. They can only expand questions at a shallow level and in one direction around a single assessment point. They cannot achieve multi-directional deduction from dimensions such as ability level, scenario extension, technical connection, and experience transfer. They also have not completed a tiered divergence design from basic cognition to in-depth application and from single scenario to multiple scenarios. This results in narrow coverage of generated questions and insufficient assessment depth, making it difficult to comprehensively assess the candidate's overall ability and job suitability.

[0006] The lack of anchoring core elements in question generation results in low relevance and accuracy: Existing technologies do not deeply refine and extract elements from job profiles, candidate backgrounds, and interview assessment needs. Without clear core question elements as anchors for question generation, the question generation process becomes random. The generated questions are prone to deviating from the actual assessment needs of the job and the candidate's real experience and background, resulting in weak job suitability and low candidate matching. It is impossible to achieve accurate question generation based on job and candidate characteristics.

[0007] The problems are divergent and lack unified deduction rules, resulting in insufficient logic and systematicity: Existing technologies do not have a standardized set of rules for integrating interview interaction logic and job assessment logic. Instead, they are mostly generated randomly without rules, which leads to a lack of progressive logical connection between the generated questions. This can easily result in problems such as repeated questions, overlapping assessment dimensions, and difficulty gaps. The resulting question set cannot support an orderly interview questioning process, which greatly reduces the systematicity and effectiveness of the interview assessment.

[0008] Static generation systems lack dynamic iterative optimization capabilities and lag in scenario adaptability: Existing interview question generation systems are mostly statically designed, lacking an iterative optimization mechanism based on actual usage feedback and self-optimization learning capabilities. They cannot dynamically adjust divergent logic based on the effectiveness of question usage during the interview process and interviewer evaluation opinions. At the same time, they have poor adaptability to changes in industry assessment trends and job requirements. When the assessment scenario or requirements change, the question generation rules and dimensional system cannot be optimized in a timely manner, resulting in generated questions that are prone to assessment failure.

[0009] The question set has a low degree of structure and lacks traceability and reusability: the interview questions generated by existing technologies are mostly single text information, which do not establish a structured relationship with the assessment elements and divergent dimensions, and there is no standardized storage and indexing mechanism. As a result, the basis for the generation of questions cannot be traced, and it is difficult to modularly retrieve and reuse them according to different interview scenarios and assessment needs. This increases the repetitive workload of subsequent interview question design and reduces the overall efficiency of question generation. Summary of the Invention

[0010] In view of this, the present invention aims to provide a method and system for generating interview questions in a divergent manner, so as to solve or alleviate the technical problems existing in the prior art, or at least provide a beneficial option.

[0011] To address the aforementioned technical problems, this application adopts the following technical solution: providing a method for generating interview questions, comprising the following steps:

[0012] S1. Obtain the current question asked during the interview interaction and the candidate's current answer data, and load the interview assessment target data, job profile, and candidate background auxiliary reference data;

[0013] S2. Perform dimensional analysis and path coverage analysis on the current question and the current answer data to identify the examination directions that the current question has opened up and the examination directions that the current answer has covered, and mark the examination directions that have not yet been developed.

[0014] S3. Based on the data of the interview assessment targets and the path coverage analysis results, construct an interview assessment space expansion diagram to represent the coverage status, dimensional relationships and divergent expansion paths of the assessment directions in a structured form.

[0015] S4. Using the direction to be examined in the spatial development diagram as the expansion anchor point, generate multiple sets of candidate divergent questions in combination with the interview assessment target data.

[0016] S5. Perform targeted evaluation on candidate divergent questions. The core judgment dimensions include: whether it can reach the direction that has not been explored, whether it can compensate for the coverage gap of the current investigation path, whether it can avoid high overlap with the already covered path, and whether it can improve the evidence stability of the investigation target. Based on the judgment results, determine the optimal divergent question.

[0017] S6. Output the optimal divergent problem and associate it with the expansion path and direction to be investigated in the exploration space unfolding graph, and update the coverage status information of the exploration space unfolding graph synchronously.

[0018] Furthermore, the auxiliary reference data in S1 includes at least core data of job profile, key data of candidate background, and core interview assessment requirements. The basic reference data is modeled using structured modeling and vector embedding techniques to form a standardized basic feature vector library, which provides feature support for subsequent assessment direction identification and question generation.

[0019] Both the current question and the current answer data are preprocessed using natural language understanding technology. The preprocessing includes word segmentation, semantic role labeling, and assessment dimension mapping to transform unstructured text into structured assessment dimensions.

[0020] Furthermore, the path coverage analysis in S2 is performed based on a pre-defined library of dimensions to be examined, specifically including:

[0021] Match the current question with the corresponding assessment dimensions to determine all potential assessment directions opened up by the question;

[0022] Extract semantic features from the current answer data, match them with potential examination directions, and identify the actual examination directions covered;

[0023] The unexpanded directions to be examined are obtained by difference calculation, and the matching degree and expansion priority of each direction to be examined are marked with the interview assessment objectives.

[0024] Furthermore, the interview assessment space unfolding graph in S3 is a dynamically updatable structured representation carrier, whose core components include at least: current question node, current answer node, covered direction node, direction to be assessed node, direction-related edge, and candidate divergent question node;

[0025] Nodes are used to carry direction name, coverage status, matching degree, and priority information. Direction-related edges are used to represent the relationship between dimensions and the expansion path. The unfolded graph supports dynamic updates of node addition, deletion, status modification, and relationship adjustment.

[0026] Furthermore, S4 generates multiple sets of candidate divergent problems, specifically:

[0027] Based on the node information of the direction to be examined in the spatial development diagram and the data of the interview assessment targets, it is generated in accordance with the anchoring rules of the direction to be examined.

[0028] The generated candidate divergent questions are strongly tied to the direction to be examined, and each group of candidate questions serves to achieve the interview assessment objectives. It supports the generation of combined questions for a single direction to be examined or multiple directions to be examined.

[0029] Furthermore, the targeted assessment in S5 is a coverage-compensation-oriented decision-making and screening process, with the assessment logic focusing on:

[0030] Reach determination: Can the candidate question directly reach at least one unexplored direction to be investigated?

[0031] Gap compensation determination: Can the candidate problem fill the coverage gap of the current investigation path?

[0032] Overlap determination: Whether the candidate question has low semantic overlap with the already asked question and the covered directions;

[0033] Stability assessment: Can the candidate questions provide new and valid evidence for the interview assessment objectives, thereby improving the stability of the evaluation?

[0034] Furthermore, the targeted evaluation employs a priority screening rule:

[0035] Prioritize candidate questions that can reach the core areas to be investigated;

[0036] Based on meeting the reach requirements, the candidate problem with the best coverage gap compensation capability is selected;

[0037] When both of the above conditions are met, the candidate problem with the lowest overlap with the already covered path and the most significant improvement in the stability of the evidence for the target being examined is selected as the optimal divergent problem.

[0038] Furthermore, it also includes examining the dynamic iteration and self-optimization steps of the spatial unfolding graph:

[0039] Collect follow-up question feedback, answer feedback and evaluation feedback data during the interview interaction process, and adjust the node information and directional relationships of the assessment space development diagram based on the feedback data;

[0040] Simultaneously optimize the candidate question generation rules and targeted evaluation and judgment rules to make the divergent generation logic adapt to actual interview interaction scenarios.

