Job-seeking interview question AI generation method based on large language model

Abstract

The invention discloses a job-hunting interview question AI generation method based on a large language model, and relates to the technical field of human resource management, and the method comprises the following steps: S1, analyzing post information, and extracting core skill points; s2, constructing candidate feature matching vectors based on the core skill points; s3, performing knowledge graph expansion by using the core skill points and candidate feature matching vectors; s4, performing semantic fusion on the basis of the extended knowledge graph to obtain a knowledge association set; s5, performing question generation and difficulty grading by using the knowledge association set; and S6, performing logic consistency detection and diversification generation by using the question set. By setting knowledge graph extension and semantic fusion based on post core skill points and candidate feature matching vectors, a knowledge association set with a wider coverage and a more reasonable structure can be established before topic generation, and it is ensured that the extended knowledge graph not only embodies post requirements but also is matched with candidate features.

Description

Technical Field

[0001] This invention relates to the field of human resource management technology, specifically to an AI-based method for generating job interview questions based on a large language model. Background Technology

[0002] With the rapid development of artificial intelligence technology, companies are increasingly relying on interview questions during the recruitment process. Traditional manual question bank construction models are gradually revealing problems such as poor adaptability, lagging updates, and insufficient intelligence. Existing interview question banks are mostly pre-set manually, making it difficult to tailor questions to the specific needs of the job. They are highly general but lack personalization, and updates to the question bank content often fail to reflect the latest industry developments, resulting in insufficient scientific rigor and relevance of the questions.

[0003] In existing technologies, although some studies have attempted to expand the question bank or perform semantic rewriting using natural language processing methods, there are still significant shortcomings in the dynamic expansion and semantic fusion of knowledge graphs. Question generation lacks job-related knowledge expansion and cross-dimensional semantic fusion, resulting in the final generated questions failing to fully cover the key points of the examination. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an AI-based method for generating job interview questions based on a large language model, in order to solve the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] In a first aspect, embodiments of the present invention provide an AI-based method for generating job interview questions based on a large language model, comprising the following steps:

[0007] S1. Analyze job information and extract core skill points;

[0008] S2. Construct candidate feature matching vectors based on core skill points;

[0009] S3. Expand the knowledge graph using core skill points and candidate feature matching vectors;

[0010] S4. Based on the extended knowledge graph, semantic fusion is performed to obtain a knowledge association set;

[0011] S5. Use the knowledge association set to generate questions and classify their difficulty to obtain a question set;

[0012] S6. Use a set of questions for logical consistency testing and diversified generation.

[0013] To further optimize this technical solution, step S3 in this method is used to map the core job skills points and candidate features to a knowledge graph. By expanding the hierarchical concepts and related nodes related to skills, a potential skill network of the candidate in the knowledge graph is formed, providing basic data for subsequent semantic fusion and question generation. Step S3 includes the following steps:

[0014] Skill node retrieval;

[0015] Weighted expansion;

[0016] Overall knowledge graph combination and node deduplication.

[0017] To further optimize this technical solution, in step S3, when performing skill node retrieval, for each skill point, relevant nodes and adjacent concepts are retrieved in the skill knowledge graph to form a preliminary extended node set;

[0018] Retrieve all directly or indirectly related nodes based on graph path length and node association relationships;

[0019] Use graph search techniques to obtain adjacent nodes and filter nodes with confidence scores higher than a threshold;

[0020] Each node contains attribute fields: skill type, level, and tag.

[0021] To further optimize this technical solution, in step S3, when performing weighted expansion, candidate feature matching values ​​are used. The nodes are weighted to reflect the candidate's potential ability in the skills network, and the calculation formula is as follows:

[0022] ;

[0023] ;

[0024] in:

[0025] The weighted node value represents the potential matching degree of the candidate features on this expanded node;

[0026] Skill points The One extended node;

[0027] For nodes With skill points semantic similarity;

[0028] The set of extended skill nodes after weighting the candidates.

[0029] To further optimize this technical solution, in step S3, when performing overall knowledge graph combination and node deduplication, all skill point extension sets are combined to form an overall extended knowledge graph:

[0030] ;

[0031] Process duplicate nodes:

[0032] If multiple skill point expansion sets contain duplicate nodes, then retain the nodes and their weights. Take the maximum value or weighted average value, and retain the node attribute information.

