High-precision AI surface test question generation method based on self-distillation and industry knowledge base

Abstract

The invention belongs to the technical field of automatic generation of face test questions, and particularly relates to a high-precision AI face test question generation method based on self-distillation and an industry knowledge base. According to the method, by dynamically adjusting the question difficulty, the actual ability of an interviewer can be reflected more accurately, the efficiency and accuracy of the interview process are improved, meanwhile, based on self-distillation and industry knowledge base-based interview question generation, it is ensured that interview questions are highly matched with post requirements, subjective evaluation errors are reduced, reliable data support is provided for recruitment decision making, and the method is suitable for popularization and application. According to the method, the intellectualization and individuation of the interview process are realized, and in addition, the difficulty and content of questions can be flexibly customized for different post characteristics, so that the pertinence and distinction degree of interview questions are further improved, the interview process of an interviewer is more efficient, the ability of the interviewer is accurately evaluated, and the recruitment quality of an enterprise is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic generation of interview questions, and specifically relates to a high-precision AI interview question generation method based on self-distillation and industry knowledge base. Background Art

[0002] As companies' requirements for talent selection become increasingly accurate, traditional interview methods can no longer meet the needs for efficiency and accuracy. With the rapid development of AI technology, it has become possible to use AI technology to generate interview questions. This not only improves interview efficiency, but also accurately matches job requirements. It can also dynamically adjust the difficulty of questions to achieve personalized customization of the interview process, ensuring that each interviewee can demonstrate their true ability at the most suitable difficulty level, thereby selecting the most suitable talent for the company.

[0003] In the existing technology, AI interview question generation mostly relies on static databases and lacks real-time updates and cross-modal matching, resulting in a low match between questions and job requirements, making it difficult to accurately assess the interviewee's ability and adapt to dynamic changes in the industry. In addition, there is a lack of dynamic adjustment mechanism during the interview process, and the difficulty of question optimization in real time based on the interviewee's performance, which affects the interview effect. Based on this, this solution provides a high-precision AI interview question generation method based on self-distillation and industry knowledge base to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a high-precision AI interview question generation method based on self-distillation and industry knowledge base. It can continuously optimize the model through self-distillation technology, combined with a real-time updated industry knowledge base, to achieve high-precision matching of questions and job requirements. The dynamic adjustment mechanism ensures the personalization of the interview process and improves the accuracy of the assessment.

[0005] The technical solutions adopted by the present invention are as follows: High-precision AI interview question generation method based on self-distillation and industry knowledge base, including: Acquire industry technical documents and job description data in real time, extract technical entities and their relationships, and build a dynamic knowledge graph with timestamps; Build a hybrid framework of contrastive learning and self-distillation, generate multiple candidate interview questions through contrastive learning, and use the self-distillation mechanism to cross-modally match technical entities with job requirements; Analyze job requirements into a multi-level classification system, refine the matching granularity, and then calculate the knowledge point coverage based on preset weight rules; Based on the knowledge point coverage and job matching, the optimal interview questions are selected from multiple candidate interview questions, and the corresponding question analysis and scoring criteria are generated simultaneously; The interviewee's answer speed and accuracy for each interview question are collected during the interview process, and the difficulty of subsequent adjacent interview questions is dynamically adjusted until the end of the interview.

[0006] In a preferred embodiment, the steps of acquiring industry technical documents and job description data in real time, extracting technical entities and their relationships, and constructing a dynamic knowledge graph with a timestamp include: Build a relationship extraction network based on the attention mechanism to identify and extract the dependencies, version relationships, and scenario application relationships between technical entities, and annotate them with timestamps; Dynamically update the edge weights of the knowledge graph using a preset time decay function; Graph neural networks are used for cross-document entity alignment. Similarity calculations are used to eliminate naming ambiguities between data sources, and unified entity identifiers are recommended. Map the skill requirements in the job description into a graph query subgraph, and generate a job capability profile through subgraph matching.

[0007] In a preferred embodiment, the step of constructing a hybrid framework of contrastive learning and self-distillation and generating multiple candidate interview questions through contrastive learning includes: Based on the technical entities and relationships in the dynamic knowledge graph, a multi-head attention mechanism is used to generate question trunks and interference items, and negative sampling technology is used to construct comparison samples; Output the corresponding technical points priority based on the job capability profile as a reference standard to assist in cross-matching knowledge points with job requirements; Based on the priority of technical points and combined with the self-distillation mechanism, the matching degree between the question trunk and interference items is optimized to generate multiple candidate interview questions.