[0041] Furthermore, the examination space unfolding graph supports multi-level dynamic expansion of the examination direction. The expansion dimensions are adaptively configured based on industry attributes, job level, and interview type. New expansion dimensions are directly associated with the original topology graph nodes, automatically generating edge attributes and integrating them into the examination space unfolding graph.

[0042] Furthermore, an interview question generation system includes:

[0043] The interactive data acquisition module is used to acquire the current question, current answer data, interview assessment target data, and auxiliary reference data, and to complete the data preprocessing and standardization.

[0044] The path coverage analysis module is used to analyze the dimensions of investigation and identify the investigation directions that have been opened, covered, and not yet developed.

[0045] The assessment space construction module is used to build and dynamically update the unfolded graph of the interview assessment space, and maintain the nodes, associated edges and coverage status;

[0046] The candidate question generation module is used to generate multiple sets of candidate divergent questions based on the direction to be investigated and the investigation objectives.

[0047] The coverage compensation assessment module is used to filter the optimal divergent problem based on rules such as reachability, gap compensation, low overlap, and stability improvement.

[0048] The output update module is used to output the optimal divergent problem and complete the association and binding between the problem and the unfolded graph, as well as the update of the unfolded graph coverage status.

[0049] The embodiments of the present invention have the following advantages due to the adoption of the above technical solutions:

[0050] This invention achieves the systematic and multi-level divergent generation of interview questions by acquiring interview interaction data, performing path coverage analysis, constructing a dynamically updated assessment space expansion graph, generating and evaluating candidate questions, outputting the optimal question and updating it accordingly. It has systematic and multi-level divergent capabilities, significantly improving the coverage, relevance, logic, dynamic adaptability and traceability of interview assessment.

[0051] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of the invention will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of the method of the present invention;

[0054] Figure 2 This is a flowchart of the path coverage analysis method of the present invention;

[0055] Figure 3 This is a flowchart illustrating the construction of a spatial unfolded diagram for the method of this invention;

[0056] Figure 4 This is a schematic diagram of the modules of the system of the present invention. Detailed Implementation

[0057] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0058] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0059] Traditional interview question generation technology faces many challenges in the development of intelligent technology. Existing methods mostly rely on general question libraries or keyword matching, resulting in a lack of systematic divergent dimension system for question generation, a crude core element anchoring mechanism, inconsistent deduction rules, and a lack of dynamic iterative optimization capabilities. This makes the generated questions narrow in coverage, low in relevance and accuracy, and difficult to meet the needs of enterprises for in-depth and personalized talent assessment.

[0060] like Figure 1 As shown, this application proposes a method for generating interview questions in a divergent manner, including the following steps:

[0061] S1. Obtain the current question asked during the interview interaction and the candidate's current answer data, and load the interview assessment target data, job profile, and candidate background auxiliary reference data;

[0062] S2. Perform dimensional analysis and path coverage analysis on the current question and the current answer data to identify the examination directions that the current question has opened up and the examination directions that the current answer has covered, and mark the examination directions that have not yet been developed.

[0063] S3. Based on the data of the interview assessment targets and the path coverage analysis results, construct an interview assessment space expansion diagram to represent the coverage status, dimensional relationships and divergent expansion paths of the assessment directions in a structured form.

[0064] S4. Using the direction to be examined in the spatial expansion diagram as the expansion anchor point, generate multiple sets of candidate divergent questions in combination with the interview assessment target data.

[0065] S5. Perform targeted evaluation on candidate divergent questions. The core judgment dimensions include: whether it can reach the direction that has not been explored, whether it can compensate for the coverage gap of the current investigation path, whether it can avoid high overlap with the already covered path, and whether it can improve the evidence stability of the investigation target. Based on the judgment results, determine the optimal divergent question.

[0066] S6. Output the optimal divergent problem and associate it with the expansion path and direction to be investigated in the exploration space unfolding graph, and update the coverage status information of the exploration space unfolding graph synchronously.

[0067] For ease of understanding, the following explains some key terms in this embodiment:

[0068] Interview assessment target data refers to the set of specific indicators, such as abilities, knowledge, and experience, that are pre-set to be assessed during the interview process to evaluate whether a candidate meets the requirements of a specific position. This data provides the core basis for question generation and evaluation.

[0069] A job profile is a dataset that provides a comprehensive and structured description of a specific job's responsibilities, required skills, qualifications, and cultural fit. It is used to guide the generation of interview questions, ensuring a high degree of alignment between the questions and the job requirements.

[0070] Candidate background supplementary data refers to data beyond the current answer that helps in understanding the candidate's overall situation, such as their educational background, work experience, project experience, and skills certifications. This data helps in a more comprehensive evaluation of the candidate and provides a personalized basis for expanding on the questions.

[0071] Dimensional analysis refers to semantic analysis of textual information (such as questions and answers) and mapping it to a pre-defined dimensional system to identify the scope and depth of the examination.

[0072] Path coverage analysis refers to the process of comparing the assessment dimensions involved in the questions asked and the answers answered with a pre-defined assessment dimension library to identify the knowledge, skills, or experience paths that have been assessed and those that have not, thereby evaluating the comprehensiveness of the assessment.

[0073] The assessment focus refers to the specific entry point or sub-dimension for in-depth exploration of a particular ability, knowledge point, or area of ​​experience during the interview process.

[0074] An interview assessment space unfolding diagram is a tool that uses a structured graphical representation to dynamically depict the coverage of all assessment directions, the relationships between dimensions, and potential divergent expansion paths during the interview assessment process. This diagram provides the interviewer with a global perspective, guiding subsequent questioning.

[0075] Areas to be explored refer to those that, at this stage of the interview, have not yet been fully explored or covered, based on the interview objectives and the findings already made. These areas will be the focus of subsequent divergent questions.

[0076] Anchor points for expansion refer to the directions to be examined that are selected as the core basis or starting point for generating new divergent questions in the interview assessment space expansion diagram.

[0077] Candidate divergent questions refer to multiple sets of alternative questions generated by algorithms based on the areas to be assessed and the objectives of the interview. These questions aim to further explore previously unexplored areas of assessment.

[0078] Targeted evaluation refers to the process of evaluating and screening generated candidate divergent questions in a targeted manner. This evaluation focuses on whether the question can effectively reach the direction of investigation, fill the gaps in investigation, avoid duplication, and improve the stability of the investigation evidence.

[0079] The optimal divergent problem refers to the candidate problem that, after targeted evaluation, is identified as the most effective way to advance the interview process and improve the quality of the assessment.

[0080] This application provides a method for generating interview questions in a divergent manner.

[0081] In step S1, the system acquires the current question and the candidate's current answer during the interview interaction, and simultaneously loads the interview assessment target data, job profile, and candidate background auxiliary reference data. The current question and answer data can be obtained through manual input or by converting them into text using speech recognition technology. The interview assessment target data, job profile, and candidate background auxiliary reference data can be loaded by the system from a preset database. As one implementation method, the interview assessment target data, job profile, and candidate background auxiliary reference data can be pre-stored in the database and automatically loaded by the system when the interview starts. For example, the interviewer can manually input the candidate's answer into the system, and the system will simultaneously retrieve the assessment target, profile information, and candidate resume information for the position from the preset database. Furthermore, the current question and answer data can be acquired through real-time speech recognition technology and automatically converted into text. This interview assessment target data, job profile, and candidate background auxiliary reference data can be configured and uploaded to the system by the system administrator before the interview. For example, the system can capture the interviewer's questions and the candidate's answers in real time through a microphone, convert them into text data, and simultaneously load the interviewer's preset assessment targets and the background information submitted by the candidate.