[0033] To further optimize this technical solution, step S4 in this method is used to semantically fuse the expanded skill nodes, making the nodes semantically coherent, and to aggregate them by combining candidate feature weights to generate a knowledge association set. Specifically, it includes the following steps:

[0034] Node vector representation and semantic similarity calculation;

[0035] Candidate weighted clustering;

[0036] Node weight fusion;

[0037] Knowledge association set generation.

[0038] To further optimize this technical solution, in step S4, when performing node vector representation and semantic similarity calculation, the extended nodes... Generate semantic vectors The vector generation method uses graph embedding, and the vector dimensions are unified and normalized to [0,1].

[0039] And calculate the pairwise semantic similarity of nodes:

[0040] ;

[0041] Similarity measures the semantic proximity between nodes and is used for subsequent clustering selection.

[0042] To further optimize this technical solution, in step S4, when performing weighted clustering of candidates, nodes with similar semantics and similar candidate matching values ​​are clustered into a set. :

[0043] ;

[0044] and These are semantic similarity and matching value thresholds, determined through existing clustering experience or cross-validation.

[0045] Clustering algorithms can employ hierarchical clustering, K-means, or density-based clustering.

[0046] Each collection It includes nodes with similar semantics and similar candidate matching weights.

[0047] To further optimize this technical solution, in step S4, when performing node weight fusion, for each cluster node set The weighted average matching value is calculated using the following formula:

[0048] ;

[0049] After weighted average, This indicates the overall skill matching degree of the candidates corresponding to the set of fusion nodes;

[0050] Outlier or missing value nodes are removed or compensated to ensure that the output weights are effective.

[0051] To further optimize this technical solution, in step S4, when generating the knowledge association set, all clustered node sets are... With corresponding weights Combined to form the final knowledge association set:

[0052] .

[0053] Secondly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the job interview question AI generation method based on a large language model as described in the first aspect of the present invention.

[0054] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the job interview question AI generation method based on a large language model as described in the first aspect of the present invention.

[0055] Compared with existing technologies, this invention provides an AI-based method for generating job interview questions based on a large language model, which has the following beneficial effects:

[0056] This AI-powered job interview question generation method, based on a large language model, establishes a broader and more structurally sound knowledge graph set before question generation by expanding and semantically fusing knowledge graphs based on core job skills and candidate feature matching vectors. During the knowledge graph expansion stage, the system not only introduces job-related hierarchical skills but also adds knowledge nodes closely related to the candidate's background, guided by candidate feature vectors. This ensures that the expanded knowledge graph reflects both job requirements and candidate characteristics. The semantic fusion process then unifies these expanded nodes with job skills and candidate features into the same semantic space, maintaining semantic consistency and computability across different sources. This allows the question generation stage to directly utilize this knowledge graph set to construct logically sound, targeted, and differentiated questions. Compared to existing technologies, this method not only solves the problems of incomplete question coverage and insufficient logical connections but also improves the questions' job specificity and candidate suitability, making the generated interview questions more scientific and practical. Attached Figure Description

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart illustrating an AI-based method for generating job interview questions based on a large language model, as proposed in this invention.

[0059] Figure 2 This is a schematic diagram illustrating the process of constructing candidate feature matching vectors in an AI-based job interview question generation method based on a large language model proposed in this invention.

[0060] Figure 3 This is a schematic diagram illustrating the knowledge graph expansion process of an AI-based job interview question generation method based on a large language model proposed in this invention.

[0061] Figure 4 This is a schematic diagram of the semantic fusion process of an AI-based job interview question generation method based on a large language model proposed in this invention.