[0008] In a preferred solution, after the candidate interview questions are output, the job competency profile and the knowledge points of the candidate interview questions are overlapped to perform an overlap test to determine the degree of fit between the interview questions and the job requirements. The specific process is as follows: Extract the core skill nodes and associated weights in the job capability profile from the knowledge graph; Perform reverse analysis on each candidate interview question to extract the set of technical entities involved and the topological relationships between them; The subgraph isomorphism method is used to detect the common connected subgraphs of the candidate knowledge point coverage subgraph and the job competency profile subgraph, and to identify overlapping knowledge points and their weight distribution; Based on the timeliness annotation of technical entities, the weights of overlapping knowledge points in the public connected subgraph are modified by applying the time decay coefficient; Calculate the sum of the weights of the corrected overlapping knowledge points and compare it with the total weight of the job competency profile to generate a fit score.

[0009] In a preferred embodiment, the steps of analyzing job requirements into a multi-level classification system, refining the matching granularity, and then calculating the knowledge point coverage based on preset weight rules include: Through natural language processing, the interviewee's job requirements are broken down into a three-level classification system, including the first-level technical field, the second-level technology stack, and the third-level skill points; Assign initial weights to the first-level technology areas, second-level technology stacks, and third-level skill points; Based on the association strength and time decay coefficient of technical entities in the knowledge graph, the initial weights of the first-level technical field, the second-level technology stack, and the third-level skill points are dynamically revised; Traverse the topological structure of knowledge points in the candidate interview questions, and cumulatively match them to the first-level technical field, second-level technology stack, and third-level skill points after the revised weights, and calculate the knowledge point coverage; Among them, when the knowledge point coverage is lower than the preset threshold, the cross-level weight compensation mechanism is triggered, and the weight is transferred from the high-weight level to the low-weight level until the overall coverage meets the standard.

[0010] In a preferred embodiment, the step of selecting the best interview question from a plurality of candidate interview questions based on knowledge point coverage and job matching includes: Conduct preliminary screening of candidate interview questions based on the pre-set dual-threshold screening criteria of knowledge point coverage and fit score; After the initial screening, simulate the coverage trajectory of different question combinations on the knowledge graph, select the question combination with the highest cumulative knowledge coverage, and output it as the optimal interview question combination; When there are multiple combinations of optimal interview questions, the timeliness enhancement mechanism is triggered, and question combinations with high timeliness are given priority based on the timeliness marking of technical entities.

[0011] In a preferred solution, after the optimal interview question combination is output, the weight distribution of technical entities in the knowledge graph is dynamically adjusted based on historical interview data feedback. The specific process is as follows: Collect historical interview data, including the accuracy rate, average time spent answering questions, and exposure frequency of related questions for each technical entity, and build a three-dimensional feedback indicator system; Generate dynamic attenuation coefficients of technical entities based on the three-dimensional feedback indicator system; The weights of technical entities in the knowledge graph are migrated based on the dynamic attenuation coefficient, and the knowledge graph is updated in real time.

[0012] In a preferred embodiment, the step of collecting the interviewee's answering speed and accuracy for each interview question during the interview and dynamically adjusting the difficulty of subsequent adjacent interview questions includes: Calculate the interviewee's answering speed and accuracy in real time after they start answering questions, and record them as the first adjustment vector and the second adjustment vector respectively; Performing a time series analysis on the first adjustment vector and the second adjustment vector to predict an optimal difficulty curve for subsequent interview questions; Based on the predicted optimal difficulty curve, the difficulty coefficient of subsequent interview questions is dynamically adjusted. After each adjustment, the adjustment effect is verified in real time through the feedback mechanism to ensure that the difficulty of the adjusted questions matches the interviewee's ability.

[0013] The present invention also provides a high-precision AI interview question generation system based on self-distillation and industry knowledge base, using the above-mentioned high-precision AI interview question generation method based on self-distillation and industry knowledge base, including: A graph construction module is used to obtain industry technical documents and job description data in real time, extract technical entities and their relationships, and build a dynamic knowledge graph with timestamps; The cross-modal matching module is used to build a hybrid framework of contrastive learning and self-distillation. It generates multiple candidate interview questions through contrastive learning and uses the self-distillation mechanism to perform cross-modal matching between technical entities and job requirements. A multi-level classification module is used to parse job requirements into a multi-level classification system, refine the matching granularity, and then calculate the knowledge point coverage based on preset weight rules; The question screening module is used to select the best interview questions from multiple candidate interview questions based on knowledge point coverage and job matching, and simultaneously generate corresponding question analysis and scoring criteria; The question adjustment module is used to collect the interviewee's answering speed and accuracy for each interview question during the interview, and dynamically adjust the difficulty of subsequent adjacent interview questions until the end of the interview.

[0014] And, an electronic device, comprising: at least one processor; and a memory communicatively coupled to the at least one processor; In which, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned high-precision AI interview question generation method based on self-distillation and industry knowledge base.