[0082] In step S2, dimensional analysis and path coverage analysis are performed on the current question and answer data to identify the assessment directions that the current question has opened and the assessment directions that the current answer has covered, and to mark the directions that have not yet been explored. Dimensional analysis can be implemented through keyword matching or a rule engine to associate text content with preset assessment dimensions. Path coverage analysis can be performed through set operations to compare the identified assessment directions with a preset assessment dimension library to determine the covered and unexplored directions. For example, the system identifies specific words in the current question and categorizes them into assessment dimensions such as "communication skills" or "technical depth." At the same time, based on the keywords appearing in the candidate's answer, it determines whether the candidate has covered the specific assessment points under these dimensions. Uncovered assessment points are marked as directions to be explored. As another implementation method, dimensional analysis can use a text classification model based on machine learning to map questions and answers to a multi-dimensional assessment system. Path coverage analysis can be based on a predefined assessment path graph, traversing the graph nodes to identify the opened, covered, and unexplored directions. For example, the system uses a trained semantic model to analyze the deeper intentions of the interviewer's questions and the knowledge points of the candidate's answers, and matches them with the preset sub-dimensions such as "analytical ability" and "decision-making ability" under "problem-solving ability", and updates the coverage status of the assessment path based on the matching results.

[0083] like Figure 2As shown, in step S3, based on the interview assessment target data and path coverage analysis results, an interview assessment space unfolding graph is constructed to represent the coverage status, dimensional relationships, and divergent expansion paths of the assessment directions in a structured form. The interview assessment space unfolding graph can be constructed as a tree structure or a list, where each node represents an assessment direction, and the node's status (e.g., "covered," "to be assessed") can be distinguished by color or label. Dimensional relationships can be represented by parent-child nodes or links. For example, the system can use "technical ability" as the root node, containing child nodes such as "programming language" and "algorithm," each child node can be further subdivided into specific knowledge points, and the knowledge points that have been assessed are marked with different colors. Furthermore, the interview assessment space unfolding graph can be constructed using a graph database or knowledge graph, where nodes represent assessment directions, and edges represent the relationships and expansion paths between dimensions. Each node can be configured to store attribute information such as coverage status, matching degree, and priority. For example, the system can construct a graph centered on "project experience", where the "project experience" node is connected to nodes such as "technology stack", "team collaboration", and "problem solving" through different types of edges. These edges indicate other dimensions of consideration that can be derived from project experience.

[0084] like Figure 3 As shown, in step S4, the direction to be examined in the expanded examination space diagram is used as the expansion anchor point, and multiple sets of candidate divergent questions are generated in combination with the interview examination target data. Candidate divergent questions can be generated using preset question templates and keyword replacement mechanisms. The system selects the corresponding template from the question template library based on the name of the direction to be examined and fills the template with keywords related to the interview examination target. For example, if the direction to be examined is "concurrent programming," the system can generate "Please describe your experience in concurrent programming" from the template "Please describe your experience in [direction to be examined]." As a preferred implementation, candidate divergent questions can be generated using a rule-based generation engine or a sequence generation model. Based on the semantic information of the direction to be examined and the interview examination target, the system combines preset generation rules or a trained model to generate multiple sets of grammatically correct and semantically relevant questions. For example, for the direction to be examined, "data structure optimization," the system can generate questions such as "How do you evaluate the performance bottleneck of a data structure?" and "When processing large-scale data, which data structure optimization strategies do you prioritize?"

[0085] In step S5, targeted evaluation is performed on candidate divergent questions. The core judgment dimensions include: whether it can reach unexplored directions for investigation, whether it can compensate for coverage gaps in the current investigation path, whether it can avoid high overlap with already covered paths, and whether it can improve the evidence stability of the investigation target. The optimal divergent question is determined based on the judgment results. Targeted evaluation can be implemented using a simple matching algorithm. The system performs keyword matching between candidate divergent questions and unexplored directions for investigation to determine their reachability; it evaluates the overlap by calculating the semantic similarity between the candidate question and the already covered paths; and it judges whether the question can provide new evidence for the investigation target using preset rules. For example, if a candidate question contains the keyword "distributed transactions," and "distributed transactions" is a direction for investigation, then the question is considered to reach that direction. If the question has a large semantic difference from previously asked questions, the overlap is judged to be low. Furthermore, targeted evaluation can employ a multi-dimensional scoring model. Each judgment dimension (reachability, gap compensation, overlap, evidence stability) can be assigned different weights. The system scores each candidate question on these dimensions and combines the scores to determine the optimal divergent question. For example, the system uses semantic embedding technology to calculate the semantic distance between candidate questions and the direction to be investigated, with higher scores for closer distances; it also calculates the semantic similarity with already covered paths, with higher scores for lower similarity; and assesses the potential to provide new evidence based on the relevance of the question type to the investigation target.

[0086] In step S6, the optimal divergent question is output and associated with the expansion path and the direction to be examined in the examination space unfolding graph, and the coverage status information of the examination space unfolding graph is updated synchronously. The optimal divergent question can be directly displayed to the interviewer. In the examination space unfolding graph, the system links the question with the corresponding direction to be examined via text, and updates the status of the direction to be examined from "to be examined" to "examined" or "partially examined". For example, the system can display the selected optimal question on the interface, associate the question with the "concurrent programming" node in the graph, and update the status of the "concurrent programming" node to "covered". As one implementation method, the optimal divergent question can be sent to the interviewer's interview tool via API interface. In the examination space unfolding graph, the system establishes bidirectional links between the optimal divergent question and the expansion path and the direction to be examined, and dynamically adjusts the coverage status and priority of the relevant nodes according to the actual examination effect of the question. For example, the system pushes the best question to the interviewer's questioning interface, and in the background assessment space expansion graph, creates a new node for the question and connects it to the corresponding assessment direction node through an "already asked" edge, while updating the coverage percentage of the assessment direction node.

[0087] This application acquires multi-source interview data, performs dimensional analysis and path coverage analysis on questions and answers, and constructs a dynamically updatable interview assessment space expansion graph, achieving structured representation and management of assessment directions. Based on this, using the assessment direction as an expansion anchor point, and combining a targeted evaluation mechanism, it generates optimal divergent questions. This effectively solves the problems of missing problem divergent logic, coarse anchoring mechanisms, and lack of dynamic optimization in existing technologies, thereby improving the systematicness, relevance, and accuracy of interview question generation, ensuring the comprehensiveness and depth of candidate assessment.

[0088] In divergent question generation methods for interviews, acquiring the current question asked, the candidate's current answer, and supplementary reference data during the interview interaction is fundamental for subsequent analysis and question generation. However, this raw data often exists in an unstructured form, and its inherent information density and correlation are insufficient. Directly using it for identification of examination direction and question generation may lead to low analysis efficiency and insufficient accuracy, making it difficult to effectively support the subsequent steps' requirements for structured and standardized features, thus affecting the quality of divergent question generation and the accuracy of the examination.