[0062] Figure 5 This diagram illustrates the logical consistency detection and diversified generation process of an AI-based job interview question generation method based on a large language model proposed in this invention. Detailed Implementation

[0063] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0064] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0065] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0066] Example 1:

[0067] Reference Figures 1-5 This is the first embodiment of the present invention, which provides an AI-based method for generating job interview questions based on a large language model, including the following steps:

[0068] S1. Analyze job information and extract core skill points;

[0069] Step S1 in this method is used to receive relevant information about the job posting, including text data such as job responsibilities, required skills, and preferred qualifications. Step S1 is executed according to the following process:

[0070] Text semantic analysis:

[0071] Using mature Natural Language Processing (NLP) technology, the job description text is segmented, part-of-speech tagging is performed, and named entity recognition (NER) is performed; key content such as skills, tools, tasks, and industry terms in the job description are extracted to form a preliminary set of skill candidates.

[0072] Using existing NLP models (such as BERT or word embedding models), a vector representation is generated for each term, and preliminary clustering and classification are performed based on semantic similarity to obtain a set of skill categories.

[0073] Skill node mapping and structured transformation:

[0074] The skill terms extracted in the previous process are matched with the existing job skill knowledge graph; through mature knowledge graph retrieval and semantic matching technologies, the text terms are mapped to standardized skill nodes in the knowledge graph, and the original text descriptions are transformed into structured skill items.

[0075] For example, "extensive experience in Python programming" is mapped to the skill node "Python programming", while also recording the skill type and related assessment dimensions (such as programming ability and project application ability).

[0076] Core skill points determined:

[0077] The mapped skill nodes are weighted and summarized to select the most critical skill points for the position, forming a set of core skill points for the position. Skill priorities are determined by mature weight calculation methods (such as weighting methods based on frequency of occurrence or importance indicators of job responsibilities), and the selection results are transformed into a set of skill points that can be directly used in the next step.

[0078] S2. Construct candidate feature matching vectors based on core skill points;

[0079] Step S2 in this method uses the core job skills points obtained in step S1 as a reference framework to perform targeted analysis and structured modeling of the information in the candidate's resume, forming a candidate feature matching vector. Step S2 includes the following steps:

[0080] Skills Points - Resume Information Mapping:

[0081] Step S2 uses the set of core job skills obtained in step S1. and candidate resume text Using named entity recognition (NER), text vectorization (such as BERT vectors), and keyword matching techniques, descriptions of job skills in the resume are extracted and mapped to corresponding skill points. The above yields a preliminary set of characteristics for each candidate at each skill level. ;

[0082] Candidate feature quantification:

[0083] Candidate descriptions based on the skill point mapping obtained in the previous step For each skill point, a mature similarity calculation technique (such as cosine similarity) is used to measure the degree of match between the candidate's resume content and the skill point definition, and a matching value is obtained. The calculation formula is as follows:

[0084] ;

[0085] in:

[0086] This indicates the skill points in the candidate's resume. Related vectorized features;

[0087] Indicates skill points Vectorized representation of;

[0088] For candidates in skill points The matching degree is in the range [0,1], and the higher the value, the higher the compatibility.

[0089] Generate candidate feature matching vectors:

[0090] Matching value for each skill point Arranged in order of job skill points, forming a vector:

[0091] ;

[0092] Output This is the candidate feature matching vector.

[0093] Existing technologies typically only extract general features from resumes (education, experience, certificates), which do not have a clear correspondence with job skills.

[0094] Step S2 establishes a targeted matching vector based on job skill points, embedding candidate features into the job skill space to achieve a one-to-one correspondence between job requirements and candidate features. This logical modeling method differs from traditional general feature extraction techniques.

[0095] S3. Expand the knowledge graph using core skill points and candidate feature matching vectors;

[0096] Step S3 in this method is used to map the core job skills to candidate features into a knowledge graph. By expanding the hierarchical concepts and related nodes related to skills, a potential skill network of the candidate in the knowledge graph is formed, providing basic data for subsequent semantic fusion and question generation. Step S3 includes the following steps:

[0097] Skill Node Retrieval:

[0098] For each skill point Retrieve relevant nodes and adjacent concepts from the existing skill knowledge graph to form a preliminary expanded node set. ;

[0099] Retrieve all related nodes based on graph path length and node association. Nodes that are directly or indirectly related;

[0100] Use mature graph search techniques (such as BFS or graph database queries) to obtain adjacent nodes, and filter those with confidence scores higher than a threshold. Nodes;

[0101] Each node contains attribute fields: skill type, level, and tag.

