An artificial intelligence-based recruitment interview evaluation system

Abstract

The present invention provides an artificial intelligence-based recruitment interview evaluation system, comprising: an information acquisition module for acquiring enterprise recruitment information, corresponding applicant resume information, and applicant interview information; a skill evaluation module for extracting job skill characteristics and applicant skill characteristics, and performing skill learning path analysis based on the established industry skill knowledge graph to further obtain the skill potential evaluation result of the applicant; a cultural fit evaluation module for extracting enterprise culture characteristics and applicant culture characteristics, and performing cultural fit analysis based on the enterprise culture characteristics and applicant culture characteristics to obtain the cultural fit evaluation result of the applicant; an interview evaluation module for analyzing the interview performance of the applicant according to the interview video record data of the applicant to obtain the interview evaluation result; and a report module for obtaining the comprehensive evaluation result of the applicant for the enterprise position. The present invention helps enterprises improve the intelligent level of applicant evaluation for specific positions.

Description

Technical Field

[0001] The present invention relates to the technical fields of artificial intelligence and enterprise management, and particularly to a recruitment interview evaluation system based on artificial intelligence. Background Art

[0002] As human resources are increasingly valued, enterprises are facing more and more challenges in talent recruitment.

[0003] When enterprises conduct traditional recruitment interviews, the resumes of candidates are mostly screened and interviewed jointly by the human resources person in charge and the position person in charge. However, the screening effect largely depends on the vision and judgment level of both human resources and position management at the same time. That is, the human resources person in charge needs to have a certain understanding of the relevant technologies of the position, so as to be able to screen out more suitable interview candidates (for example, when simply using the skills required for the position as a hard standard for resume screening, many very potential candidates will be missed); for the position person in charge, they also need to have the awareness of enterprise development and management at the same time, in order to select employees who conform to the company's culture and temperament, so as to improve the adaptability of newly recruited personnel. At the same time, the interview results mainly depend on the subjective judgment of the interviewers, and there will also be a situation where the objectivity of the recruitment results is insufficient.

[0004] Therefore, based on the above situation, it is usually difficult to guarantee the recruitment effect of enterprises at present, which has led to a waste of recruitment human resources and an increase in employment costs, and cannot meet the needs of modern enterprises. Summary of the Invention

[0005] In view of the above problems, the present invention aims to provide a recruitment interview evaluation system based on artificial intelligence.

[0006] The object of the present invention is achieved by the following technical solutions:

[0007] The present invention provides a recruitment interview evaluation system based on artificial intelligence, including an information acquisition module, a skill evaluation module, a cultural fit evaluation module, an interview evaluation module and a report module; wherein,

[0008] The information acquisition module is used to acquire enterprise recruitment information, the corresponding resume information of candidates and the interview information of candidates;

[0009] The skill evaluation module is used to extract the position skill characteristics and the candidate skill characteristics respectively according to the enterprise recruitment information and the candidate resume information; and conduct skill learning path analysis on the position skill characteristics and the candidate skill characteristics based on the established industry skill knowledge graph, and further obtain the skill potential evaluation result of the candidate based on the obtained learning path analysis result;

[0010] The cultural adaptation assessment module is used to extract corporate culture characteristics and candidate culture characteristics based on corporate recruitment information and candidate resume information, and conduct cultural adaptation analysis based on corporate culture characteristics and candidate culture characteristics to obtain the candidate's cultural adaptation assessment results;

[0011] The interview evaluation module is used to analyze the interview performance of the applicant based on the interview video record data of the applicant to obtain the interview evaluation result;

[0012] The report module is used to obtain a comprehensive assessment result of the applicant for the company's position based on the applicant's skill potential assessment results, cultural adaptability assessment results and interview assessment results.

[0013] Preferably, the information acquisition module includes an enterprise recruitment information input unit, an applicant resume acquisition unit and an applicant interview acquisition unit; wherein,

[0014] The enterprise recruitment information input unit is used to obtain enterprise recruitment information by scanning or inputting, wherein the enterprise recruitment information includes recruitment positions, job descriptions corresponding to the recruitment positions, and corporate culture introduction information; wherein the job description includes job responsibilities, qualifications and other requirements;

[0015] The applicant resume acquisition unit is used to acquire the applicant's resume information, wherein the resume information includes the applicant's basic information, skills introduction, work experience and personal introduction, etc.;

[0016] The candidate interview acquisition unit is used to record the candidate's interview video record data during the candidate's job interview.

[0017] Preferably, the skill assessment module includes a skill feature extraction unit, a skill path analysis unit and a skill potential assessment unit; wherein,

[0018] The skill feature extraction unit is used to extract job skill features and candidate skill features according to the enterprise recruitment information and candidate resume information respectively;

[0019] The skill path analysis unit is used to analyze the skill learning path of the job skill characteristics and the candidate skill characteristics based on the established industry skill knowledge graph, and obtain the best learning path between the candidate skill characteristics and the job skill characteristics;

[0020] The skill potential estimation unit is used to calculate the skill potential value of the applicant as a skill potential evaluation result based on the obtained optimal learning path.

[0021] Preferably, the skill assessment module further includes a graph building unit;

[0022] Among them, the graph construction unit is used to build a skill knowledge graph based on the job skill database, industry technical manual database, skill training course database, and talent resume database, and construct a learning relationship network between industry skills through the skill knowledge graph; specifically including:

[0023] 1) Perform entity extraction based on the job skill database, industry technical manual database, and skill training course database to obtain network node information corresponding to skills and construct a skill knowledge graph network; among them, the extracted entities are industry skills, and each point in the constructed knowledge graph network corresponds to an industry skill;

[0024] 2) Calculate the weights between each network point based on the constructed skill knowledge graph network, and the weight calculation function used is:

[0025] ;

[0026] Among them, ω(i, j) represents the connection weight between skill node i and skill node j , sim(i, j) represents the correlation degree between skill i and skill j , obtained based on the talent resume database, where , NumJ represents the total number of talent resumes, Numj(i, j) represents skill i and skill j appearing in the same resume at the same time; Numz(i, j) represents the number of times skill i and skill j appear in the same content during the entity extraction process, NumZ represents the total number of contents; DDEC(i, j) represents the time difference factor between skill i and skill j , obtained based on the time difference between the time information of their first records; DHOT(i, j) represents the popularity difference factor between skill i and skill j , obtained based on the difference in their current appearance frequencies in the content.

