Personalized intelligent interview system

Abstract

The invention discloses a personalized intelligent interview system, and relates to the technical field of artificial intelligence and recruitment interview, and the system comprises a position information analysis module which is used for inputting an original position description text, carrying out the analysis through employing a natural language processing NLP technology, and extracting structured position demand data containing skill items, characteristic items and quantification levels; the resume / position intelligent matching module is used for receiving the structured position demand data, receiving a job seeker resume text, performing intelligent matching analysis by constructing a skill semantic vector space, and outputting a multi-dimensional matching score matrix and an investigation key index set; according to the job demand and the job seeker resume, personalized interview topic screening and optimization are realized, and the interview pertinence and accuracy are improved; the job seeker is comprehensively evaluated, and the resume / position matching information is combined, so that a more comprehensive evaluation and scoring basis is provided, and an enterprise is helped to make a more accurate recruitment decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and recruitment interviews, and particularly relates to an individualized intelligent interview system. BACKGROUND

[0002] In today's highly competitive job market, enterprises have an increasingly urgent need to recruit suitable talents. Traditional interview methods have gradually become difficult to meet the requirements of efficient and accurate recruitment. In this context, AI interviews have emerged. AI interviews are an innovative interview method that combines artificial intelligence technology and video interviews, greatly promoting the intelligent development of the recruitment process.

[0003] During the interview process, on the one hand, advanced technologies such as voice recognition and image recognition are used to analyze the candidate's answer accuracy, language expression ability, logical thinking, emotional state, expression management, and body language in real time, and to give objective evaluations. On the other hand, it breaks through the limitations of time and space, improves recruitment efficiency, and reduces the cost of manpower, material resources, and financial resources. In addition, AI interviews set standardized processes and use uniform scoring standards to evaluate each candidate, effectively avoiding the interference of human factors, reducing subjective bias and scoring standard differences, and improving the fairness and accuracy of recruitment.

[0004] For example, application No. CN115619360A discloses a talent selection and recruitment system and method based on artificial intelligence, application No. CN115413348B discloses a system and method for automatically verifying and quantifying interview question answers, and application No. CN110347787B discloses an interview method, device, and terminal equipment based on an AI-assisted interview scene, promoting the intelligent development of interviews, but there are still certain limitations: for example, during the interview process, the selection of written tests and oral question and answer questions becomes a key issue. Inappropriate questions cannot accurately tap the candidate's ability and potential, thereby affecting the judgment of the candidate's matching degree with the position.

[0005] The existing technology has the following shortcomings: the current common practice is for enterprise HRs to select some questions in advance for a position or randomly select a certain number of questions from various types of question banks. Although this method can initially meet the basic needs of the interview, it has obvious shortcomings in terms of accuracy and individualization. Because it is not possible to make differentiated question selection based on the nature and requirements of different positions, as well as the personal experiences and experiences of job seekers, different positions have different requirements for the ability and quality of candidates, and the personal backgrounds and experiences of job seekers also vary greatly. Therefore, if personalized interview questions cannot be provided, it is difficult to comprehensively and accurately assess whether the candidate is truly suitable for the position, which may result in the enterprise missing out on excellent talents or recruiting mismatched personnel, affecting the development of the enterprise.

[0006] The above information disclosed in the BACKGROUND section merely to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0007] The purpose of the present application is to provide a personalized intelligent interview system, which solves the problems in the above background technology through the synergistic effect of the position information analysis module, the resume / position intelligent matching module, the question bank analysis and annotation preliminary screening module, the interview question screening and optimization module, the virtual interview subsystem and the interview result comprehensive evaluation module.