[0015] The technical effects achieved by the present invention are: By dynamically adjusting the difficulty of questions, the present invention can more accurately reflect the actual ability of the interviewee and improve the efficiency and accuracy of the interview process. At the same time, the interview question generation based on self-distillation and industry knowledge base can ensure that the interview questions are highly consistent with job requirements, reduce subjective evaluation errors, provide reliable data support for recruitment decisions, and realize the intelligence and personalization of the interview process. In addition, according to the characteristics of different positions, the difficulty and content of questions can be flexibly customized, thereby further improving the pertinence and discrimination of interview questions, making the interview process more efficient, accurately evaluating the interviewee's ability, and ensuring the quality of corporate recruitment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2 It is a schematic diagram of the system modules of the present invention; Figure 3 It is a schematic structural diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0018] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "one embodiment" or "embodiment" 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 a preferred embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it constitute a separate or selective embodiment that is mutually exclusive of other embodiments.

[0020] See also Figure 1 As shown, the present invention provides a high-precision AI interview question generation method based on self-distillation and industry knowledge base, including: S1. Real-time acquisition of industry technical documents and job description data, extraction of technical entities and their relationships, and construction of a dynamic knowledge graph with timestamps. In step S1, with the development of information technology, the generation method of interview questions has gradually shifted from traditional manual proposition to intelligent generation. Especially under the impetus of big data and artificial intelligence technology, AI interviews can dynamically adjust the content and difficulty of interview questions based on the real-time updated industry knowledge base to achieve personalized generation of interview questions. In this embodiment, technical documents and job description data in the industry are first collected. The technical documents cover the knowledge and skill requirements in various technical fields. Then, key technical entities and their associations are extracted from the technical documents and job descriptions, such as the dependencies between technical frameworks, programming languages ​​and algorithm models. Natural language processing technology is used to perform semantic analysis and structured storage on this information, and then a dynamic knowledge graph with timestamps is constructed to provide a real-time updated knowledge base for subsequent interview question generation. The steps of obtaining industry technical documents and job description data in real time, extracting technical entities and their associations, and constructing a dynamic knowledge graph with timestamps include: Build a relationship extraction network based on the attention mechanism to identify and extract the dependencies, version relationships, and scenario application relationships between technical entities, and annotate them with timestamps; Dynamically update the edge weights of the knowledge graph using a preset time decay function; Graph neural networks are used for cross-document entity alignment. Similarity calculations are used to eliminate naming ambiguities between data sources, and unified entity identifiers are recommended. Map the skill requirements in the job description into a graph query subgraph, and generate a job capability profile through subgraph matching; Specifically, when constructing the knowledge graph for generating interview questions, we first construct a relationship extraction network based on the attention mechanism (for example, by introducing the self-attention mechanism of the Transformer architecture, we deeply model the dependencies, version iteration relationships, and scenario application relationships between technical entities). The relationship extraction network can identify and extract the dependencies, version iteration relationships, and scenario application relationships between technical entities. At the same time, it can also mark each relationship with a timestamp to record its update time, so that the knowledge graph can reflect industry dynamics in real time and ensure that the interview questions are synchronized with the latest technology trends. After the knowledge graph is constructed, the preset time decay function will be used to dynamically update the edge weights in the knowledge graph. The time decay function calculates the timeliness of each relationship according to the timestamp to ensure that the interview questions can focus on the current key or popular technical points. The expression of the time decay function is: , where represents the updated edge weight, represents the initial edge weight, Represents the attenuation coefficient, which is a preset constant. Indicates the implementation timestamp, It represents the timestamp when the relationship in the knowledge graph is created. In addition, graph neural networks are used for cross-document entity alignment. Naming ambiguity between different data sources is eliminated through similarity calculation, and a unified identifier is assigned to each technical entity. The similarity calculation method can use cosine similarity or Jaccard similarity and other methods. Multi-dimensional matching is performed based on the contextual description information and attribute characteristics of the technical entity to improve the consistency and accuracy of the knowledge graph and lay the foundation for the subsequent interview question generation.