[0089] This application further proposes a scheme for refining the data acquired in step S1. Specifically, the auxiliary reference data in S1 includes at least core job profile data, key candidate background data, and core interview assessment requirements. These data are crucial information for constructing the interview assessment framework and understanding candidate characteristics. To fully utilize these auxiliary reference data, this application employs structured modeling and vector embedding techniques for information modeling. Structured modeling involves organizing these data in a predefined pattern, such as breaking down job profiles into fields like skill requirements, experience requirements, and responsibilities, and candidate background data into fields like education, work experience, and project achievements, while defining clear dimensions and indicators for interview assessment requirements. Based on this, vector embedding techniques, such as using deep learning models, are used to map these structured data into a high-dimensional vector space, thereby forming a standardized basic feature vector library.

[0090] Meanwhile, this application preprocesses the current question and the candidate's current answer data generated during the interview interaction using natural language understanding technology. This preprocessing aims to transform unstructured text information into structured assessment dimensions that can be understood and analyzed by machines. Specific preprocessing steps include: first, word segmentation, dividing continuous text into words or phrases with independent semantics; second, semantic role labeling, identifying predicates and their arguments (such as agent, patient, time, place, etc.) in the sentence, thereby revealing the deep semantic structure and event information of the sentence; and finally, assessment dimension mapping, associating and matching the segmented and semantically labeled text information with a pre-defined interview assessment dimension library, for example, mapping "leading a team to complete a complex project" mentioned in the answer to assessment dimensions such as "leadership" and "project management ability." Through these preprocessing steps, the transformation from unstructured text to structured assessment dimensions is achieved, enabling the system to accurately grasp the intent of the current question and the assessment points covered by the current answer.

[0091] The above technical solution involves structured modeling and vector embedding of auxiliary reference data, and natural language understanding preprocessing of the current question and answer data. This transforms the raw, unstructured interview-related data into standardized feature vectors and structured assessment dimensions. This significantly improves data quality and analyzability, providing accurate and efficient input for the subsequent dimension analysis and path coverage analysis in step S2.

[0092] This application further proposes path coverage analysis in S2, which is performed based on a preset assessment dimension library. Specifically, it includes: matching the assessment dimension corresponding to the current question to determine all potential assessment directions opened by the question; extracting the semantic features of the current answer data and matching them with the potential assessment directions to mark the actual covered assessment directions; obtaining the unexpanded assessment directions through difference calculation, and marking the matching degree and expansion priority of each assessment direction with the interview assessment target.

[0093] Path coverage analysis is a crucial step in identifying the areas already assessed and those yet to be assessed during the interview process. Executing this analysis based on a pre-defined assessment dimension library means it is not arbitrary but relies on a predefined, structured knowledge system. This assessment dimension library can be a multi-level tree or graph structure, containing various assessment dimensions related to job requirements, competency models, and knowledge domains, such as communication skills, problem-solving abilities, professional knowledge, and project experience. Each dimension can be further refined into sub-dimensions or specific assessment points.

[0094] When matching the current interview question with the corresponding assessment dimensions to determine all potential assessment directions, the aim is to understand the scope of assessment the current interview question points to. In practice, Natural Language Processing (NLP) techniques, such as keyword extraction, topic modeling, and semantic similarity calculation, can be used to match the current interview question with various dimensions in a pre-defined assessment dimension library. For example, if the question includes "What was the biggest challenge you encountered in a project? How did you solve it?", the system might match it with assessment dimensions such as "problem-solving ability," "stress resistance," and "project management experience." Through this matching, all potential assessment directions that the question might touch upon in the assessment dimension library can be identified—that is, what aspects of discussion the question could theoretically elicit. This provides a foundation for subsequently determining whether the answer covers these directions.

[0095] When extracting semantic features from the current answer data and matching them with potential assessment directions to identify the actual assessment directions covered, this process is used to evaluate which assessment points the candidate's answer actually addresses. After acquiring the current answer data, natural language processing techniques, such as named entity recognition, sentiment analysis, and semantic vector embedding, are also used to extract key information and semantic features from the answer. Subsequently, these semantic features are compared and matched with the potential assessment directions identified in the previous step.

[0096] By using set difference calculations, unexplored directions for further investigation are identified, and each direction is labeled with its matching degree with the interview assessment objectives and its expansion priority. After determining all potential directions for investigation opened up by the current question and the directions actually covered by the current answer data, set difference operations can clearly identify those "unexplored directions for further investigation" that have been raised but not fully covered by the answers, or those that have not been touched upon but fall within the scope of the interview assessment objectives.

[0097] Through the above technical solution, this application overcomes the problems of inaccurate identification and incomplete coverage of assessment directions in traditional interviews. Path coverage analysis based on a pre-set assessment dimension library ensures the standardization and systematic nature of assessment direction identification, avoiding bias from subjective judgment. By accurately matching the semantic features of the questions and answers, it is possible to quantitatively identify the potential assessment directions opened up by the questions and the assessment directions actually covered by the answers, thereby clearly defining the unexplored assessment directions.

[0098] This application further proposes that the interview assessment space unfolding graph in S3 is a dynamically updatable structured representation carrier, whose core components include at least: current question node, current answer node, covered direction node, direction to be assessed node, direction association edge, and candidate divergent question node; the node is used to carry direction name, coverage status, matching degree, and priority information, the direction association edge is used to represent the association relationship and expansion path between dimensions, and the unfolding graph supports dynamic updates of node addition and deletion, status modification, and association relationship adjustment.

[0099] Specifically, the interview assessment space unfolding diagram is designed as a medium capable of organizing and storing information using predefined data structures (e.g., graph structures), and its content can be modified, added to, or deleted based on real-time input during the interview process or changes in the system's internal logic to ensure it always remains consistent with the actual progress of the interview. This dynamically updatable characteristic allows the unfolding diagram to flexibly adapt to the non-linear and interactive nature of interviews.

[0100] The core components of this unfolded graph include various types of nodes and directional edges. The "Current Question" node records the interviewer's current question, serving as the starting point for subsequent analysis and question generation. The "Current Answer" node contains the candidate's response to the current question, facilitating semantic and path coverage analysis. "Covered Direction" nodes identify directions that have been thoroughly examined during the interview, avoiding repetitive questioning. "Direction to be Examined" nodes explicitly indicate directions that have not yet been fully explored or addressed in the interview, serving as core anchor points for generating subsequent divergent questions.

[0101] The nodes not only carry the name of the direction they represent, but also contain key information such as coverage status, matching degree, and priority. Coverage status indicates whether the assessment direction has been fully examined, such as "not covered," "partially covered," or "covered." Matching degree reflects the degree to which the assessment direction aligns with the overall interview assessment objectives. Priority information guides the order and focus of question generation. These information, as node attributes, make the current status and importance of each assessment direction immediately clear.

[0102] The directional association edges are used to connect different nodes to represent the logical relationships, hierarchical relationships, or expansion paths between the dimensions of examination. For example, an edge can represent a "precondition" or "association," thereby constructing the topological structure of the examination space and guiding the logical direction of problem divergence. This structured association helps the system understand the intrinsic connections between different examination points, thus generating more logical and coherent divergent problems.

[0103] To enable dynamic updates, the expanded graph supports adding and deleting nodes, modifying their status, and adjusting their relationships. This means that during the interview process, new nodes representing different assessment directions can be dynamically added or irrelevant nodes can be deleted based on new information or assessment needs; the coverage status, matching degree, or priority of nodes can be updated in real time; and the edges between nodes can be modified or added based on the progress of the interview or new insights to more accurately reflect the dynamic relationships between assessment dimensions.