[0008] In order to achieve the above purpose, the present application provides the following technical scheme: a personalized intelligent interview system, comprising a position information analysis module: inputting an original position description text, using natural language processing (NLP) technology to analyze and extract structured position requirement data containing skill items, trait items and quantitative grades; A resume / position intelligent matching module: receiving the structured position requirement data and receiving a job seeker resume text, performing intelligent matching analysis by constructing a skill semantic vector space, and outputting a multi-dimensional matching score matrix and a set of key indicators for examination; A question bank analysis and annotation preliminary screening module: using a bidirectional long short-term memory network (BiLSTM-CRF) model to capture the context information of the question text in the question bank, automatically annotating the question type label, training an XGBoost regression model based on historical answer data to predict the difficulty of the question, and generating a preliminary screening question set with metadata according to the received structured position requirement data and the matching degree of the question type label; An interview question screening and optimization module: inputting the set of key indicators for examination to perform personalized adjustment on the preliminary screening question set with metadata, obtaining a personalized adjusted question sequence and a fine-tuning question text; An interview intelligent arrangement module: constructing an interview state transition model to design a four-stage state mechanism, receiving the personalized adjusted question sequence, generating a smooth transition question sequence based on the question difficulty coefficient, and generating a dynamic question and answer path graph according to the smooth transition question sequence; A virtual interview subsystem: executing the dynamic question and answer path graph, and collecting multi-modal data streams including voice, expression and posture through a camera and a microphone; An interview result comprehensive evaluation module: receiving the multi-modal data streams and fusing the multi-dimensional matching score matrix, constructing a three-dimensional scoring system, setting an anomaly detection mechanism, calculating the final matching degree using a weighted fusion function, and generating an evaluation report with an anomaly label.

[0009] Optionally, the position information analysis module extracts the structured position requirement data as follows: Skill and trait extraction: use the BERT model variant RoBERTa to perform named entity recognition on the original job description text to extract skill items and trait items; Skill-level mapping relationship construction: based on dependency syntax analysis technology, analyze the grammatical relationship between words in the original job description text, and construct the "skill-level" mapping relationship, and quantify the requirements of the position to each skill as a 1-5 point requirement level matrix; Implicit demand dimension identification: use DBSCAN clustering algorithm to mine the implicit requirements in the original job description text and supplement them to the structured job requirement data; Output JSON format structured job requirement data.

[0010] Optionally, the resume / job intelligent matching module has a built-in skill matching engine, gap analysis engine and authenticity verification engine, wherein the skill matching engine: constructs a skill semantic vector space, maps and converts skill items in the job requirement matrix into semantic vectors, and generates a skill coverage score by calculating the cosine similarity between the skill keyword vector in the resume and the job requirement matrix; The gap analysis engine: pre-set rule base, detailed analysis of project experience in the resume, comparative analysis of project experience and job requirements, identification of coverage gaps between project experience and job requirements, and generation of key indicator set according to the coverage gap; The authenticity verification engine: based on the term frequency-inverse document frequency TF-IDF weight analysis method, analyzes the terms in the resume to trigger authenticity labels.

[0011] Optionally, the library analysis and labeling preliminary screening module includes a question type label automatic labeling unit, a difficulty coefficient prediction unit, and a question preliminary screening strategy unit, wherein the question type label automatic labeling unit: inputs the question text into the bidirectional long short-term memory network BiLSTM-CRF model, and automatically outputs a three-level label system; The difficulty coefficient prediction unit: uses historical answer data to train an XGBoost regression model, analyzes historical answer accuracy, average answer time, and keyword complexity, and outputs a normalized difficulty coefficient; The question preliminary screening strategy unit: calculates the matching degree of the question type label and the job requirement matrix, and selects questions with a matching degree greater than 80%; the selected questions are divided into closed and open channel preliminary screening lists.

[0012] Optionally, the test question screening and optimization module includes a question optimization unit, a question dynamic rewriting unit, and a question difficulty calibration unit, wherein the question optimization unit: extracts skill gaps from the key indicator set, and preferentially selects questions with the highest relevance to the skill gaps in the preliminary screening list; The question dynamic rewriting unit: for the rewriting of closed questions, replace the general terms in the standard questions with the project terms in the resume of the job seeker, and for the rewriting of open questions, add follow-up sub-questions based on skill gap, and generate follow-up chains; The question difficulty calibration unit: based on the distribution curve of the normalized difficulty coefficient, the selected questions are combined and adjusted.