[0102] Weighted expansion:

[0103] Using candidate feature matching values The nodes are weighted to reflect the candidate's potential ability in the skills network, and the calculation formula is as follows:

[0104] ;

[0105] ;

[0106] in:

[0107] The weighted node value represents the potential matching degree of the candidate features on this expanded node;

[0108] Skill points The One extended node;

[0109] For nodes With skill points Semantic similarity (calculated through knowledge graph relationships or word vectors);

[0110] The set of extended skill nodes after weighting the candidates.

[0111] Overall knowledge graph assembly and node deduplication:

[0112] All skill points are expanded and combined to form a holistic expanded knowledge graph:

[0113] ;

[0114] Process duplicate nodes:

[0115] If multiple skill point expansion sets contain duplicate nodes, then retain the nodes and their weights. Take the maximum value or weighted average value, while retaining node attribute information;

[0116] Output This serves as the input for the subsequent semantic fusion step S4.

[0117] Existing knowledge graph expansion methods mostly rely on static node retrieval or preset skill relationships, without incorporating candidate features;

[0118] Step S3 achieves dynamic personalized mapping of the skill network by weighted expansion of candidate matching vectors, making the skill node distribution of each candidate in the knowledge graph highly correlated with its features, and clarifying the node attributes, weight calculation and output structure.

[0119] S4. Based on the extended knowledge graph, semantic fusion is performed to obtain a knowledge association set;

[0120] Step S4 in this method is used to semantically fuse the expanded skill nodes, making the nodes semantically coherent, and then aggregate them by combining candidate feature weights to generate a knowledge association set. Specifically, it includes the following steps:

[0121] Node vector representation and semantic similarity calculation:

[0122] For extended nodes Generate semantic vectors The vector generation method adopts mature language models (such as BERT, GraphSAGE) or graph embedding methods, and the vector dimension is unified and normalized to [0,1].

[0123] Calculate the pairwise semantic similarity of nodes:

[0124] ;

[0125] Similarity measures the semantic proximity between nodes and is used for subsequent clustering selection.

[0126] Candidate weighted clustering:

[0127] Cluster nodes with similar semantics and similar candidate matching values ​​into sets. :

[0128] ;

[0129] and These are semantic similarity and matching value thresholds, determined through existing clustering experience or cross-validation.

[0130] Clustering algorithms can include hierarchical clustering, K-means, or density clustering;

[0131] Each collection It includes nodes with similar semantics and similar candidate matching weights.

[0132] Node weight fusion;

[0133] For each cluster node set The weighted average matching value is calculated using the following formula:

[0134] ;

[0135] After weighted average, This indicates the overall skill matching degree of the candidates corresponding to the set of fusion nodes;

[0136] Outlier or missing value nodes are removed or compensated to ensure that the output weights are effective.

[0137] Knowledge association set generation:

[0138] The set of all clustered nodes With corresponding weights Combined to form the final knowledge association set:

[0139] ;

[0140] Output The data structure is clear, and each element can be used to generate questions.

[0141] Existing technologies typically only perform static semantic aggregation and do not introduce candidate feature weights.

[0142] Step S4 combines candidate matching values ​​with weighted clustering and node fusion to make the knowledge association set both semantically coherent and directly related to candidate features.

[0143] S5. Use the knowledge association set to generate questions and classify their difficulty to obtain a question set;

[0144] Step S5 in this method serves to process the knowledge association set output in step S4. This process transforms the questions into a set of directly usable interview questions, and categorizes the questions by difficulty based on the importance of the skills and their suitability for the candidate. This includes the following steps:

[0145] Question stem generation:

[0146] The knowledge association set output in step S4 For input, leveraging mature large language model text generation technology, each skill set is... The semantic content is transformed into examineable question stem descriptions; for example, using a knowledge graph-based question-answering generation method, the semantic tags of skill nodes and their hierarchical relationships are organized into question-answer or case-based questions; this step yields a set of initial question drafts, with each question corresponding to... A combination of skills.