[0021] S2. Build a hybrid framework of contrastive learning and self-distillation, generate multiple candidate interview questions through contrastive learning, and use the self-distillation mechanism to perform cross-modal matching between technical entities and job requirements; In step S2, based on the knowledge graph, a hybrid framework of contrastive learning and self-distillation needs to be constructed to generate multiple candidate interview questions. The self-distillation mechanism is used to achieve cross-modal semantic alignment between technical entities and job requirements, thereby improving the matching degree between interview questions and jobs. The steps of constructing a hybrid framework of contrastive learning and self-distillation and generating multiple candidate interview questions through contrastive learning include: Based on the technical entities and relationships in the dynamic knowledge graph, a multi-head attention mechanism is used to generate question trunks and interference items, and negative sampling technology is used to construct comparison samples; Output the corresponding technical points priority based on the job capability profile as a reference standard to assist in cross-matching knowledge points with job requirements; Based on the priority of technical points and combined with the self-distillation mechanism, the matching degree between the question trunk and distractors is optimized to generate multiple candidate interview questions; Specifically, when generating candidate interview questions, we will first capture the complex interactions between technical entities based on the technical entities in the dynamic knowledge graph and the relationships between them, and then generate a logical and targeted question trunk. In order to increase the diversity and challenge of the interview questions, we will also use negative sampling technology to select entities that are related to the question trunk but not fully matched from the knowledge graph as interference items, and construct comparison samples. The comparison samples can help the interviewee identify key information in the answering process and improve the evaluation value of the interview. Then, the corresponding technical point priority will be output based on the job capability portrait (mapped to the query subgraph in the knowledge graph through early processing). The priority of technical points is used as a reference standard for generating interview questions, which can ensure that the interview questions revolve around the core skill requirements of the position. When generating questions, the knowledge points and job requirements will be cross-matched according to the priority of technical points to ensure that the interview questions not only cover the key skills of the position, but also can evaluate the actual mastery of the interviewees. Then, the self-distillation mechanism is combined to optimize the matching degree between the question trunk and the interference items. The self-distillation mechanism is a method of transferring the knowledge learned by the deep learning model to a simple model through knowledge distillation technology. It is used here to improve the quality and pertinence of the interview questions. By iteratively optimizing the question trunk and interference items for multiple times, multiple candidate interview questions for subsequent interviews are finally generated.

[0022] It should be noted that after the candidate interview questions are output, the job competency profile and the knowledge points of the candidate interview questions are overlapped and tested to determine the degree of fit between the interview questions and the job requirements. The specific process is as follows: Extract the core skill nodes and associated weights in the job capability profile from the knowledge graph; Perform reverse analysis on each candidate interview question to extract the set of technical entities involved and the topological relationships between them; The subgraph isomorphism method is used to detect the common connected subgraphs of the candidate knowledge point coverage subgraph and the job competency profile subgraph, and to identify overlapping knowledge points and their weight distribution; Based on the timeliness annotation of technical entities, the weights of overlapping knowledge points in the public connected subgraph are modified by applying the time decay coefficient; Calculate the sum of the weights of the corrected overlapping knowledge points and compare it with the total weight of the job competency profile to generate a fit score; In the above, after the candidate interview questions are generated, an overlap analysis will be conducted between the job competency profile and the knowledge point coverage of the questions to ensure that the interview questions match the job competency requirements. When executing the analysis, the core skill nodes in the job competency profile and their associated weights in the knowledge graph will be extracted from the knowledge graph first. The associated weights reflect the importance of different skills to the job, and then each candidate interview question will be reversely parsed. This process mainly involves parsing the set of technical entities involved in the question and the topological relationship between the technical entities in the knowledge graph to form a knowledge point coverage subgraph corresponding to the question. The subgraph isomorphism algorithm is then used to detect the common connected subgraph between the candidate knowledge point coverage subgraph and the job competency profile subgraph. Subgraph isomorphism is an algorithm in graph theory used to judge Whether a graph is a subgraph of another graph and finding all matching node mappings is a well-known technical means in this field. In this embodiment, it is used to identify the common knowledge points and their weight distribution in the interview questions and job capability portraits. Combined with the timeliness of the technical entity, the time decay coefficient will also be applied to correct the weights of overlapping knowledge points in the common connected subgraph to ensure that the latest updated technical points will get higher weights. The sum of the weights of the corrected overlapping knowledge points is then calculated and compared with the total weight of the job capability portrait to generate a fit score. The fit score can be calculated using methods such as cosine similarity to quantify the degree of match between job requirements and the coverage of knowledge points in interview questions, providing a corresponding basis for subsequent question screening.