[0027] Preferably, the skill path analysis unit includes:

[0028] 1) According to the job skills a and candidate skills b obtained under the same classification;

[0029] 2) Based on the constructed industry skill knowledge graph, retrieve the skills of the candidate from skill b to skill aFor the learning paths, calculate the learning path scores of each path respectively. The learning path scoring function adopted is:

[0030] ;

[0031] In the formula, Point(i, b, a) represents the learning path score of the b th learning path between the skill node a and the skill node i . c, d ∈ i represents that the variable c, d belongs to the adjacent skill nodes in the i th learning path, where the path direction is from c to d ; ω(c, d) represents the connection weight between the skill node c and the skill node d in the industrial skill knowledge graph. sim(c, d) represents the correlation degree between the skill c and the skill d . β(d) represents the learning cost factor of the skill d , where the value of β(d) is the average time required to master the skill d ;

[0032] 3) According to the learning path scores of each learning path, select the path with the maximum learning path score as the best learning path from skill b to skill a.

[0033] Preferably, the skill potential estimation unit includes:

[0034] Calculate the potential value of the applicant's skill corresponding to the recruitment position skill according to the obtained best learning path. The potential value calculation function adopted is:

[0035] ;

[0036] In the formula, Qua(a, b) represents the potential value of the applicant's skill b corresponding to the recruitment position skill a . The variable d represents the skill node on the best learning path b from the skill to the skill a in the skill knowledge graph, where I ; d ≠ b ; ω(d) represents the node weight of the skill node d in the skill knowledge graph. sim(d) represents the skill node I in the best learning path dThe degree of association with the previous skill node; β(d) Indicates the skill d 's learning cost factor; indicates the set cost attenuation coefficient, where ; σ 2 Indicates the corresponding skill a and skill b The node weight variance of all skill nodes under the same classification;

[0037] Based on the potential values of the candidate's skills corresponding to each recruitment position, obtain the skill potential evaluation result of the recruiter.

[0038] Preferably, the cultural fit evaluation module includes:

[0039] Construct an input set from the corporate culture introduction information and the candidate's personal introduction information, and input the input set into the trained cultural fit analysis model to obtain the cultural fit evaluation result of the candidate output by the cultural fit analysis model;

[0040] Among them, the form of the input set is {corporate culture introduction text, candidate personal introduction text};

[0041] Among them, the cultural fit analysis model is built based on the BERT model, including an input layer, an encoding layer, and an output layer connected in sequence. The input layer performs word segmentation, text padding, and character encoding processing on the corporate culture introduction text and the candidate's personal introduction text in the input set respectively, and obtains two character encoding vector sequences of the text; the encoding layer uses 12 consecutive Transformers, and each Transformer uses a multi-head attention mechanism for feature extraction, and a fully connected layer is added after the last Transformer to output a 256-dimensional cultural feature vector; the encoding layer extracts cultural feature vectors from the two character encoding vector sequences of the text respectively, and the output layer compares the similarity of the two obtained cultural feature vectors. The similarity comparison function used is the cos function, and the similarity result of the two is used as the cultural fit evaluation result.

[0042] Preferably, the interview evaluation module includes a video extraction unit, an emotion evaluation unit, a stress evaluation unit, and an interview evaluation unit; among them,

[0043] The video extraction unit is used to obtain the interview video record data of the candidate, and obtain the corresponding video image information and audio information according to the obtained interview video record data; among them, the interview video record data is obtained by shooting with an intelligent camera during the candidate's interview;

[0044] The emotion assessment unit is used to track the facial features of the applicant based on the facial tracking model according to the video image data of the applicant in the first critical time period, and calculate the expression energy value of the applicant according to the tracking results of the facial feature points. The expression energy value calculation function used is:

[0045] ;

[0046] In the formula, EVM represents the expression energy value of the applicant for the interview question. The higher the expression energy value, the more unstable the applicant's emotion; the variable t = 1, 2, … T , t represents the moment in the critical time period, T represents the total duration of the critical time period; ∆P(t) represents t the average change amount of the facial key point coordinates at each moment compared with the previous moment when ∆T t represents t the optical flow change amount of the applicant's facial area at the moment

[0047] The stress assessment unit is used to obtain the stress response value of the applicant based on the audio analysis technology according to the audio data of the applicant in the second critical time period. The stress response value calculation function used is:

[0048] ;

[0049] In the formula, SRC represents the stress response value of the applicant for the interview question. The higher the stress response value, the greater the stress of the applicant; σ(f t ) and μ(f t ) respectively represent the standard deviation and average value of the frequencies of the audio data at each moment of the applicant in the second critical time period, N pauce represents the number of abnormal pauses of the applicant in the second critical time period. Based on the VAD algorithm, the audio data is detected. When the audio gap is greater than the preset standard time T pauce a record of an abnormal pause is made, where T pauce ∈ [0.5s, 1s] ; T represents the total duration of the audio data; α represents the preset voice fluctuation weight, where α∈ [0.4,0.6] ; β represents the preset pause frequency weight, where β∈[0.3,0.7] ;

[0050] The interview evaluation unit is used to calculate an interview evaluation value as the interview evaluation result according to the obtained expression energy value and stress response value, and the interview evaluation value calculation function adopted is:

[0051] ;

[0052] Where FC represents the interview evaluation value of the applicant, γ and δ respectively represent the set normalization weight factors, which are statistically obtained according to the historical maximum values of EVM and SRC respectively, EVM represents the expression energy value of the applicant for the interview question, SRC represents the stress response value of the applicant for the interview question.