[0013] Optionally, the interview intelligent arrangement module constructs an interview state transition model, and designs a four-stage state mechanism, including a warm-up stage, a core competence stage, a stress test stage, and a closing stage. Based on the question sequence adjusted individually, a smooth transition question sequence is generated combined with the question difficulty coefficient, ensuring that the difficulty fluctuation value of adjacent questions is less than or equal to 0.3, so as to avoid sudden difficulty changes that cause excessive stress on the job seeker. It also includes an emotion analysis and question adjustment unit that receives the tension curve of the virtual interview subsystem in real time, and when it detects that the job seeker's tension has exceeded the threshold for 10 seconds in a row, it inserts a relaxation question in time to ensure the smooth progress of the interview.

[0014] Optionally, the virtual interview subsystem includes an answer text transcription unit, a body movement parameter extraction unit, and an expression confidence matrix calculation unit, wherein the answer text transcription unit: through an automatic speech recognition (ASR) engine, transcribes the job seeker's answer speech into text and marks the time stamp, speech speed and pause frequency; The body movement parameter extraction unit: uses the OpenPose model to extract the body movement frequency parameters of the job seeker, outputs 18 joint coordinates, and calculates the gesture change rate; The expression confidence matrix calculation unit: calculates the expression confidence matrix based on the ResNet50 network, and outputs the concentration / tension time curve.

[0015] Optionally, the interview result comprehensive evaluation module constructs a three-dimensional scoring system as follows: The three-dimensional scoring system includes a professional competence dimension, an expression ability dimension, and a sincerity dimension; The professional competence dimension calculates the job seeker's question score; The expression ability dimension calculates the expression score according to the benchmark speed, actual speed and average tension in the job seeker's answer session; The sincerity dimension calculates the consistency score according to the Jaccard similarity between the resume project description and the oral project description; The final matching degree is obtained by using a weighted fusion function to calculate the question score, expression score and consistency score.

[0016] Optionally, the abnormality detection mechanism comprises an answer plagiarism detector, an experience contradiction detector and a behavior abnormality marker, wherein the answer plagiarism detector measures the difference between two texts by calculating the edit distance between the answer text and the standard answer; The experience contradiction detector compares the items mentioned in the dialogue with the date, technical stack information recorded in the resume; The behavior abnormality marker establishes a baseline for body movement according to the mean and standard deviation of the gesture change rate of the first five questions, and marks a suspicious point of behavior fluctuation when the frequency mutation value of the body movement in the subsequent questions exceeds three times the standard deviation of the baseline.

[0017] A computer device comprises a memory, a processor and a computer readable storage medium; the memory stores a computer program, and the processor implements the steps of the above-mentioned personalized intelligent interview system when executing the computer program.

[0018] In the above technical solution, the present application provides technical effects and advantages: The present application analyzes the skills, characteristics and quantitative grades of job seekers required by the position by intelligently analyzing the position requirements and descriptions according to the position requirements and the resumes of job seekers, realizes personalized interview question screening and optimization, improves the pertinence and accuracy of the interview, comprehensively evaluates the job seekers, provides more comprehensive evaluation and scoring basis in combination with the resume / position matching information, helps enterprises make more accurate recruitment decisions, detects abnormal conditions in the answers and marks reminders, which helps enterprises to discover potential problems and avoid recruiting unsuitable personnel, and realizes an automated interview process and real-time evaluation report, which saves the time of enterprises and job seekers and improves the recruitment efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art according to these drawings.

[0020] Figure 1 The module block diagram of the personalized intelligent interview system of the present application. DETAILED DESCRIPTION

[0021] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive aspects to those skilled in the art.