[0023] S3. Analyze job requirements into a multi-level classification system, refine the matching granularity, and then calculate the knowledge point coverage based on preset weighting rules; In step S3, after clarifying the job requirements, the job requirements are further analyzed using a multi-level classification system, and the job ability requirements are refined into the smallest matching unit, thereby improving the matching accuracy. The coverage of each knowledge point is calculated using a preset weight rule to clarify the importance of each knowledge point in the job requirements. The steps of analyzing the job requirements into a multi-level classification system, refining the matching granularity, and then calculating the knowledge point coverage based on the preset weight rule include: Through natural language processing, the interviewee's job requirements are broken down into a three-level classification system, including the first-level technical field, the second-level technology stack, and the third-level skill points; Assign initial weights to the first-level technology areas, second-level technology stacks, and third-level skill points; Based on the association strength and time decay coefficient of technical entities in the knowledge graph, the initial weights of the first-level technical field, the second-level technology stack, and the third-level skill points are dynamically revised; Traverse the topological structure of knowledge points in the candidate interview questions, and cumulatively match them to the first-level technical field, second-level technology stack, and third-level skill points after the revised weights, and calculate the knowledge point coverage; When the knowledge point coverage falls below the preset threshold, the cross-level weight compensation mechanism is triggered, transferring weight from the high-weight level to the low-weight level until the overall coverage reaches the target. Specifically, this embodiment parses job requirements into a three-level classification system. Of course, in actual applications, it can be adjusted according to actual needs, such as expanding to a four-level or five-level classification system. The three-level classification system in this embodiment includes a first-level technical field, a second-level technology stack, and a third-level skill point. The first-level technical field describes the macro-technical field in which the position is located, and the second-level technology stack is further refined to a specific technical framework or tool set. The third-level skill point is the most fine-grained skill requirement, representing the specific skills that the position needs to master. After the decomposition is completed, initial weights will be assigned to the three levels respectively. The initial weight setting can be based on historical data, expert experience or industry standards to reflect the importance of skills at different levels to the position. Subsequently, the initial weight will be dynamically corrected according to the correlation strength and time decay coefficient of the technical entity in the knowledge graph. The specific correction method can use a weighted average algorithm to transfer the correlation strength as the weight. The time decay coefficient is used to adjust the impact of technological obsolescence on the weight. The corrected weight is more in line with the actual job requirements. After the weight correction is completed, the knowledge point topology structure in the candidate interview questions will be traversed, and the knowledge points involved in the questions will be matched with the three-level classification system. The weights of the matched knowledge points will be accumulated layer by layer to calculate the knowledge point coverage. Knowledge point coverage = the sum of the weights of the matched knowledge points / the total weight of the three-level classification system. The higher the knowledge point coverage, the more comprehensive the interview questions cover the job requirements. This is used to quantify the degree of coverage of the interview questions for the job requirements, and when the knowledge point coverage is lower than the preset threshold, the cross-level weight compensation mechanism will be triggered. The cross-level weight mechanism allows weights to be transferred from high-weight levels to low-weight levels to improve the overall coverage and ensure that the interview questions can comprehensively and accurately reflect the job requirements until the overall coverage reaches the preset standard, providing a corresponding basis for the subsequent interview question screening.

[0024] S4. Select the optimal interview questions from multiple candidate interview questions based on knowledge point coverage and job matching, and simultaneously generate corresponding question analysis and scoring criteria; In step S4, after the knowledge point coverage is output, the candidate interview questions are comprehensively evaluated in combination with the job matching degree, so as to screen out the best interview question from multiple candidate interview questions. After the best interview question is screened out, the corresponding question analysis and scoring criteria are generated for use in the interview process. The step of screening out the best interview question from multiple candidate interview questions based on the knowledge point coverage and job matching degree includes: Conduct preliminary screening of candidate interview questions based on the pre-set dual-threshold screening criteria of knowledge point coverage and fit score; After the initial screening, simulate the coverage trajectory of different question combinations on the knowledge graph, select the question combination with the highest cumulative knowledge coverage, and output it as the optimal interview question combination; When there are multiple optimal interview question combinations, the timeliness reinforcement mechanism is triggered. Based on the timeliness labeling of technical entities, question combinations with high timeliness are given priority. Specifically, in the process of screening the optimal interview question combination, the candidate interview questions will first be preliminarily screened according to the pre-set dual-threshold screening criteria of knowledge point coverage and fit score, so as to ensure that the candidate interview questions not only cover the key knowledge points required for the position, but also have a high degree of fit with the position capability profile. The pre-set dual threshold can be set according to historical data, expert experience or industry standards to ensure the rationality and effectiveness of the screening criteria. After preliminary screening, qualified candidate interview questions can be obtained. In order to further optimize the interview question combination, the coverage trajectory of different question combinations on the knowledge graph will be simulated in the candidate questions, and the coverage of different interview question combinations on the knowledge graph will be evaluated. By calculating the cumulative knowledge coverage, the question combination with the highest coverage will be selected and output as the optimal interview question combination. However, in actual applications, there may be multiple sets of optimal interview question combinations. At this time, the timeliness enhancement mechanism will be triggered. According to the timeliness labeling of the technical entity, the interview question combination with high timeliness will be given priority as the final output interview question combination.