[0053] Preferably, the reporting module includes:

[0054] Output separately according to the obtained skill potential evaluation result, cultural fit evaluation result and interview evaluation result of the applicant to generate a comprehensive evaluation result of the applicant; or,

[0055] Conduct a comprehensive evaluation according to the obtained skill potential evaluation result, cultural fit evaluation result and interview evaluation result of the applicant to obtain a comprehensive evaluation result of the applicant.

[0056] The beneficial effects of the present invention are as follows: The present invention proposes a recruitment interview evaluation system based on artificial intelligence, which can, based on the recruitment information of the enterprise and the resume information of the applicant, intelligently evaluate the suitability between the applicant and the position from three dimensions of skill potential, cultural fit and interview evaluation based on artificial intelligence technology. Among them, for the skill characteristics of the applicant, the idea of skill development is combined, thereby improving the adaptability of skill screening for the position and helping to find more potential personnel; by evaluating the cultural fit between the applicant and the corporate culture, the suitability between the applicant's values and the corporate culture can be quantified, which helps to avoid the situation of personnel loss due to insufficient recognition of the corporate culture by new employees due to reasons such as personality. And finally, combined with the interview video record data of the applicant, an objective feedback on the applicant's on-site interview situation is provided, which helps to improve the objective level of feedback on the applicant's interview status. By comprehensively displaying the degree of fit between the applicant and the enterprise through multi-dimensional evaluation results, it helps the enterprise to improve the comprehensiveness, pertinence and objectivity of the evaluation of applicants for specific positions, and also improves the efficiency and intelligent level of enterprise recruitment. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The present invention will be further described with reference to the accompanying drawings. However, the embodiments in the drawings do not constitute any limitation to the present invention. For those of ordinary skill in the art, other drawings can be obtained based on the following drawings without creative efforts.

[0058] Figure 1 It is a framework structure diagram of a recruitment interview evaluation system based on artificial intelligence shown in an embodiment of the present invention;

[0059] Figure 2 is Figure 1 A schematic diagram of the framework structure of the information acquisition module in the embodiment;

[0060] Figure 3 is Figure 1 A schematic diagram of the framework structure of the skill evaluation module in the embodiment. Detailed implementation manners

[0061] The present invention will be further described in combination with the following application scenarios.

[0062] See Figure 1 A recruitment interview evaluation system based on artificial intelligence shown in the embodiment, including an information acquisition module, a skill evaluation module, a cultural fit evaluation module, an interview evaluation module, and a report module; wherein,

[0063] The information acquisition module is used to obtain enterprise recruitment information, corresponding applicant resume information, and applicant interview information;

[0064] The skill evaluation module is used to extract job skill characteristics and applicant skill characteristics respectively according to the enterprise recruitment information and the applicant resume information; and perform skill learning path analysis on the job skill characteristics and the applicant skill characteristics based on the established industry skill knowledge graph, and further obtain the skill potential evaluation result of the applicant based on the obtained learning path analysis result;

[0065] The cultural fit evaluation module is used to extract enterprise culture characteristics and applicant culture characteristics respectively according to the enterprise recruitment information and the applicant resume information, and perform cultural fit analysis based on the enterprise culture characteristics and the applicant culture characteristics to obtain the cultural fit evaluation result of the applicant;

[0066] The interview evaluation module is used to analyze the interview performance of the applicant according to the interview video record data of the applicant to obtain the interview evaluation result;

[0067] The report module is used to obtain the comprehensive evaluation result of the applicant for the enterprise position according to the skill potential evaluation result, the cultural fit evaluation result, and the interview evaluation result of the applicant.

[0068] The above-mentioned embodiment of the present invention proposes a recruitment interview evaluation system based on artificial intelligence, which can be based on the recruitment information of the enterprise and the resume information of the applicant, and based on artificial intelligence technology, intelligently evaluate the adaptability of the applicant to the position from three dimensions of skill potential, cultural adaptability and interview evaluation, wherein the skill characteristics of the applicant are combined with the idea of ​​skill development, thereby improving the adaptability of job skill screening, which is helpful to find more potential personnel; by evaluating the adaptability of the applicant to the corporate culture, the values ​​of the applicant and the degree of suitability of the corporate culture can be quantified, which helps to avoid the situation of personnel loss caused by the lack of recognition of the corporate culture by new employees due to personality and other reasons. And finally, combined with the interview video recording data of the applicant, objective feedback is given to the on-the-spot interview situation of the applicant, which helps to improve the objective level of feedback on the interview status of the applicant. The comprehensive display of the adaptability of the applicant to the enterprise through the multi-dimensional evaluation results helps the enterprise to improve the comprehensiveness, pertinence and objectivity of the evaluation of applicants for specific positions, and also improves the efficiency and intelligence level of enterprise recruitment.

[0069] Among them, the system proposed in the present invention can be built based on a cloud server or a local server built within an enterprise. By establishing a corresponding database as support, and combining the above-mentioned functional modules and on-site equipment to obtain the required data, the system can provide complete recruitment interview evaluation services.

[0070] Preferably, the system also includes a database module; wherein,

[0071] The database module includes the databases required for building or imaging the system, including job skill database, industry technical manual database, skill training course database, talent resume database, etc. The database can be built based on specific data resources, or by grabbing relevant content from existing open source databases.

[0072] Preferably, see Figure 2 The information acquisition module includes an enterprise recruitment information input unit and an applicant resume acquisition unit; wherein,

[0073] The enterprise recruitment information input unit is used to obtain enterprise recruitment information by scanning or inputting, wherein the enterprise recruitment information includes recruitment positions, job descriptions corresponding to the recruitment positions, and corporate culture introduction information; wherein the job description includes job responsibilities, qualifications, other requirements, etc.;

[0074] The applicant resume acquisition unit is used to acquire the applicant's resume information, wherein the resume information includes the applicant's basic information, skills introduction, work experience and personal introduction.