[0104] By designing the interview assessment space unfolding graph as a dynamically updatable structured representation and clearly defining its core constituent nodes and directional association edges, this application can capture dynamic changes during the interview process in real time and with precision. Specifically, the current question node and the current answer node can instantly reflect the latest progress of the interview, while the status updates of covered direction nodes and pending direction nodes ensure the accuracy and timeliness of the assessment direction. The directional association edges clearly represent the logical relationships and expansion paths between dimensions, providing clear navigation for subsequent question generation. This dynamic update mechanism allows the unfolding graph to continuously synchronize with the actual interview process, effectively avoiding the problem of question generation being out of sync with the interview context, thereby significantly improving the accuracy and relevance of divergent question generation and the depth and efficiency of interview assessment.

[0105] This application further proposes to generate multiple sets of candidate divergent questions in the above S4, specifically: based on the node information of the direction to be examined in the examination space expansion diagram and the interview examination target data, the candidate divergent questions are generated in accordance with the anchoring rules of the direction to be examined; the generated candidate divergent questions are strongly bound to the direction to be examined, and each set of candidate questions serves to achieve the interview examination target, supporting the generation of combined questions for a single direction to be examined or multiple directions to be examined.

[0106] Specifically, the node information of the direction to be examined in the assessment space expansion diagram refers to the detailed data contained in the node corresponding to each direction to be examined in the interview assessment space expansion diagram, such as the name of the direction to be examined, its level in the assessment system, its matching degree with the interview assessment target, its current expansion priority, and its relationship with other directions to be examined.

[0107] The generated candidate divergent questions are strongly linked to the areas to be assessed, meaning that after the questions are generated, the system establishes a clear relationship, indicating which areas(s) each candidate question is designed for. This strong linking mechanism ensures that each generated question has a clear assessment purpose, avoiding the generation of questions irrelevant to the assessment objectives. Furthermore, each set of candidate questions serves to achieve the interview assessment goals, emphasizing that all generated candidate questions should not only cover the areas to be assessed but also, as a whole, promote a comprehensive evaluation of the interview assessment objectives.

[0108] The above technical solution uses the node information of the direction to be examined in the spatial development diagram and the target data of the interview as the core basis, and generates candidate divergent questions in accordance with the anchoring rules of the direction to be examined, ensuring the logic and directionality of the question generation process. At the same time, the generated candidate divergent questions are strongly bound to the direction to be examined, and it is clear that each group of candidate questions serves to achieve the interview assessment goal, effectively avoiding blindness and deviation in the question generation process.

[0109] In the method for generating divergent interview questions, step S5 requires targeted evaluation of multiple sets of candidate divergent questions to determine the optimal divergent question. However, if the evaluation logic is not clear and focused enough, the selected optimal divergent question may fail to effectively address the blind spots in the interview assessment or may have a high degree of overlap with already assessed content, thereby reducing interview efficiency and the comprehensiveness of the assessment. This makes it difficult for interviewers to accurately guide the interview process and ensure a comprehensive and in-depth assessment of the candidate.

[0110] This application further proposes that the targeted assessment in S5 is a coverage compensation-oriented judgment and screening, and its assessment logic focuses on reach determination, gap compensation determination, overlap determination, and stability determination.

[0111] Specifically, the coverage-compensation-oriented decision-making and screening aims to ensure that the selected questions can maximally compensate for the deficiencies of the current investigation. The reach determination refers to assessing whether the candidate questions can directly reach at least one unexplored direction to be investigated. This is typically achieved by analyzing the matching degree between the semantic content of the candidate questions and the keywords, topics, or concepts of the unexplored direction.

[0112] Gap filling assessment refers to evaluating whether candidate questions can fill coverage gaps in the current assessment path. Coverage gaps in the current assessment path may manifest as a lack of sufficiently deep information in a certain assessment dimension, or a key skill point that has not yet been effectively verified. This assessment determines the filling capability by analyzing the information that candidate questions can elicit and comparing it with identified weaknesses or insufficiently verified sub-dimensions in the current assessment path.

[0113] Overlap assessment refers to evaluating whether candidate questions have low semantic overlap with previously asked questions or covered topics. This assessment aims to avoid asking the same questions repeatedly or examining previously obtained information, thereby improving interview efficiency. Overlap can be quantified by calculating the semantic similarity between candidate questions and historical questions or covered topics. For example, word vector models or topic models can be used to measure the semantic distance between texts, ensuring that the selected questions elicit new, non-redundant information.

[0114] Stability assessment refers to evaluating whether candidate questions can provide new and valid evidence for the interview's objectives, thereby improving the stability of the assessment. This assessment focuses on whether candidate questions can verify or supplement the assessment of the candidate from different angles or at a deeper level, making the final assessment results more comprehensive and reliable.

[0115] Through the aforementioned technical solution, this application ensures that the selected optimal divergent questions not only effectively broaden the scope of interview assessments, touching upon previously underexplored areas, but also precisely fill coverage gaps in the current assessment path, avoiding redundant questioning and information redundancy. This significantly improves the efficiency and comprehensiveness of the interview process, enabling interviewers to obtain more systematic and in-depth information about candidates, thereby providing richer and more stable evidence to support the interview assessment objectives, ultimately improving the accuracy and reliability of the interview evaluation.

[0116] This application further proposes that the targeted evaluation adopts a priority screening rule: priority is given to selecting candidate questions that can reach the core direction to be examined; on the basis of meeting the reach requirements, the candidate questions with the best coverage gap compensation ability are selected; when the first two conditions are met, the candidate question with the lowest overlap with the already covered path and the most significant improvement in the stability of the evidence of the examined target is selected as the optimal divergent question.

[0117] This priority selection rule aims to provide a structured decision-making framework for targeted evaluation in the divergent generation method of interview questions. It ensures that when choosing among multiple candidate divergent questions, they can be systematically sorted and selected according to a pre-defined order of importance, thereby selecting the question that best matches the current interview assessment objective. "Core areas to be assessed" refer to unexplored dimensions in the interview assessment target data that are marked as highly important, high-priority, or have a decisive impact on job competency. These core areas may be derived through pre-defined weights, expert experience annotations, or analysis based on historical data. A candidate question "reaching" the core areas to be assessed means that the question can directly guide the candidate to answer content related to that core area semantically or in terms of assessment intent, such as through keyword matching, topic model analysis, or judging the model using pre-trained question-and-answer tools. Given that the core areas to be assessed have been reached, "optimal gap-filling ability" means that the candidate question can fill the gaps or deficiencies in the current interview assessment path to the greatest extent possible. This can be achieved by quantitatively evaluating the number of unassessed sub-dimensions covered by the candidate question, the importance of these sub-dimensions, or their contribution to completing a specific assessment chain.

[0118] Through the above technical solution, this application effectively addresses the problem of efficiently selecting the optimal divergent question under multi-dimensional evaluation. First, by prioritizing the core areas to be assessed, it ensures that the interview always focuses on the most critical points, avoiding deviations in the assessment direction and significantly improving the strategic nature and effectiveness of the interview. Second, with the core areas secured, further optimization optimizes the selection of questions with the best gap-filling ability, making the interview assessment path more complete and comprehensive, avoiding fragmented assessment, and ensuring a systematic evaluation of the candidate's abilities. Finally, by balancing low overlap and high evidence stability, it not only avoids repeated questioning and improves interview efficiency but also enhances the reliability and persuasiveness of the interview evaluation results. This ensures that the ultimately selected optimal divergent question accurately addresses the key issues, comprehensively supplements information, and provides high-quality decision-making basis, thereby significantly improving the overall quality of the interview and the accuracy of the decision.