[0025] In addition, after the optimal interview question combination is output, the weight distribution of technical entities in the knowledge graph is dynamically adjusted based on historical interview data feedback. The specific process is as follows: Collect historical interview data, including the accuracy rate, average time spent answering questions, and exposure frequency of related questions for each technical entity, and build a three-dimensional feedback indicator system; Generate dynamic attenuation coefficients of technical entities based on the three-dimensional feedback indicator system; Migrate the weights of technical entities in the knowledge graph based on the dynamic attenuation coefficient and update the knowledge graph in real time; Specifically, after the optimal interview question combination is output, the feedback of these interview questions in previous interviews will be counted, including the answer accuracy, average time and exposure frequency of related questions corresponding to each technical entity. Based on this, a three-dimensional feedback indicator system is constructed. The answer accuracy reflects the interviewee's mastery of the technical entity, the average time reveals the interviewee's thinking depth and proficiency when answering questions, and the exposure frequency of related questions reflects the frequency of appearance of the technical entity in the interview questions, which together reflect the actual performance of the technical entity in the interview process. Subsequently, the dynamic attenuation coefficient of the technical entity will be generated according to this three-dimensional feedback indicator system. The dynamic attenuation coefficient is a comprehensive indicator obtained by weighted calculation of the answer accuracy, average time and exposure frequency, and is used to measure The importance of technical entities in the time dimension is attenuated. A higher attenuation coefficient indicates that the influence of the technical entity in the interview is weakening. In addition, the dynamic attenuation coefficient is used to adjust the distribution of technical entity weights in the knowledge graph to reflect the importance and timeliness of the technical entity in the interview. Specifically, technical entities with high answer accuracy will be given higher weights, while technical entities with longer average answering time or too high exposure frequency of related questions will have their weights reduced accordingly. This ensures that the weight distribution of technical entities in the knowledge graph is more reasonable, that is, the weights of technical entities in the knowledge graph are migrated according to the dynamic attenuation coefficient, and the knowledge graph is updated in real time to ensure that the knowledge graph can be continuously optimized with the accumulation of interview data, providing a more accurate and reliable basis for the generation of subsequent interview questions.

[0026] S5. Collect the interviewee's answer speed and accuracy for each interview question during the interview, and dynamically adjust the difficulty of subsequent adjacent interview questions until the interview is completed; In step S5, during the interview process, in order to determine the interviewee's actual level, the interviewee's answering speed and accuracy for each interview question are collected in real time, so as to dynamically adjust the difficulty of subsequent adjacent interview questions. The step of collecting the interviewee's answering speed and accuracy for each interview question during the interview process and dynamically adjusting the difficulty of subsequent adjacent interview questions includes: Calculate the interviewee's answering speed and accuracy in real time after they start answering questions, and record them as the first adjustment vector and the second adjustment vector respectively; Perform time series analysis on the first adjustment vector and the second adjustment vector to predict the optimal difficulty curve of subsequent interview questions; Based on the predicted optimal difficulty curve, the difficulty coefficient of subsequent interview questions is dynamically adjusted. After each adjustment, the adjustment effect is verified in real time through a feedback mechanism to ensure that the difficulty of the adjusted questions matches the interviewee's ability. Specifically, when adjusting the difficulty of subsequent interview questions, the interviewee's answering speed and accuracy will first be collected in real time after the interviewee starts answering. The answering speed is the time it takes the interviewee to answer the question from the beginning to the completion of the answering, which is recorded as the first adjustment vector. The answering accuracy is the correct rate of the interviewee's answers to the question, which is recorded as the second adjustment vector, reflecting the interviewee's mastery of the knowledge points. Afterwards, the first adjustment vector and the second adjustment vector will be subjected to a time series analysis to determine the trend of the interviewee's ability change. The time series analysis can use a sliding window algorithm. Specifically, by setting a time window, the answering speed and accuracy data within the window are weighted averaged to obtain the interviewee's ability fluctuations in different time periods, and then based on it, predictive processing is performed to determine the difficulty of subsequent interview questions, and the optimal difficulty curve is drawn to facilitate subsequent adjustment of the question difficulty. At the same time, it can also provide the interviewer with a visual dynamic adjustment reference to determine the degree of match between the interviewee's ability and the difficulty of the question.

[0027] See also Figure 2 , a high-precision AI interview question generation system based on self-distillation and industry knowledge base, using the above-mentioned high-precision AI interview question generation method based on self-distillation and industry knowledge base, including: A graph construction module is used to obtain industry technical documents and job description data in real time, extract technical entities and their relationships, and build a dynamic knowledge graph with timestamps; The cross-modal matching module is used to build a hybrid framework of contrastive learning and self-distillation. It generates multiple candidate interview questions through contrastive learning and uses the self-distillation mechanism to perform cross-modal matching between technical entities and job requirements. A multi-level classification module is used to parse job requirements into a multi-level classification system, refine the matching granularity, and then calculate the knowledge point coverage based on preset weight rules; The question screening module is used to select the best interview questions from multiple candidate interview questions based on knowledge point coverage and job matching, and simultaneously generate corresponding question analysis and scoring criteria; The question adjustment module is used to collect the interviewee's answering speed and accuracy for each interview question during the interview, and dynamically adjust the difficulty of subsequent adjacent interview questions until the end of the interview.