[0075] When a recruitment interview assessment is required, first enter the relevant information of the enterprise's recruitment and the relevant information of the applicant as a basis, so that the system can evaluate the skill potential and cultural fit of the recruiter based on the above information.

[0076] In one scenario, the personal introduction in the applicant's resume information specifically describes the applicant's personality, hobbies, work attitude, etc.

[0077] Preferably, the information acquisition module further includes an applicant interview acquisition unit; among them,

[0078] The applicant interview acquisition unit is used to record the interview video record data of the applicant during the process of the applicant's job interview.

[0079] For the interview assessment of the applicant, the video data of the interview process is recorded during the applicant's completion of the interview as the basis for subsequent interview assessment.

[0080] Among them, in traditional recruitment screening technologies, for example, some artificial intelligence-based establishment screenings are usually based on set skill conditions, such as "for a Java engineer, 5 years or more of Java development experience is required" for hard judgment of skill conditions. That is, when screening resumes, the resumes that meet "5 years or more of Java development experience" are taken as the screening results. However, in the field of programming, if there is another industry veteran who is only proficient in Python, then this veteran will be excluded. But in actual situations, an experienced senior programmer can quickly master the use of Java on the basis of being proficient in Python. Therefore, it will lead to the screening results not conforming to the industry situation and missing potential talents. Therefore, the present invention proposes a method that can analyze the potential value between the skills the applicant has mastered and the skills required for the recruitment position to evaluate, so as to improve the adaptability and effectiveness of talent matching and selection.

[0081] Preferably, refer to Figure 3 , the skill assessment module includes a skill feature extraction unit, a skill path analysis unit, and a skill potential assessment unit; among them,

[0082] The skill feature extraction unit is used to extract the position skill features and the applicant skill features according to the enterprise recruitment information and the applicant resume information respectively;

[0083] The skill path analysis unit is used to perform a skill learning path analysis on the position skill features and the applicant skill features based on the established industry skill knowledge graph to obtain the best learning path between the applicant skill features and the position skill features;

[0084] The skill potential estimation unit is used to calculate the skill potential value of the applicant as the skill potential evaluation result according to the obtained optimal learning path.

[0085] The above-mentioned embodiment of the present invention proposes a technical solution for evaluating the skill potential of applicants. First, skill feature extraction is performed on the energy-saving requirements of the recruitment position and the skills already mastered by the applicant. Based on the extracted features, the optimal learning path between the two is obtained through the skill knowledge graph. By obtaining the learning path, it is possible to screen the fastest path for the applicant to learn the skill requirements of the position with the skills already mastered, so as to further evaluate the potential value of the applicant for the skills that do not meet the requirements based on the optimal learning path. Through the quantification of the potential value, it is helpful to further screen out applicants who have a certain foundation and can quickly master the skills required for the position, and improve the adaptability of the skill potential evaluation to the position requirements. Compared with the traditional resume screening method based on hard criteria, by adding skill potential evaluation, the intelligent level of position skill screening can be improved.

[0086] In a scenario, for the Java skills mastered by the applicant, the calculated optimal learning path for building a distributed system is from Java to Spring Cloud to the distributed system, and the calculated skill potential value is 0.92, which is higher than the preset standard. Therefore, it can be considered that the applicant can master the "distributed system building" with less learning cost based on the currently mastered technology "proficient in Java" to meet the requirements of the corresponding position.

[0087] Preferably, the skill feature extraction unit includes:

[0088] Based on the BERT model, the recruitment position and the corresponding position description are input into the BERT model. The BERT model encodes the position description text, and extracts the corresponding skill keywords as the position skill features through classification; and the applicant's resume information is input into the BERT model to obtain the skill keywords extracted by the BERT model as the applicant's skill features.

[0089] Among them, in addition to the BERT model, the skill feature extraction unit can also select other feature word extraction models to extract the position skill features and the applicant's skill features. For example, keyword extraction models already disclosed in the prior art such as the TextRank model, TF-IDF model, SingleRank model, RAKE model, and YAKE model can be used to implement the extraction of skill feature words. The present invention does not make specific limitations here.

[0090] Preferably, the skill evaluation module further includes a graph construction unit;

[0091] Among them, the graph construction unit is used to construct a skill knowledge graph based on the job skill database, industry technology manual database, skill training course database, and talent resume database, and construct a learning relationship network between industry skills through the skill knowledge graph; specifically including:

[0092] 1) Entity extraction is performed based on the job skill database, industry technology manual database, and skill training course database to obtain network node information corresponding to skills and construct a skill knowledge graph network; among them, the extracted entities are industry skills, and each point in the constructed knowledge graph network corresponds to an industry skill;

[0093] 2) Calculate the weights between each network point based on the constructed skill knowledge graph network, and the weight calculation function used is:

[0094] ;

[0095] Among them, ω(i, j) represents the connection weight between skill node i and skill node j , sim(i, j) represents the correlation degree between skill i and skill j , obtained based on the talent resume database, where , NumJ represents the total number of talent resumes, Numj(i, j) represents skill i and skill j appearing in the same resume at the same time; Numz(i, j) represents the number of times that skill i and skill j appear in the same content during the entity extraction process, NumZ represents the total number of contents; DDEC(i, j) represents the time difference factor between skill i and skill j , obtained based on the time difference between the first recorded time information of the two; DHOT(i, j) represents the popularity difference factor between skill i and skill j , obtained based on the difference in the current (e.g., this year) appearance frequency of the two in the content.

[0096] In one scenario, entity extraction can be implemented using existing data extraction models, such as data extraction models based on models like HMM and HERT. This application does not make specific limitations here.

[0097] Among them, the above embodiments of the present invention also propose a technical solution for building a skill knowledge graph. First, based on data extraction technology, the node information of the knowledge graph network is determined, and then through the proposed weight calculation function, the connection weights between skill nodes are determined to reflect the association degree and popularity relationship between skills, thereby completing the construction of the skill knowledge graph and laying a foundation for subsequent skill potential value evaluation.