[0147] Knowledge point coverage check:

[0148] By using keyword matching and semantic similarity calculation, we can detect whether the generated questions cover the relevant content. It covers all key skill points; if any omissions are found, the questions are supplemented or fine-tuned by calling the same large language model to obtain a complete set of questions.

[0149] Difficulty levels:

[0150] Based on the set of questions + candidate matching value As input, questions are categorized using established difficulty prediction models (such as those based on ItemResponse Theory, IRT, or deep learning-based question difficulty prediction models). The difficulty is assessed based on factors including skill level, logical complexity of the question, and candidate fit. For example: when When the value is high, more complex questions are generated to differentiate candidates' abilities; when... When the value is low, medium to low difficulty questions are generated to ensure basic assessment. This step yields a set of questions and corresponding difficulty labels (such as basic, intermediate, and advanced).

[0151] Step S5, after the above steps, finally outputs a set of questions labeled with difficulty levels. ,in For the generated questions, Difficulty level.

[0152] S6. Use a question set for logical consistency testing and diversified generation;

[0153] In this scheme, step S6 is used to ensure that the question set output by step S5 is logically complete and consistent, while avoiding question homogenization through diversified generation mechanisms. This ensures that the final output question set satisfies both the fairness of the assessment and covers the characteristic differences of different candidates. Step S6 includes two steps: logical consistency detection and diversified generation. The logical consistency detection step includes the following process:

[0154] First, a logical dependency rule base is constructed, based on knowledge association sets. The collection of knowledge points in and their weights Establish logical rules:

[0155] like and If there is a sequential dependency relationship, then it involves The topic must involve The question will appear after the topic;

[0156] like and If they are mutually exclusive, they must not appear in the same question or the same question group at the same time;

[0157] Weight Here, the higher the value, the higher the priority of the rule in the rule determination. The rule with the higher value will be given priority in conflict resolution.

[0158] Then, calculate the mapping relationship between questions and knowledge points, and use the question tags from step S5 to clarify each question. Corresponding knowledge point subset ;

[0159] By matching and The dependency rules in the code check for order contradictions or mutual exclusion contradictions.

[0160] Finally, consistency checks and filtering are performed; questions with contradictions are removed. Based on the weight of the knowledge points involved Prioritize and retain questions with high weight, while removing or adjusting questions with low weight.

[0161] Output a set of logically consistent questions .

[0162] The diversified generation process includes the following steps:

[0163] First, candidate rewrites are generated, targeting the consistent set. For each question Generate a candidate rewrite set The methods include:

[0164] Based on synonym replacement and grammatical structure transformation (a mature NLP rewriting method);

[0165] Diverse generation based on pre-trained language models (e.g., generating multiple versions by controlling decoding parameters);

[0166] Question type transformation based on human-generated templates (multiple choice questions ↔ fill-in-the-blank questions, etc.).

[0167] Then, coverage calculation is performed. Calculate the subset of knowledge points covered by each candidate question. ;

[0168] Require With the original question Corresponding knowledge point subset Maintain consistency to ensure that rewriting the questions does not change the knowledge assessment objectives.

[0169] Then, a difference measurement was performed on the candidate topics. With the original question Calculate the difference index Edit distance or semantic embedding cosine distance can be used;

[0170] The difference must be higher than a set threshold (e.g., 0.3–0.4) to ensure that the generated questions have substantial differences in expression.

[0171] Finally, candidate screening and optimization are performed, and a comprehensive scoring function is set:

[0172] ,

[0173] in , For experience weights (e.g.) =0.6, =0.4);

[0174] from Select the candidates with the highest scores to form the final output set.

[0175] Unlike existing techniques that simply rewrite questions or clean test banks, this step involves associating knowledge sets. Introducing a consistency check process, utilizing knowledge point weights Control the priority of conflict handling, and ensure the quality and rationality of generated questions through a dual-indicator screening mechanism of "coverage + difference".

[0176] Example 2:

[0177] This embodiment also provides a computer device applicable to a job interview question AI generation method based on a large language model, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the job interview question AI generation method based on a large language model as proposed in the above embodiment.

[0178] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements an AI method for generating job interview questions based on a large language model as proposed in the above embodiment.