[0028] The execution process of the above-mentioned interview question generation system is consistent with the high-precision AI interview question generation method based on self-distillation and industry knowledge base, so it will not be repeated here.

[0029] See also Figure 3 , an electronic device, the electronic device comprising: at least one processor; and a memory communicatively coupled to the at least one processor; Among them, the memory stores a computer program that can be executed by at least one processor, and the computer program is executed by at least one processor so that the at least one processor can execute the above-mentioned high-precision AI interview question generation method based on self-distillation and industry knowledge base.

[0030] The processor of the above-mentioned electronic device may be a central processing unit (CPU), a graphics processing unit (GPU) or a digital signal processor (DSP), etc. The memory of the above-mentioned electronic device may include random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other types of memory. Of course, the electronic device may also include an operator, such as an arithmetic logic unit, etc., for performing basic arithmetic and logical operations, as well as a network interface for network communication, and input and output devices for input and output, etc. The above-mentioned network interface may be a wired interface, such as an Ethernet interface, or a wireless interface, such as a WIFI interface, a Bluetooth interface, etc. The input and output devices may include a keyboard, a mouse, a touch screen, a display, a printer, etc. In specific implementation, the structure of the above-mentioned electronic device may be adjusted according to actual usage to adapt to different application scenarios and needs.

[0031] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0032] The foregoing is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make various improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained herein shall, unless otherwise specified or limited, be implemented in accordance with conventional means in the art.

Claims

1. A high-precision AI interview question generation method based on self-distillation and industry knowledge base, characterized by: include: Acquire industry technical documents and job description data in real time, extract technical entities and their relationships, and build a dynamic knowledge graph with timestamps; Build a hybrid framework of contrastive learning and self-distillation, generate multiple candidate interview questions through contrastive learning, and use the self-distillation mechanism to cross-modally match technical entities with job requirements; Analyze job requirements into a multi-level classification system, refine the matching granularity, and then calculate the knowledge point coverage based on preset weight rules; Based on the knowledge point coverage and job matching, the optimal interview questions are selected from multiple candidate interview questions, and the corresponding question analysis and scoring criteria are generated simultaneously; The interviewee's answer speed and accuracy for each interview question are collected during the interview process, and the difficulty of subsequent adjacent interview questions is dynamically adjusted until the end of the interview.

2. The high-precision AI interview question generation method based on self-distillation and industry knowledge base according to claim 1 is characterized by: The steps of obtaining industry technical documents and job description data in real time, extracting technical entities and their relationships, and constructing a dynamic knowledge graph with a timestamp include: Build a relationship extraction network based on the attention mechanism to identify and extract the dependencies, version relationships, and scenario application relationships between technical entities, and annotate them with timestamps; Dynamically update the edge weights of the knowledge graph using a preset time decay function; Graph neural networks are used for cross-document entity alignment. Similarity calculations are used to eliminate naming ambiguities between data sources, and unified entity identifiers are recommended. Map the skill requirements in the job description into a graph query subgraph, and generate a job capability profile through subgraph matching.

3. The high-precision AI interview question generation method based on self-distillation and industry knowledge base according to claim 1 is characterized by: The steps of constructing a hybrid framework of contrastive learning and self-distillation and generating multiple candidate interview questions through contrastive learning include: Based on the technical entities and relationships in the dynamic knowledge graph, a multi-head attention mechanism is used to generate question trunks and interference items, and negative sampling technology is used to construct comparison samples; Output the corresponding technical points priority based on the job capability profile as a reference standard to assist in cross-matching knowledge points with job requirements; Based on the priority of technical points and combined with the self-distillation mechanism, the matching degree between the question trunk and interference items is optimized to generate multiple candidate interview questions.

4. The high-precision AI interview question generation method based on self-distillation and industry knowledge base according to claim 1 is characterized by: After the candidate interview questions are output, the job competency profile and the knowledge points of the candidate interview questions are overlapped and tested to determine the degree of fit between the interview questions and the job requirements. The specific process is as follows: Extract the core skill nodes and associated weights in the job capability profile from the knowledge graph; Perform reverse analysis on each candidate interview question to extract the set of technical entities involved and the topological relationships between them; The subgraph isomorphism method is used to detect the common connected subgraphs of the candidate knowledge point coverage subgraph and the job competency profile subgraph, and to identify overlapping knowledge points and their weight distribution; Based on the timeliness annotation of technical entities, the weights of overlapping knowledge points in the public connected subgraph are modified by applying the time decay coefficient; Calculate the sum of the weights of the corrected overlapping knowledge points and compare it with the total weight of the job competency profile to generate a fit score.