[0098] Preferably, the skill path analysis unit includes:

[0099] 1) According to the job skills and applicant skills obtained under the same classification a and b ;

[0100] 2) Based on the built industry skill knowledge graph, retrieve the learning path of the applicant from skill b to skill a . Calculate the learning path scores of each path respectively. The learning path scoring function used is:

[0101] ;

[0102] In the formula, Point(i, b, a) represents the learning path score of the b th learning path from skill node a to skill node i . c, d ∈ i represents that the variable c, d belongs to the adjacent skill nodes in the i th learning path, where the path direction is from c to d ; ω(c, d) represents the connection weight between skill nodes c and d in the industry skill knowledge graph. sim(c, d) represents the association degree between skill c and skill d . β(d) represents the learning cost factor of skill d , where the value of β(d) is the average time (months) required to master skill d ;

[0103] 3) According to the learning path scores of each learning path, select the path with the largest learning path score as the best learning path from skill b to skill a.

[0104] After obtaining the characteristics of the skills currently mastered by the applicant and the skill requirements of the position, based on the network structure in the skill knowledge graph, calculate the best learning path between the two. The present invention also particularly proposes a learning path scoring function, which can be quantified based on the degree of association (difficulty) and learning cost of different learning paths, so as to serve as the basis for evaluating the best learning path. Through the above method, the optimal path from the skills currently mastered by the applicant to the skill requirements of the position can be accurately represented based on the skill knowledge graph, thus providing support for subsequent potential value evaluation.

[0105] Preferably, the skill potential estimation unit includes:

[0106] Calculate the potential value of the applicant's skills corresponding to the skills of the recruitment position according to the obtained best learning path, and the potential value calculation function adopted is:

[0107] ;

[0108] In the formula, Qua(a, b) represents the potential value of the applicant's skills b corresponding to the skills of the recruitment position a , the variable d represents the skill node on the best learning path b from skill a to skill I in the skill knowledge graph, where d ≠ b ; ω(d) represents the node weight of the skill node d in the skill knowledge graph, sim(d) represents the degree of association between the skill node I in the best learning path d and the previous skill node; β(d) represents the learning cost factor of skill d ; represents the set cost attenuation coefficient, where ; σ 2 represents the node weight variance of all skill nodes under the same classification of the corresponding skills a and skill b ;

[0109] Obtain the skill potential evaluation result of the recruiter according to the potential value of the applicant corresponding to each recruitment position skill QU .

[0110] Generally, after obtaining the potential values of the candidate's skills corresponding to each recruitment position, the corresponding skill potential evaluation results are obtained by addition. However, in some other scenarios, the corresponding skill potential evaluation results can also be determined based on the minimum value (only considering the shortest board) or the average value (considering comprehensively) of the potential values of each recruitment skill, and can be reasonably set according to different recruitment ideas and recruitment scenarios.

[0111] Among them, the skills under the same classification pointed out in the present invention, where the classification refers to the skill field. For example, for programming skills, such as Java, distributed systems, cloud computing, etc., all belong to the skills under the same classification. For a qualified position, foreign language skills, such as English, Spanish, etc., may also be required, which belong to the skills under another classification.

[0112] In the above embodiment of the present invention, after obtaining the best learning path between the skills mastered by the candidate and the skill requirements of the position, based on the proposed potential value calculation function, considering factors such as the importance and learning cost of each skill node in the optimal path, the cost and difficulty that the candidate needs to pay to learn the skills required for the position are further quantified, so as to objectively and accurately evaluate the candidate's skill potential. The reliability and intelligence level of the candidate potential consideration are improved. It can provide a basis for the subsequent recruitment evaluation of candidates as an important consideration factor.

[0113] Considering the cultural bias of the candidate, such as the attitude towards work, etc., will all become important factors for whether the candidate can adapt to the position and corporate culture after joining the enterprise. Therefore, the present invention also proposes a method for evaluating the adaptability between the candidate and the corporate culture, so as to improve the pertinence and reliability of the recruitment interview evaluation from the dimension of cultural adaptation.

[0114] Preferably, the cultural adaptation evaluation module includes:

[0115] Construct an input set with the corporate culture introduction information and the candidate's personal introduction information, and input the input set into the trained cultural adaptation analysis model to obtain the cultural adaptation evaluation result of the candidate output by the cultural adaptation analysis model;

[0116] Among them, the form of the input set is {corporate culture introduction text, candidate's personal introduction text};

[0117] Among them, the cultural adaptation analysis model is built based on the BERT model, including an input layer, an encoding layer, and an output layer connected in sequence. The input layer performs word segmentation, text padding, and character encoding on the corporate culture introduction text and the applicant's personal introduction text in the input set respectively, and obtains the character encoding vector sequences of the two texts; the encoding layer uses 12 consecutive Transformers, and each Transformer uses the multi-head attention mechanism for feature extraction, and a fully connected layer is added after the last Transformer to output a 256-dimensional cultural feature vector; the encoding layer extracts the cultural feature vectors from the character encoding vector sequences of the two texts respectively, and the output layer compares the similarity of the two obtained cultural feature vectors. The similarity comparison function used is the cosine function to obtain their similarity. CA The result is used as the cultural adaptation evaluation result.

[0118] In one scenario, taking a 120-word corporate culture introduction text as an example, by inputting the corporate culture introduction text into the cultural adaptation analysis model, 80 words are obtained through word segmentation based on the input layer, and through character encoding processing, an 80×768 character encoding vector sequence is obtained; and through the attention processing of 12 consecutive Transformers, the semantic features of the corporate culture are obtained, and the cultural feature vectors are extracted through the fully connected layer. The activation function used in the fully connected layer is Sigmoid, and the finally output 256-dimensional corporate culture feature vector is obtained; similarly, the above processing is performed on the applicant's personal introduction text to obtain the 256-dimensional applicant's cultural feature vector corresponding to the applicant's personal introduction text; finally, the similarity between the two (cultural feature vectors) is calculated based on the similarity calculation function, so as to obtain the similarity between the applicant's cultural tendency and the corporate culture tendency; the corporate culture tendency can be understood as including, for example, "work-life balance" or "pursuing competition", etc.