[0119] This application further proposes a divergent generation method for interview questions, which also includes a dynamic iteration and self-optimization step for the examination space unfolding graph: collecting subsequent question feedback, answer feedback and evaluation feedback data during the interview interaction process, adjusting the node information and directional correlation of the examination space unfolding graph based on the feedback data; and simultaneously optimizing the candidate question generation rules and targeted evaluation and judgment rules to make the divergent generation logic adapt to the actual interview interaction scenario.

[0120] Specifically, the phrase "collecting follow-up question feedback, answer feedback, and evaluation feedback data during the interview interaction process, and adjusting the node information and directional relationships of the assessment space unfolding diagram based on the feedback data" refers to the system actively or passively collecting data related to the interview process during or after the interview. Follow-up question feedback may include whether the interviewer actually adopted the system-recommended divergent questions, or questions raised independently by the interviewer outside of the system's recommendations. This data can be obtained through system logs or explicit selection by the interviewer. Answer feedback data refers to the candidate's specific answers to follow-up questions, which can be obtained through speech recognition-to-text technology or direct text input. Evaluation feedback data covers the interviewer's evaluation of the candidate's answers, judgment of coverage of specific assessment directions, and satisfaction with the entire interview process. This data can be collected through structured scoring sheets, free text evaluations, or post-interview questionnaires. Based on this feedback data, the system can dynamically adjust the assessment space unfolding diagram. For example, if a particular area of ​​expertise is successfully explored in multiple interviews and receives positive feedback from interviewers, its priority or matching degree in the expanded graph of the assessment space can be increased accordingly; conversely, if a particular area is difficult to effectively assess in actual interviews or is proven unimportant, its weight may be reduced. Simultaneously, if certain related edges frequently appear in the actual interview path and are verified to be effective, their association strength or weight in the expanded graph can be enhanced. These adjustments can be made automatically by the system based on preset machine learning models or rules, or semi-automatically with the confirmation of the interviewers.

[0121] Meanwhile, the phrase "synchronously optimizing candidate question generation rules and targeted evaluation rules to adapt the divergent generation logic to actual interview interaction scenarios" refers to the system continuously improving the internal logic used to generate candidate divergent questions and select the optimal divergent question based on collected feedback data. If the system finds that certain types of areas to be examined are consistently difficult to generate high-quality candidate questions, or that the generated candidate questions are not highly relevant to the interview assessment target data, the candidate question generation rules can be adjusted. This may involve updating question templates, adjusting keyword weights, introducing new semantic similarity calculation models, or training the generation model through reinforcement learning to better adapt to the interview scenario. Similarly, if the optimal divergent question selected by the system performs poorly in actual interviews, for example, if it has a high overlap with already covered paths or fails to effectively reach unexplored areas to be examined, the targeted evaluation rules can be adjusted. This includes adjusting the weights of various dimensions such as reach determination, gap compensation determination, overlap determination, and stability determination, and even introducing new judgment factors, such as "question novelty" or "potential for in-depth exploration." Through this continuous optimization, the divergent generation logic can better understand and predict the interviewer's needs, generating questions that are more in line with the actual interview process and key assessment points, thereby improving the system's practicality and effectiveness.

[0122] Through the above technical solution, this application effectively addresses the problem of insufficient adaptability of divergent question generation methods in dynamic interview scenarios. By continuously collecting feedback data on subsequent questions, answers, and evaluations during the interview interaction process, the system can understand the actual progress and effectiveness of the interview in real-time or near real-time. Based on this feedback data, the node information and directional relationships of the examination space unfolding graph are dynamically adjusted, enabling the unfolding graph to more accurately reflect the current state of the interview and the priority of the directions to be examined. Simultaneously, the candidate question generation rules and targeted evaluation judgment rules are optimized to ensure that the system can generate more targeted and effective questions according to changes in the actual interview scenario, and select the truly optimal divergent questions. This dynamic iteration and self-optimization mechanism allows the divergent question generation logic to continuously learn and evolve, thereby better adapting to actual interview interaction scenarios, significantly improving the quality of generated questions and the efficiency and depth of the interview, and ensuring a comprehensive and accurate assessment of candidates.

[0123] This application further proposes that the direction to be examined in the above-mentioned examination space unfolding graph supports multi-level dynamic expansion. The expansion dimensions are adaptively configured based on industry attributes, job level, and interview type. New expansion dimensions are directly associated with the original topology graph nodes, and edge attributes are automatically generated and integrated into the examination space unfolding graph.

[0124] Specifically, the assessment space expansion diagram supports multi-level dynamic expansion of the assessment directions. This means that the interview assessment space is not a flat or fixed-level structure, but can be further refined and deepened based on the identified assessment directions as needed, forming deeper assessment dimensions. For example, an assessment direction of "communication skills" can be further expanded into sub-directions such as "listening skills," "expression skills," and "influence." Each sub-direction can be further subdivided, thereby ensuring the depth and breadth of the interview and avoiding omission of key assessment points. This multi-level expansion can be represented by data structures such as tree structures, graph databases, or multi-dimensional arrays. When the system identifies that a certain assessment direction requires deeper exploration, it will trigger the expansion mechanism to generate new child nodes or sub-dimensions under that direction.

[0125] Meanwhile, the expanded dimensions are adaptively configured based on industry attributes, job levels, and interview types, emphasizing the intelligence and personalization of the expansion process. Different industries, job levels, and interview types have drastically different focuses and depths for assessing candidates. Adaptive configuration means that the system can automatically select or recommend the most suitable expanded dimensions based on the current interview scenario, ensuring the relevance and effectiveness of the generated questions. The system can maintain a dimension configuration rule base, associating contextual information such as industry attributes, job levels, and interview types with a predefined set of expanded dimensions. For example, for "IT industry - senior engineer - technical interview," the system might prioritize expanding dimensions such as "algorithm complexity" and "system architecture design"; while for "finance industry - management position - behavioral interview," it might expand dimensions such as "risk control awareness" and "team leadership." These configurations can be preset templates or dynamically learned and optimized based on historical interview data through machine learning models.

[0126] Furthermore, newly added extended dimensions are directly associated with existing nodes in the topology graph, automatically generating edge attributes and integrating them into the unfolded graph of the examination space. This means that when a new extended dimension is determined, it does not exist in isolation, but rather as a new node or a child node of an existing node, connecting with other relevant nodes in the graph (such as its parent direction to be examined, related interview objectives, etc.) through edges. These edges not only represent connection relationships but also carry edge attributes, such as association strength, extension priority, and logical dependencies, thereby enriching the semantic information of the unfolded graph of the examination space. When the system generates a new extended dimension, it automatically creates a new node in the data structure of the unfolded graph of the examination space based on its source and its relationship with the interview objectives, and establishes connections with existing nodes. Edge attributes can be generated based on preset rules or calculated by algorithms.