[0022] EMBODIMENTS The application provides a personalized intelligent interview system as shown in the specification Figure 1 The personalized intelligent interview system comprises a position information analysis module: inputting an original position description text, adopting natural language processing (NLP) technology to analyze and extract structured position requirement data containing skill items, trait items and quantitative grades, and transmitting the data to a resume / position intelligent matching module, a question bank analysis and annotation preliminary screening module and an interview question screening and optimization module; The resume / position intelligent matching module: receiving the structured position requirement data and receiving a job seeker resume text, performing intelligent matching analysis by constructing a skill semantic vector space, outputting a multi-dimensional matching score matrix and an investigation key index set, and transmitting the matrix and the set to the interview question screening and optimization module; The question bank analysis and annotation preliminary screening module: adopting a bidirectional long short-term memory network (BiLSTM-CRF) model to capture context information of question text in the question bank, automatically annotating a question type label, training an XGBoost regression model based on historical answer data to predict question difficulty, generating a preliminary screening question set with metadata according to the received structured position requirement data and the matching degree of the question type label, and transmitting the set to the interview question screening and optimization module, wherein the metadata comprises a question type, difficulty and related skill label; The interview question screening and optimization module: inputting the investigation key index set to perform personalized adjustment on the preliminary screening question set with metadata, obtaining a question sequence and fine-tuning question text after personalized adjustment, and transmitting the sequence and the text to an interview intelligent arrangement module; The interview intelligent arrangement module: constructing an interview state transition model to design a four-stage state mechanism, receiving the question sequence after personalized adjustment, generating a smooth transition question sequence in combination with a question difficulty coefficient, and generating a dynamic question and answer path graph according to the smooth transition question sequence, wherein the dynamic question and answer path graph comprises a state transition rule and a real-time intervention interface, and is transmitted to a virtual interview subsystem; The virtual interview subsystem: executing the dynamic question and answer path graph, collecting multi-modal data streams containing voice, expression and posture through a camera and a microphone, and transmitting the data streams to an interview result comprehensive evaluation module; The interview result comprehensive evaluation module: receiving the multi-modal data streams and fusing the multi-dimensional matching score matrix, setting an anomaly detection mechanism by constructing a three-dimensional scoring system, calculating a final matching degree by using a weighted fusion function, and generating an evaluation report with an anomaly label.

[0023] Specifically, the steps of extracting the structured position requirement data by the position information analysis module are as follows: Skill item and trait item extraction: adopting a BERT model variant RoBERTa to perform named entity recognition on the original position description text, and extracting skill items and trait items, wherein the skill items include programming languages and tool usage, and the trait items include communication ability, team cooperation and stress resistance; Skill-level mapping relationship construction: based on dependency syntax analysis technology, the syntactic relationship between words in the original position description text is analyzed, and the "skill-level" mapping relationship is constructed, and the requirement of the position to each skill is quantified as a 1-5 point demand level matrix; Implicit demand dimension identification: using DBSCAN clustering algorithm to mine the implicit demand in the original position description text, and supplementing to the structured position demand data; Output JSON format structured position demand data.

[0024] Specifically, the resume / position intelligent matching module is built-in skill matching engine, gap analysis engine and authenticity verification engine, wherein the skill matching engine: construct skill semantic vector space, map and convert skill items in position demand matrix into 300-dimensional semantic vector, generate skill coverage score by calculating the cosine similarity between skill keyword vector in resume and position demand matrix; The gap analysis engine: preset rule base, detailed analysis of project experience in the resume, comparative analysis of project experience and job requirements, identification of the coverage gap between project experience and job requirements, and generation of key indicator set according to the coverage gap; The authenticity verification engine: based on term frequency-inverse document frequency TF-IDF weight analysis method, the terms in the resume are analyzed, the authenticity mark is triggered, and the related content is reminded for key verification in subsequent evaluation.

[0025] Specifically, the question bank analysis and annotation preliminary screening module includes a question type label automatic annotation unit, a difficulty coefficient prediction unit and a question preliminary screening strategy unit, wherein the question type label automatic annotation unit: input the question text into the bidirectional long short-term memory network BiLSTM-CRF model, automatically output a three-level label system, including a first-level label of technical type / behavior type / scenario type, a second-level label of algorithm question / database question / conflict processing / target achievement, and a third-level label of open type / closed type; The difficulty coefficient prediction unit: uses historical answer data to train an XGBoost regression model, analyzes historical answer accuracy, average answer time, and keyword complexity, and outputs a normalized difficulty coefficient; The question preliminary screening strategy unit: calculates the matching degree of question type label and position demand matrix, and selects questions with a matching degree greater than 80%, and divides the selected questions into closed and open two-channel preliminary screening lists, wherein the closed questions have clear answers, and the open questions focus more on the thinking ability and expression ability of the candidate.