[0119] In the above embodiment of the present invention, by extracting the cultural tendency from the corporate culture introduction information and the applicant's personal introduction information, the cultural attributes of the applicant can be evaluated from the perspective of corporate culture adaptation, so as to assist in finding applicants with cultural adaptation, improve the adaptation degree of the applicant to the enterprise after subsequent employment, reduce the probability of personnel mismatch and short-term personnel loss caused by cultural conflicts, and also reduce the overall cost of talent recruitment.

[0120] Preferably, the interview evaluation module includes a video extraction unit, an emotion evaluation unit, a stress evaluation unit, and an interview evaluation unit; among them,

[0121] The video extraction unit is used to obtain the interview video record data of the applicant, and obtain the corresponding video image information and audio information according to the obtained interview video record data; wherein the interview video record data is obtained by shooting with an intelligent camera during the applicant's interview process;

[0122] The emotion assessment unit is used to track the facial features of the applicant based on the facial tracking model according to the video image data of the applicant in the first key time period, and calculate the expression energy value of the applicant according to the tracking results of the facial feature points. The expression energy value calculation function adopted is:

[0123] ;

[0124] In the formula, EVM represents the expression energy value of the applicant for the interview question. The higher the expression energy value, the more unstable the applicant's emotion; the variable t = 1, 2, … T , t represents the moment in the key time period, T represents the total duration of the key time period; ∆P(t) represents t the average change amount of the facial key point coordinates at each moment compared to the previous moment when ∆T t represents t the optical flow change amount of the applicant's facial area at the moment

[0125] The stress assessment unit is used to obtain the stress response value of the applicant based on the audio analysis technology according to the audio data of the applicant in the second key time period. The stress response value calculation function adopted is:

[0126] ;

[0127] In the formula, SRC represents the stress response value of the applicant for the interview question. The higher the stress response value, the greater the stress of the applicant; σ(f t ) and μ(f t ) respectively represent the standard deviation and average value of the frequencies of the audio data at each moment of the applicant in the second key time period, N pauce represents the number of abnormal pauses of the applicant in the second key time period. Among them, the audio data is detected based on the VAD algorithm. When the audio gap is greater than the preset standard time T pauce , record an abnormal pause once. Among them, T pauce ∈ [0.5s, 1s] ;T Represents the total duration of the audio data; α Represents the preset voice fluctuation weight, where α∈ [0.4,0.6] ; β Represents the preset pause frequency weight, where β∈[0.3,0.7] ;

[0128] The interview evaluation unit is used to calculate the interview evaluation value as the interview evaluation result according to the obtained expression energy value and stress response value, and the interview evaluation value calculation function adopted is:

[0129] ;

[0130] Where FC Represents the interview evaluation value of the applicant, γ and δ respectively represent the set normalization weight factors, which are statistically obtained according to the historical maximum values of EVM and SRC respectively, EVM Represents the expression energy value of the applicant for the interview question, SRC Represents the stress response value of the applicant for the interview question.

[0131] In a scenario, the face tracking model is implemented by a face recognition and tracking model based on tracking 68 face key points.

[0132] For the interview scenario, the first key time period is 1 - 3 s after the interviewer asks a question. The video record data corresponding to the corresponding moments are intercepted respectively, and the corresponding expression energy value calculation and the emotion evaluation result can be obtained. The second key time period is the time period from when the applicant starts to answer the question asked by the interviewer to the end of answering the question.

[0133] In a scenario, the interview video record data can be obtained by setting up a camera at the interview site to record the interview process, or by recording and extracting the video interview process, so as to obtain the interview video record data.

[0134] The above embodiments of the present invention provide a technical solution for evaluating the interview status of candidates based on the interview video record data of candidates. First, based on the obtained interview video record data, images and audio of the video are extracted respectively, laying a foundation for subsequent emotion evaluation and stress evaluation. Based on the progress of the interview, key time periods are also marked, so as to conduct targeted evaluation of the interview status of candidates. Among them, based on the proposed expression energy calculation function, the facial expression change characteristics of candidates after answering interview questions are extracted, and the expression mutation degree of candidates is reflected through expression energy, so as to reflect the emotional stability state of candidates when facing recruitment questions (when candidates are in a tense state, due to the instantaneous reaction of facial muscles, by capturing the characteristics of this instantaneous facial expression change, the tension degree of candidates for the interview can be reflected, and thus the interview status of candidates can be reflected, such as being nervous or hiding). Further, based on the proposed stress response value calculation function, the audio data of candidates during the answering process of interview questions can be analyzed, and the stress state of the interviewee's answer to the interview question can be evaluated through the frequency characteristics change and pause characteristics of the audio feature signal. Finally, the interview status is evaluated according to the obtained expression energy value and stress response value, so as to give an objective feedback on the interview status of candidates, avoiding the deviation of the interview evaluation of candidates due to the subjective evaluation of the interviewer, and helping to improve the objective level of recruitment interview evaluation.

[0135] Preferably, the reporting module includes:

[0136] Output the obtained skill potential evaluation result, cultural fit evaluation result and interview evaluation result of the candidate respectively to generate the comprehensive evaluation result of the candidate; or,

[0137] Conduct a comprehensive evaluation based on the obtained skill potential evaluation result, cultural fit evaluation result and interview evaluation result of the candidate to obtain the comprehensive evaluation result of the candidate; among them, the comprehensive evaluation function used is:

[0138] ;

[0139] In the formula, Score represents the comprehensive evaluation value of the candidate, QU represents the skill potential evaluation result of the candidate, CA represents the cultural fit evaluation result of the candidate, FC represents the interview evaluation result of the candidate. ω 1 、 ω 2 and ω 3They respectively represent the set skill potential weight factor, cultural fit weight factor, and interview weight factor.