[0127] Through the aforementioned technical solution, the areas to be examined in the spatial unfolding graph can support multi-level dynamic expansion and adaptive configuration based on industry attributes, job levels, and interview types. This allows the divergent generation method for interview questions to flexibly deepen and refine the examination dimensions according to specific interview scenarios, avoiding the limitations of fixed dimensions. The newly added expanded dimensions can directly establish associations with existing topological graph nodes and automatically generate edge attributes, ensuring the dynamic updating and semantic integrity of the spatial unfolding graph. This dynamic and adaptive expansion mechanism significantly improves the relevance, comprehensiveness, and depth of interview question generation, enabling the system to more accurately capture candidates' true abilities in different contexts. This provides interviewers with more insightful divergent questions, effectively improving the accuracy and effectiveness of interview assessments.

[0128] like Figure 4As shown in the figure, this application also discloses an interview question divergent generation system, including an interactive data acquisition module, a path coverage analysis module, an examination space construction module, a candidate question generation module, a coverage compensation evaluation module, and an output update module.

[0129] The interactive data acquisition module is used to acquire the current question, current answer data, interview assessment target data, and auxiliary reference data, and to complete the data preprocessing and standardization.

[0130] The path coverage analysis module is used to analyze the dimensions of investigation and identify the investigation directions that have been opened, covered, and not yet developed.

[0131] The assessment space construction module is used to build and dynamically update the unfolded graph of the interview assessment space, and maintain the nodes, associated edges and coverage status;

[0132] The candidate question generation module is used to generate multiple sets of candidate divergent questions based on the direction to be investigated and the investigation objectives.

[0133] The coverage compensation assessment module is used to filter the optimal divergent problem based on rules such as reachability, gap compensation, low overlap, and stability improvement.

[0134] The output update module is used to output the optimal divergent problem and complete the association and binding between the problem and the unfolded graph, as well as the update of the unfolded graph coverage status.

[0135] The following example will provide a more detailed explanation of the above technical solution:

[0136] In a technical interview scenario for a "Senior Java Development Engineer" position, the system needs to assist the interviewer in generating in-depth and targeted divergent questions.

[0137] First, during the data acquisition phase, the system receives questions from the current interview interaction, such as, "Please describe the biggest technical challenge you encountered in your most recent project and how you solved it?" Simultaneously, the system acquires the candidate's current answer to this question, such as, "I encountered a distributed transaction consistency problem in my project, which I successfully solved by introducing the Seata framework and combining it with the TCC pattern." During this process, the system also loads interview assessment data for the position (such as assessing the candidate's complex system design ability, problem-solving ability, technical depth, and framework application experience), the job profile (such as requirements for proficiency in microservice architecture, distributed systems, high concurrency handling, and familiarity with Spring Cloud, Dubbo, Seata, etc.), and supplementary reference data regarding the candidate's background (such as 5 years of Java development experience and participation in multiple large-scale distributed projects as shown in the resume). This supplementary reference data, including core data from the job profile, key data from the candidate's background, and core interview assessment requirements, is modeled using structured modeling and vector embedding techniques to form a standardized basic feature vector library, providing feature support for subsequent assessment direction identification and question generation. The current question and the current answer data are preprocessed using natural language understanding technology, including word segmentation, semantic role labeling, and assessment dimension mapping, to transform unstructured text into structured assessment dimensions.

[0138] Next, the system performs dimensional analysis and path coverage analysis on the current question and answer data. Based on a preset assessment dimension library, the system matches the potential assessment directions opened up by the current question's "greatest technical challenge," such as "problem-solving ability," "technical depth," "project experience," and "framework application." Subsequently, the system extracts the semantic features of the current answer data, such as "distributed transaction consistency issues, Seata framework, TCC mode," and matches them with potential assessment directions, marking the actually covered assessment directions, such as "distributed system experience," "Seata framework application," and "problem-solving ability (partial)." Through difference calculation, the system obtains the assessment directions that have not yet been explored in the current interview, such as "high concurrency processing experience," "depth of microservice architecture design," "understanding of the underlying principles of the Seata framework," and "experience in applying other distributed components." The system also marks the matching degree and expansion priority of each assessment direction with the interview assessment objectives. For example, "high concurrency processing experience" has a high matching degree with the job profile of "senior Java development engineer" and is given a higher expansion priority.

[0139] Subsequently, based on the interview assessment target data and path coverage analysis results, the system constructs an interview assessment space expansion diagram. This expansion diagram is a dynamically updatable structured representation carrier, whose core components include the current question node ("Biggest Technical Challenge"), the current answer node ("Distributed Transaction Consistency, Seata"), covered direction nodes ("Seata Framework Application", "Distributed System Experience"), to-be-assessed direction nodes ("High Concurrency Processing Experience", "Microservice Architecture Design Depth", "Seata Underlying Principles"), and direction-related edges. These nodes are used to carry direction names, coverage status, matching degree, and priority information, while direction-related edges are used to represent the relationships and expansion paths between dimensions. For example, "Seata Framework Application" can be expanded to "Seata Underlying Principles". This expansion diagram supports dynamic updates such as adding and deleting nodes, modifying status, and adjusting relationships. Compared with the static and unrelated question sets in existing technologies, it provides a more systematic assessment view.

[0140] Then, the system uses the areas to be examined in the spatial expansion diagram as anchor points, and generates multiple sets of candidate divergent questions based on the interview assessment target data. For example, using "experience in high-concurrency processing" as the anchor point, candidate question A is generated: "When designing a high-concurrency system, what technical solutions and challenges do you usually consider? Please give examples."; using "depth of microservice architecture design" as the anchor point, candidate question B is generated: "Besides distributed transactions, what other complex problems have you encountered in microservice architecture design, and how did you solve them?"; using "Seata underlying principles" as the anchor point, candidate question C is generated: "What is your understanding of the transaction coordination mechanism and data consistency guarantee principle of the Seata framework?" These generated candidate divergent questions are strongly bound to the areas to be examined, and each set of candidate questions serves to achieve the interview assessment target. It supports the generation of questions in combination of single or multiple areas to be examined, avoiding the shortcomings of existing technologies in terms of lack of core element anchoring, low targeting and accuracy in question divergence.

[0141] Next, the system performs targeted evaluation on the candidate divergent problems. This evaluation is a coverage-compensation-oriented decision-making and screening process, and the evaluation logic focuses on:

[0142] 1. Reach determination: Can candidate question A directly reach the unexplored direction of "experience in handling high concurrency"?

[0143] 2. Gap Compensation Determination: Can candidate question A fill the coverage gap of "high-concurrency processing" in the current examination path?

[0144] 3. Overlap Determination: Whether the semantic overlap between candidate question A and already asked questions or covered directions (such as "distributed transactions") is low.

[0145] 4. Stability assessment: Can candidate question A provide new and valid evidence for the interview assessment objective (such as the ability to design complex systems) and improve the stability of the assessment?

[0146] The system employs a priority-based selection rule: Candidate questions that can reach the core area of ​​investigation (such as "high-concurrency processing experience") are prioritized. Based on meeting the reach requirements, the candidate question with the best coverage gap compensation capability is selected. If both the first two conditions are met, the candidate question with the lowest overlap with already covered paths and the most significant improvement in the stability of the evidence for the investigation target is selected as the optimal divergence question. Through this systematic evaluation, the system determines candidate question A as the optimal divergence question, avoiding the problems of existing technologies where problem divergence lacks unified deduction rules and suffers from insufficient logic and systematicity.