[0026] Specifically, the interview question screening optimization module includes a question optimization unit, a question dynamic rewriting unit, and a question difficulty calibration unit. The question optimization unit extracts skill gaps from the key indicator set, and preferentially selects questions with the highest correlation to the skill gaps in the preliminary screening list to ensure that the interview questions can target the key abilities and skills of the candidate. The question dynamic rewriting unit rewrites closed questions by replacing general terms in standard questions with project terms in the candidate's resume, and rewrites open questions by adding follow-up sub-questions based on skill gaps and generating follow-up chains. The question difficulty calibration unit adjusts the selected questions based on the distribution curve of the normalized difficulty coefficient to ensure that the question combination can cover the easy, medium, and difficult levels in a gradient, and the difficulty fluctuation value between adjacent questions should not be too large to ensure the reasonableness and coherence of the interview.

[0027] Specifically, the interview intelligent arrangement module constructs an interview state transition model and designs a four-stage state mechanism, including a warm-up stage, a core competence stage, a stress test stage, and a wrap-up stage. The goal of the warm-up stage is to relieve the candidate's nervousness and gradually enter the interview state. The goal of the core competence stage is to focus on the candidate's professional skills and key abilities. The goal of the stress test stage is to simulate the stress scenario in the workplace and examine the candidate's adaptability and psychological endurance. The goal of the wrap-up stage is to summarize and feedback the interview to confirm the candidate's cultural fit. Based on the question sequence adjusted individually, a smooth transition question sequence is generated based on the question difficulty coefficient to ensure that the difficulty fluctuation value between adjacent questions is ≤0.3 to avoid sudden difficulty changes that may cause excessive stress on the candidate. It also includes an emotion analysis and question adjustment unit that receives the nervousness curve of the virtual interview subsystem in real time. When the candidate's continuous 10-second nervousness exceeds the threshold, a relaxation question is inserted in time to help the candidate relieve nervousness and ensure the smooth progress of the interview.

[0028] Specifically, the virtual interview subsystem includes a question text transcription unit, a body movement parameter extraction unit, and an expression confidence matrix calculation unit. The question text transcription unit converts the candidate's answer speech into text through an automatic speech recognition (ASR) engine and marks the timestamp, speech speed, and pause frequency to facilitate analysis and evaluation of the answer content. It also records the answer time as a reference for evaluating expression ability and thinking speed. The body movement parameter extraction unit uses the OpenPose model to extract the job seeker's body movement frequency parameters, outputs the coordinates of 18 joint points, and calculates the gesture change rate to reflect the job seeker's emotional state and self-confidence; The expression confidence matrix calculation unit calculates the expression confidence matrix based on the ResNet50 network, outputs the concentration / tension time series curve, and analyzes the job applicant's facial expression to determine the changes in their concentration and tension. Facial expressions include 7-dimensional emotion vectors of anger, disgust, fear, happiness, sadness, surprise, and neutrality.

[0029] Specifically, the steps for constructing a three-dimensional scoring system in the comprehensive evaluation module of interview results are as follows: The three-dimensional scoring system includes the dimensions of professional ability, expression ability, and integrity; The professional ability dimension is calculated by calculating the candidate's question score, where question score = answer similarity × question weight; The expressiveness dimension is calculated based on the candidate's baseline speaking speed, actual speaking speed, and average nervousness during the test. Expressiveness score = (baseline speaking speed - |actual speaking speed - baseline speaking speed|) / baseline speaking speed × (1 - average nervousness); The integrity dimension calculates the consistency score based on the Jaccard similarity between the resume item description and the oral item description, where the consistency score = Jaccard similarity. The Jaccard similarity is used to measure the consistency between the content stated in the resume and the oral answer in the interview. The final matching degree is obtained by using a weighted fusion function to calculate the question score, expressiveness score and consistency score, where the final matching degree = w1 question score + w1 expressiveness score + w3 consistency score, and w1+w2+w3=1.