[0140] Among them, according to actual needs, the comprehensive evaluation result of the applicant can be output separately based on the three evaluation results for the applicant, so as to obtain the applicant's interview evaluation report; among them, the larger the values of the three evaluation results, the better the applicant's results in the corresponding evaluation items.

[0141] In another scenario, it is also possible to further comprehensively quantify based on the three evaluation results to obtain the interview evaluation score in the same dimension, so as to intuitively and directly represent the situation of the applicant.

[0142] Based on the output evaluation results, it can provide data support for the enterprise to determine the final candidate for the position recruitment, and assist the manager in selecting the most suitable manager for the enterprise position.

[0143] In one scenario, the three evaluation results are respectively output in a normalized percentage system. For example, the skill potential evaluation result is 90, the cultural fit evaluation result is 20, and the interview evaluation result is 80. If the lowest sub-item is lower than the lowest standard (for example, 30), it is determined that the applicant is not suitable for the enterprise position; or for example, the skill potential evaluation result is 70, the cultural fit evaluation result is 75, and the interview evaluation result is 80, all of which meet the set minimum standards (for example, not less than 70), so the applicant is hired.

[0144] In another scenario, the three evaluation results can also be part of many evaluation results, that is, combined with the evaluation results obtained by other means to comprehensively evaluate the suitability of the applicant for the position.

[0145] In another scenario, the comprehensive evaluation value of the applicant obtained can be used as a hard indicator (for example, it needs to be higher than the minimum standard) or a reference indicator to affect the decision on whether the applicant is finally hired, so as to meet the application requirements in different recruitment scenarios and recruitment positions.

[0146] It should be noted that in each embodiment of the present invention, each functional unit / module can be integrated in a processing unit / module, or each unit / module can exist physically alone, or two or more units / module can be integrated in one unit / module. The above integrated unit / module can be implemented in the form of hardware or in the form of a software functional unit / module.

[0147] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any appropriate combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: application specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field programmable gate array (FPGA), processor, controller, microcontroller, microprocessor, other electronic units designed to implement the functions described herein, or a combination thereof. For software implementation, part or all of the processes of the embodiments can be completed by instructing the relevant hardware through a computer program. When implemented, the above program can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. The computer-readable medium includes computer storage media and communication media, where the communication media includes any medium that facilitates the transfer of a computer program from one place to another. The storage media can be any available medium that can be accessed by a computer. The computer-readable medium can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disc storage, magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the protection scope of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. An artificial intelligence-based recruitment interview evaluation system, characterized in that, It includes an information acquisition module, a skill evaluation module, a cultural fit evaluation module, an interview evaluation module, and a reporting module. Among them, The information acquisition module is used to obtain enterprise recruitment information, the corresponding resume information of candidates, and the interview information of candidates. The skill evaluation module is used to extract job skill characteristics and candidate skill characteristics respectively according to the enterprise recruitment information and the candidate resume information; and conduct a skill learning path analysis on the job skill characteristics and candidate skill characteristics based on the established industry skill knowledge graph, and further obtain the skill potential evaluation result of the candidate based on the obtained learning path analysis result. The cultural fit evaluation module is used to extract enterprise culture characteristics and candidate culture characteristics respectively according to the enterprise recruitment information and the candidate resume information, and conduct a cultural fit analysis based on the enterprise culture characteristics and candidate culture characteristics to obtain the cultural fit evaluation result of the candidate. The interview evaluation module is used to analyze the interview performance of the candidate according to the interview video record data of the candidate to obtain the interview evaluation result. The reporting module is used to obtain the comprehensive evaluation result of the candidate for the enterprise position according to the skill potential evaluation result, the cultural fit evaluation result, and the interview evaluation result of the candidate. Among them, the skill evaluation module includes a graph construction unit, a skill characteristic extraction unit, a skill path analysis unit, and a skill potential evaluation unit. Among them, The graph construction unit is used to build a skill knowledge graph according to the job skill database, the industry technical manual database, the skill training course database, and the talent resume database, and construct a learning relationship network between industry skills through the skill knowledge graph. The skill characteristic extraction unit is used to extract job skill characteristics and candidate skill characteristics respectively according to the enterprise recruitment information and the candidate resume information. The skill path analysis unit is used to conduct a skill learning path analysis on the job skill characteristics and candidate skill characteristics based on the established industry skill knowledge graph to obtain the best learning path between the candidate skill characteristics and the job skill characteristics. The skill potential estimation unit is used to calculate the skill potential value of the candidate as the skill potential evaluation result according to the obtained best learning path. Among them, the graph construction unit specifically includes: 1) Entity extraction is performed based on the job skill database, the industry technical manual database, and the skill training course database to obtain network node information corresponding to skills and construct a skill knowledge graph network; among them, the extracted entity is an industry skill, and each point in the constructed knowledge graph network corresponds to an industry skill. 2) Calculate the weights between each network point based on the constructed skill knowledge graph network, and the weight calculation function used is: ； Among them, ω(i,j) represents the connection weight between skill nodes i and skill node j ; sim(i,j) represents the correlation degree between skill i and skill j , which is obtained based on the talent resume database. Among them , NumJ represents the total number of talent resumes, Numj(i,j) represents the total number of times that skill i and skill j appear in the same resume; Numz(i,j) represents the number of times that skill i and skill j appear in the same content during entity extraction, NumZ represents the total number of contents; DDEC(i,j) represents the time difference factor between skill i and skill j , which is obtained according to the difference in the time information when they are first recorded; DHOT(i,j) represents the popularity difference factor between skill i and skill j , which is obtained according to the difference in their current appearance frequencies in the content; Among them, the skill path analysis unit includes: 1) According to the job skills under the same classification obtained a and the candidate skills b ; 2) Retrieve the learning path of the applicant from skill b to skill a based on the established industry skill knowledge graph, and calculate the learning path scores of each path respectively. The learning path scoring function used is as follows: ； In the formula, Point(i,b,a) represents the learning path score of the b th learning path between skill node a and skill node i ; c,d∈i represents that the variable c, d belongs to adjacent skill nodes in the i th learning path, where the path direction is from c to d ; ω(c,d) represents the connection weight between skill node c and skill node d in the industrial skill knowledge graph; sim(c,d) represents the correlation degree between skill c and skill d ; β(d) represents the learning cost factor of skill d , where the value of β(d) is the average time required to master skill d ; 3) According to the learning path scores of each learning path, select the path with the largest learning path score as the best learning path from skill b to skill a. Among them, the skill potential estimation unit includes: According to the obtained optimal learning path, the potential value of the candidate's skills corresponding to the recruitment position skills is calculated, and the potential value calculation function used is: ； In the formula, Qua(a,b) represents the skills of the applicant b corresponding to the skills of the recruitment position a The potential value of, the variable d represents the skills in the skill knowledge graph b to skill a The best learning path of I The skill nodes on, where d≠b ; ω(d) represents the skill node d The node weight of in the skill knowledge graph, sim(d) represents the best learning path I The skill nodes in d The correlation degree between and the previous skill node; β(d) represents the skill d The learning cost factor of; represents the set cost attenuation coefficient, where ; σ 2 represents the corresponding skills a and skills b The node weight variance of all skill nodes under the same classification of; According to the potential value of the applicant's skills corresponding to each recruitment position, the recruiter's skill potential assessment result is obtained.