[0179] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0180] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0181] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0182] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0183] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0184] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for generating job interview questions using an AI based on a large language model, characterized in that, Includes the following steps: S1. Analyze job information and extract core skill points; S2. Construct candidate feature matching vectors based on core skill points; S3. Expand the knowledge graph using core skill points and candidate feature matching vectors; S4. Based on the extended knowledge graph, semantic fusion is performed to obtain a knowledge association set; S5. Use the knowledge association set to generate questions and classify their difficulty to obtain a question set; S6. Use a set of questions for logical consistency testing and diversified generation.

2. The AI-generated job interview questions based on a large language model according to claim 1, characterized in that, In this method, step S3 is used to map the core job skills to candidate features onto a knowledge graph. By expanding the hierarchical concepts and related nodes related to skills, a potential skill network of the candidate in the knowledge graph is formed, providing basic data for subsequent semantic fusion and question generation. Step S3 includes the following steps: Skill node retrieval; Weighted expansion; Overall knowledge graph combination and node deduplication.

3. The AI-generated job interview questions based on a large language model according to claim 2, characterized in that, In step S3, when retrieving skill nodes, for each skill point, relevant nodes and adjacent concepts are retrieved in the skill knowledge graph to form a preliminary extended node set. Retrieve all directly or indirectly related nodes based on graph path length and node association relationships; Use graph search techniques to obtain adjacent nodes and filter nodes with confidence scores higher than a threshold; Each node contains attribute fields: skill type, level, and tag.

4. The AI-generated job interview questions based on a large language model according to claim 2, characterized in that, In step S3, when performing weighted expansion, candidate feature matching values ​​are used. The nodes are weighted to reflect the candidate's potential ability in the skills network, and the calculation formula is as follows: ； ； in: The weighted node value represents the potential matching degree of the candidate features on this expanded node; Skill points The One extended node; For nodes With skill points semantic similarity; The set of extended skill nodes after weighting the candidates.

5. The AI-generated job interview questions based on a large language model according to claim 2, characterized in that, In step S3, when performing overall knowledge graph combination and node deduplication, all skill point expansion sets are combined to form an overall expanded knowledge graph: ； Process duplicate nodes: If multiple skill point expansion sets contain duplicate nodes, then retain the nodes and their weights. Take the maximum value or weighted average value, and retain the node attribute information.

6. The AI-generated job interview questions based on a large language model according to claim 1, characterized in that, Step S4 in this method is used to semantically fuse the expanded skill nodes, making the nodes semantically coherent, and to aggregate them by combining candidate feature weights to generate a knowledge association set. Specifically, it includes the following steps: Node vector representation and semantic similarity calculation; Candidate weighted clustering; Node weight fusion; Knowledge association set generation.

7. The AI-generated job interview questions based on a large language model according to claim 6, characterized in that, In step S4, when performing node vector representation and semantic similarity calculation, the extended nodes... Generate semantic vectors The vector generation method uses graph embedding, and the vector dimensions are unified and normalized to [0,1]. And calculate the pairwise semantic similarity of nodes: ； Similarity measures the semantic proximity between nodes and is used for subsequent clustering selection.

8. The AI-generated job interview questions based on a large language model according to claim 6, characterized in that, In step S4, when performing candidate weighted clustering, nodes with similar semantics and similar candidate matching values ​​are clustered into sets. : ； and These are semantic similarity and matching value thresholds, determined through existing clustering experience or cross-validation. Clustering algorithms can employ hierarchical clustering, K-means, or density-based clustering. Each collection It includes nodes with similar semantics and similar candidate matching weights.

9. The AI-generated job interview questions based on a large language model according to claim 6, characterized in that, In step S4, when performing node weight fusion, each cluster node set... The weighted average matching value is calculated using the following formula: ； After weighted average, This indicates the overall skill matching degree of the candidates corresponding to the set of fusion nodes; Outlier or missing value nodes are removed or compensated to ensure that the output weights are effective.

10. The AI-generated job interview questions based on a large language model according to claim 1, characterized in that, In step S4, when generating the knowledge association set, all clustered node sets are... With corresponding weights Combined to form the final knowledge association set: 。

Images