[0030] Specifically, the anomaly detection mechanism includes an answer plagiarism detector, an experience contradiction detector, and a behavior anomaly marker. The answer plagiarism detector measures the degree of difference between the answer text and the standard answer by calculating the edit distance between the two texts. When the similarity between the answer text and the standard answer is greater than 90%, an alert is triggered, reminding the interviewer that the answer content may be plagiarized. The experience discrepancy detector compares the projects mentioned in the conversation with the dates and technology stack information recorded in the resume. If there are obvious discrepancies between the spoken descriptions, such as conflicting project dates or mismatched technology stacks, a detailed analysis is conducted to determine whether there is a possibility of experience fraud. The behavior abnormality marker: according to the mean and standard deviation of the gesture change rate of the first 5 questions, a body movement baseline is established, when the body movement frequency mutation value in the subsequent questions exceeds three times the standard deviation of the baseline, the behavior fluctuation suspicious point is marked, and the abnormal mutation of the body movement frequency may imply that the job seeker has tension, dishonesty and other behaviors, which need further attention.

[0031] The above formulas are dimensionless values calculated, and the formulas are obtained by collecting a large amount of data to simulate a formula of the latest real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0032] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center and the like containing one or more available medium sets. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD) or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0033] It should be understood that in various embodiments of the present application, the size of the sequence number of the above processes does not mean the order of execution, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0034] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0035] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A personalized intelligent interview system, characterized by: It includes a job information analysis module: it inputs the original job description text, uses natural language processing (NLP) technology to parse it, and extracts structured job requirement data including skills, traits, and quantitative grades; Resume / position intelligent matching module: Receives structured job requirement data and applicant resume text, performs intelligent matching analysis by constructing a skill semantic vector space, and outputs a multi-dimensional matching score matrix and a set of key assessment indicators; Question bank analysis and annotation preliminary screening module: This module uses a bidirectional long short-term memory network (BiLSTM-CRF) model to capture the contextual information of question text in the question bank, automatically annotate question type labels, and train an XGBoost regression model based on historical answer data to predict question difficulty. Based on the received structured job requirement data and the matching degree of question type labels, it generates a preliminary screening question set with metadata. Interview question screening and optimization module: Input the key assessment indicator set to perform personalized adjustments to the initial screening question set with metadata, and obtain the personalized adjusted question sequence and fine-tuned question text; Interview intelligent orchestration module: This module builds an interview state transition model and designs a four-stage state mechanism. It receives a personalized question sequence, combines the question difficulty coefficient, and generates a smooth transition question sequence. Based on this smooth transition question sequence, it generates a dynamic question-answering path diagram. Virtual interview subsystem: Executes a dynamic question-and-answer path map and collects multimodal data streams including voice, facial expressions, and gestures through cameras and microphones; Interview result comprehensive evaluation module: receives multimodal data streams and integrates multi-dimensional matching scoring matrices. By constructing a three-dimensional scoring system, setting up an anomaly detection mechanism, and using a weighted fusion function to calculate the final matching degree, an evaluation report with anomaly markings is generated.

2. A personalized intelligent interview system according to claim 1, characterized in that: The steps of extracting structured job requirement data by the job information analysis module are as follows: Extraction of skills and traits: RoBERTa, a variant of the BERT model, is used to perform named entity recognition on the original job description text to extract skills and traits. Skill-level mapping: Using dependency parsing technology, we analyze the grammatical relationships between words in the original job description to construct a skill-level mapping. We then quantify the skill requirements for each position into a 1-5 requirement level matrix. Identification of implicit demand dimensions: Use the DBSCAN clustering algorithm to mine implicit requirements in the original job description text and supplement them into the structured job requirement data; Output structured job requirement data in JSON format.

3. A personalized intelligent interview system according to claim 2, characterized in that: The resume / position intelligent matching module has built-in skill matching engines, gap analysis engines, and authenticity verification engines. The skill matching engine: constructs a skill semantic vector space, maps skill items in the position requirement matrix into semantic vectors, and generates a skill coverage score by calculating the cosine similarity between the skill keyword vectors in the resume and the position requirement matrix; The gap analysis engine: a preset rule base is used to conduct a detailed analysis of the project experience in the resume, compare the project experience with the job requirements, identify the coverage gap between the project experience and the job requirements, and then generate a set of key assessment indicators based on the coverage gap; The authenticity verification engine analyzes the terms in the resume based on the term frequency-inverse document frequency TF-IDF weight analysis method to trigger the authenticity mark.