5. The high-precision AI interview question generation method based on self-distillation and industry knowledge base according to claim 1 is characterized by: The steps of analyzing job requirements into a multi-level classification system, refining the matching granularity, and then calculating the knowledge point coverage based on preset weight rules include: Through natural language processing, the interviewee's job requirements are broken down into a three-level classification system, including the first-level technical field, the second-level technology stack, and the third-level skill points; Assign initial weights to the first-level technology areas, second-level technology stacks, and third-level skill points; Based on the association strength and time decay coefficient of technical entities in the knowledge graph, the initial weights of the first-level technical field, the second-level technology stack, and the third-level skill points are dynamically revised; Traverse the topological structure of knowledge points in the candidate interview questions, and cumulatively match them to the first-level technical field, second-level technology stack, and third-level skill points after the revised weights, and calculate the knowledge point coverage; Among them, when the knowledge point coverage is lower than the preset threshold, the cross-level weight compensation mechanism is triggered, and the weight is transferred from the high-weight level to the low-weight level until the overall coverage meets the standard.

6. The high-precision AI interview question generation method based on self-distillation and industry knowledge base according to claim 1 is characterized by: The step of selecting the best interview question from multiple candidate interview questions based on knowledge point coverage and job matching includes: Conduct preliminary screening of candidate interview questions based on the pre-set dual-threshold screening criteria of knowledge point coverage and fit score; After the initial screening, simulate the coverage trajectory of different question combinations on the knowledge graph, select the question combination with the highest cumulative knowledge coverage, and output it as the optimal interview question combination; When there are multiple combinations of optimal interview questions, the timeliness enhancement mechanism is triggered, and question combinations with high timeliness are given priority based on the timeliness marking of technical entities.

7. The high-precision AI interview question generation method based on self-distillation and industry knowledge base according to claim 1 is characterized by: After the optimal interview question combination is output, the weight distribution of technical entities in the knowledge graph is dynamically adjusted based on historical interview data feedback. The specific process is as follows: Collect historical interview data, including the accuracy rate, average time spent answering questions, and exposure frequency of related questions for each technical entity, and build a three-dimensional feedback indicator system; Generate dynamic attenuation coefficients of technical entities based on the three-dimensional feedback indicator system; The weights of technical entities in the knowledge graph are migrated based on the dynamic attenuation coefficient, and the knowledge graph is updated in real time.

8. The high-precision AI interview question generation method based on self-distillation and industry knowledge base according to claim 1 is characterized by: The step of collecting the interviewee's answering speed and accuracy for each interview question during the interview and dynamically adjusting the difficulty of subsequent adjacent interview questions includes: Calculate the interviewee's answering speed and accuracy in real time after they start answering questions, and record them as the first adjustment vector and the second adjustment vector respectively; Performing a time series analysis on the first adjustment vector and the second adjustment vector to predict an optimal difficulty curve for subsequent interview questions; Based on the predicted optimal difficulty curve, the difficulty coefficient of subsequent interview questions is dynamically adjusted. After each adjustment, the adjustment effect is verified in real time through the feedback mechanism to ensure that the difficulty of the adjusted questions matches the interviewee's ability.

9. A high-precision AI interview question generation system based on self-distillation and industry knowledge base, featuring: The method for generating high-precision AI interview questions based on self-distillation and industry knowledge base according to any one of claims 1 to 8 comprises: A graph construction module is used to obtain industry technical documents and job description data in real time, extract technical entities and their relationships, and build a dynamic knowledge graph with timestamps; The cross-modal matching module is used to build a hybrid framework of contrastive learning and self-distillation. It generates multiple candidate interview questions through contrastive learning and uses the self-distillation mechanism to perform cross-modal matching between technical entities and job requirements. A multi-level classification module is used to parse job requirements into a multi-level classification system, refine the matching granularity, and then calculate the knowledge point coverage based on preset weight rules; The question screening module is used to select the best interview questions from multiple candidate interview questions based on knowledge point coverage and job matching, and simultaneously generate corresponding question analysis and scoring criteria; The question adjustment module is used to collect the interviewee's answering speed and accuracy for each interview question during the interview, and dynamically adjust the difficulty of subsequent adjacent interview questions until the end of the interview.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; In which, the memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the high-precision AI interview question generation method based on self-distillation and industry knowledge base as described in any one of claims 1 to 8.

Images