[0147] Finally, the system outputs the optimal divergent problem, "What technical solutions and challenges do you typically consider when designing high-concurrency systems? Please provide examples." and associates it with the expansion path and direction of the investigation space unfolding graph. Simultaneously, the system updates the coverage status information of the investigation space unfolding graph, marking "high-concurrency processing experience" as covered or partially covered.

[0148] Furthermore, the system includes dynamic iteration and self-optimization steps for the assessment space unfolding graph. During subsequent interview interactions, the system collects feedback data from interviewers' follow-up questions, candidates' answers, and interviewers' evaluations. Based on this feedback data, the system dynamically adjusts the node information and directional relationships of the assessment space unfolding graph. For example, based on the interviewer's further follow-up questions about "high-concurrency processing experience," the system refines the sub-dimensions under that direction. Simultaneously, the system optimizes the candidate question generation rules and targeted evaluation judgment rules, making the divergent generation logic more adaptable to actual interview interaction scenarios. This solves the problems of existing static generation systems lacking dynamic iterative optimization capabilities and lagging scenario adaptability. The assessment space unfolding graph supports multi-level dynamic expansion of the directions to be assessed. The expansion dimensions are adaptively configured based on industry attributes (such as Internet), job level (such as senior), and interview type (such as technical interview). Newly added expansion dimensions are directly associated with the original topology graph nodes, automatically generating edge attributes and integrating them into the assessment space unfolding graph. This further enhances the scalability and hierarchy of question generation, making up for the lack of a systematic divergent dimension system in existing technical question generation.

[0149] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for generating interview questions in a divergent manner, characterized in that, Includes the following steps: S1. Obtain the current question asked during the interview interaction and the candidate's current answer data, and load the interview assessment target data, job profile, and candidate background auxiliary reference data; S2. Perform dimensional analysis and path coverage analysis on the current question and the current answer data to identify the examination directions that the current question has opened up and the examination directions that the current answer has covered, and mark the examination directions that have not yet been developed. S3. Based on the data of the interview assessment targets and the path coverage analysis results, construct an interview assessment space expansion diagram to represent the coverage status, dimensional relationships and divergent expansion paths of the assessment directions in a structured form. S4. Using the direction to be examined in the examination space expansion diagram as the expansion anchor point, generate multiple sets of candidate divergent questions in combination with the interview examination target data. S5. Perform targeted evaluation on candidate divergent questions. The core judgment dimensions include: whether it can reach the direction that has not been explored, whether it can compensate for the coverage gap of the current investigation path, whether it can avoid high overlap with the already covered path, and whether it can improve the evidence stability of the investigation target. Based on the judgment results, determine the optimal divergent question. S6. Output the optimal divergent problem and associate it with the expansion path and direction to be investigated in the exploration space unfolding graph, and update the coverage status information of the exploration space unfolding graph synchronously.

2. The method for generating interview questions according to claim 1, characterized in that: The auxiliary reference data in S1 includes at least core data of job profile, key data of candidate background, and core interview assessment requirements. The basic reference data is modeled using structured modeling and vector embedding techniques to form a standardized basic feature vector library, which provides feature support for subsequent assessment direction identification and question generation. Both the current question and the current answer data are preprocessed using natural language understanding technology. The preprocessing includes word segmentation, semantic role labeling, and assessment dimension mapping to transform unstructured text into structured assessment dimensions.

3. The method for generating interview questions according to claim 1, characterized in that: The path coverage analysis in S2 is performed based on a preset examination dimension library, specifically including: Match the current question with the corresponding assessment dimensions to determine all potential assessment directions opened up by the question; Extract semantic features from the current answer data, match them with potential examination directions, and identify the actual examination directions covered; The unexpanded directions to be examined are obtained by difference calculation, and the matching degree and expansion priority of each direction to be examined are marked with the interview assessment objectives.

4. The method for generating interview questions according to claim 1, characterized in that: The interview assessment space unfolding diagram in S3 is a dynamically updatable structured representation carrier, whose core components include at least: current question node, current answer node, covered direction node, direction node to be assessed, direction-related edge, and candidate divergent question node. The nodes are used to carry direction name, coverage status, matching degree, and priority information. The direction association edges are used to represent the association relationship and expansion path between dimensions. The unfolded graph supports dynamic updates of node addition, deletion, status modification, and association relationship adjustment.

5. The method for generating interview questions according to claim 1, characterized in that: In S4, multiple sets of candidate divergent problems are generated, specifically: Based on the node information of the direction to be examined in the spatial development diagram and the data of the interview assessment targets, it is generated in accordance with the anchoring rules of the direction to be examined. The generated candidate divergent questions are strongly tied to the direction to be examined, and each group of candidate questions serves to achieve the interview assessment objectives. It supports the generation of combined questions for a single direction to be examined or multiple directions to be examined.

6. The method for generating interview questions according to claim 1, characterized in that: The targeted assessment in S5 is a coverage-compensation-oriented decision-making and screening process, and the assessment logic focuses on: Reach determination: Can the candidate question directly reach at least one unexplored direction to be investigated? Gap compensation determination: Can the candidate problem fill the coverage gap of the current investigation path? Overlap determination: Whether the candidate question has low semantic overlap with the already asked question and the covered directions; Stability assessment: Can the candidate questions provide new and valid evidence for the interview assessment objectives, thereby improving the stability of the evaluation? 7. The method for generating interview questions according to claim 6, characterized in that: The aforementioned targeted assessment employs a priority screening rule: Prioritize candidate questions that can reach the core areas to be investigated; Based on meeting the reach requirements, the candidate problem with the best coverage gap compensation capability is selected; When both of the above conditions are met, the candidate problem with the lowest overlap with the already covered path and the most significant improvement in the stability of the evidence for the target being examined is selected as the optimal divergent problem.

8. The method for generating interview questions according to claim 1, characterized in that: It also includes examining the dynamic iteration and self-optimization steps of the spatial unfolding graph: Collect follow-up question feedback, answer feedback and evaluation feedback data during the interview interaction process, and adjust the node information and directional correlation of the assessment space unfolding diagram based on the feedback data; Simultaneously optimize the candidate question generation rules and targeted evaluation and judgment rules to make the divergent generation logic adapt to actual interview interaction scenarios.

9. The method for generating interview questions according to claim 1, characterized in that: The direction to be examined in the examination space unfolding graph supports multi-level dynamic expansion. The expansion dimensions are adaptively configured based on industry attributes, job level, and interview type. New expansion dimensions are directly associated with the original topology graph nodes, and edge attributes are automatically generated and integrated into the examination space unfolding graph.

10. A system for generating divergent interview questions, characterized in that, The system, applied to the method as described in any one of claims 1-9, comprises: The interactive data acquisition module is used to acquire the current question, current answer data, interview assessment target data, and auxiliary reference data, and to complete the data preprocessing and standardization. The path coverage analysis module is used to analyze the dimensions of investigation and identify the investigation directions that have been opened, covered, and not yet developed. The assessment space construction module is used to build and dynamically update the unfolded graph of the interview assessment space, and maintain the nodes, associated edges and coverage status; The candidate question generation module is used to generate multiple sets of candidate divergent questions based on the direction to be investigated and the investigation objectives. The coverage compensation assessment module is used to filter the optimal divergent problem based on rules such as reachability, gap compensation, low overlap, and stability improvement. The output update module is used to output the optimal divergent problem and complete the association and binding between the problem and the unfolded graph, as well as the update of the unfolded graph coverage status.

Images