4. A personalized intelligent interview system according to claim 3, characterized in that: The library analysis and annotation preliminary screening module includes a question type label automatic labeling unit, a difficulty coefficient prediction unit and a question preliminary screening strategy unit, wherein the question type label automatic labeling unit: inputs the question text into the bidirectional long short-term memory network BiLSTM-CRF model and automatically outputs a three-level label system; The difficulty coefficient prediction unit uses historical answer data to train the XGBoost regression model, and outputs a normalized difficulty coefficient by analyzing the historical answer accuracy, average answer time, and keyword complexity; The question initial screening strategy unit calculates the matching degree between the question type label and the position requirement matrix, and screens out questions with a matching degree greater than 80%, and divides the screened questions into a closed and open dual-channel initial screening list.

5. A personalized intelligent interview system according to claim 4, characterized in that: The test question screening and optimization module includes a question optimization unit, a question dynamic rewriting unit, and a question difficulty calibration unit, wherein the question optimization unit extracts skill gaps from the key assessment indicators and gives priority to questions with the highest relevance to the skill gaps in the initial screening list; The dynamic question rewriting unit: for closed questions, it rewrites the general terms in the standard questions with the project terms in the job seeker's resume; for open questions, it rewrites the questions by adding follow-up sub-questions based on skill gaps and generating a chain of follow-up questions; The question difficulty calibration unit performs combination adjustment on the selected questions based on the distribution curve of the normalized difficulty coefficient.

6. A personalized intelligent interview system according to claim 5, characterized in that: The interview intelligent orchestration module constructs an interview state transition model and designs a four-stage state mechanism, including a warm-up stage, a core competency stage, a stress test stage, and a closing stage; Based on the personalized adjusted question sequence and the question difficulty coefficient, a smooth transition question sequence is generated to ensure that the difficulty fluctuation value of adjacent questions is ≤0.3, so as to avoid excessive pressure on job seekers caused by sudden changes in difficulty. It also includes a sentiment analysis and question adjustment unit, which receives the tension curve of the virtual interview subsystem in real time. When it detects that the job seeker's tension exceeds the threshold for 10 consecutive seconds, it inserts relaxation questions in time to ensure the smooth progress of the interview.

7. A personalized intelligent interview system according to claim 6, characterized in that: The virtual interview subsystem includes an answer text transcription unit, a body movement parameter extraction unit, and an expression confidence matrix calculation unit. The answer text transcription unit uses an automatic speech recognition (ASR) engine to transcribe the applicant's answer voice into text and mark it with timestamps, speaking speed, and pause frequency. The body movement parameter extraction unit uses the OpenPose model to extract the job seeker's body movement frequency parameters, outputs the coordinates of 18 joint points, and calculates the gesture change rate; The expression confidence matrix calculation unit calculates the expression confidence matrix based on the ResNet50 network and outputs a concentration / tension timing curve.

8. The personalized intelligent interview system according to claim 7, characterized in that: The steps of constructing a three-dimensional scoring system for the interview result comprehensive evaluation module are as follows: The three-dimensional scoring system includes the dimensions of professional ability, expression ability, and integrity; The professional competence dimension is calculated by calculating the candidate's question score; The expressiveness dimension is calculated based on the candidate's baseline speaking speed, actual speaking speed, and average tension during the answering process. The integrity dimension is a consistency score calculated based on the Jaccard similarity between the resume item description and the spoken item description; The final matching degree is obtained by calculating the question score, expressiveness score and consistency score using a weighted fusion function.

9. The personalized intelligent interview system according to claim 8, characterized in that: The anomaly detection mechanism includes an answer plagiarism detector, an experience contradiction detector, and a behavior anomaly marker, wherein the answer plagiarism detector: calculates the edit distance between the answer text and the standard answer to measure the degree of difference between the two texts; The experience contradiction detector compares the projects mentioned in the conversation with the dates and technology stack information recorded in the resume; The behavioral anomaly marker establishes a limb movement baseline based on the mean and standard deviation of the gesture change rate in the first five questions. When the limb movement frequency mutation value in subsequent questions exceeds three times the standard deviation of the baseline, it marks a suspicious point of behavioral fluctuation.

10. A computer device comprising: A memory, a processor and a computer-readable storage medium; the memory stores a computer program, characterized in that when the processor executes the computer program, the steps of a personalized intelligent interview system according to any one of claims 1 to 9 are implemented.

Images