2. The recruitment interview evaluation system based on artificial intelligence according to claim 1, characterized in that, The information acquisition module includes an enterprise recruitment information input unit, an applicant resume acquisition unit, and an applicant interview acquisition unit; among them, The enterprise recruitment information input unit is used to obtain enterprise recruitment information by scanning or inputting, wherein the enterprise recruitment information includes recruitment positions, job descriptions corresponding to the recruitment positions, and corporate culture introduction information; wherein the job description includes job responsibilities, qualifications and other requirements; The applicant resume acquisition unit is used to acquire the applicant's resume information, wherein the resume information includes the applicant's basic information, skills introduction, work experience and personal introduction; The candidate interview acquisition unit is used to record the candidate's interview video record data during the candidate's job interview.

3. The recruitment interview evaluation system based on artificial intelligence according to claim 1, wherein The cultural fit assessment model includes: The company culture introduction information and the applicant's personal introduction information are used to construct an input set, and the input set is input into the trained cultural adaptation analysis model to obtain the applicant's cultural adaptation evaluation result output by the cultural adaptation analysis model; The input set is in the form of {corporate culture introduction text, applicant personal introduction text}; Among them, the cultural adaptation analysis model is built based on the BERT model, including an input layer, an encoding layer and an output layer connected in sequence, wherein the input layer performs word segmentation, text completion and character encoding processing on the corporate culture introduction text and the applicant's personal introduction text in the input set, respectively, and obtains the character encoding vector sequence of the two texts respectively; the encoding layer adopts 12 consecutive Transformers, each of which adopts a multi-head attention mechanism for feature extraction, and a fully connected layer is added after the last Transformer to output a 256-dimensional cultural feature vector; the encoding layer extracts cultural feature vectors from the character encoding vector sequences of the two texts respectively, and the output layer compares the similarity of the two cultural feature vectors obtained, wherein the similarity comparison function adopted is the cos function, and the similarity result of the two is obtained as the cultural adaptation evaluation result.

4. An artificial intelligence-based recruitment interview evaluation system according to claim 1, characterized in that, The interview assessment module includes a video extraction unit, an emotion assessment unit, a stress assessment unit and an interview assessment unit; among them, The video extraction unit is used to obtain the interview video recording data of the applicant, and obtain the corresponding video image information and audio information according to the obtained interview video recording data; wherein the interview video recording data is based on the video recording obtained by the smart camera during the interview of the applicant; The emotion evaluation unit is used to track the facial features of the applicant based on the video image data of the applicant in the first key time period based on the face tracking model, and calculate the expression energy value of the applicant based on the tracking results of the facial feature points, wherein the expression energy value calculation function used is: ； In the formula, EVM represents the facial expression energy value of the applicant in response to the interview question. The higher the facial expression energy value, the more unstable the applicant's emotion; the variable t=1,2,…T , t represents the moment within the key time period, T represents the total duration of the key time period; ∆P(t) represents t the average change amount of the facial key point coordinates at each moment compared to the previous moment, ∆T t represents t the optical flow change amount of the applicant's facial area at the moment The pressure assessment unit is used to obtain the stress response value of the applicant based on the audio analysis technology according to the audio data of the applicant in the second key time period. The stress response value calculation function adopted is as follows: ； In the formula, SRC represents the stress response value of the applicant for the interview question. The higher the stress response value, the greater the stress of the applicant; σ(f t ) and μ(f t ) respectively represent the standard deviation and the average value of the frequencies of the audio data at each moment during the second key time period of the applicant, N pauce represents the number of abnormal pauses of the applicant during the second key time period. Based on the VAD algorithm, the audio data is detected. When the audio gap is greater than the preset standard time T pauce , an abnormal pause is recorded, where T pauce ∈[0.5s,1s] ; T represents the total duration of the audio data; α represents the preset voice fluctuation weight, where α∈[0.4, 0.6] ; β represents the preset pause frequency weight, where β∈[0.3,0.7] ; The interview assessment unit is used to calculate the interview assessment value as the interview assessment result according to the obtained expression energy value and stress response value. The interview assessment value calculation function adopted is as follows: ; wherein FC represents the interview evaluation value of the applicant, γ and δ respectively represent the set normalization weight factors, which are statistically obtained according to the historical maximum values of EVM and SRC respectively, EVM represents the expression energy value of the applicant for the interview question, SRC represents the stress response value of the applicant for the interview question.

5. The recruitment interview evaluation system based on artificial intelligence according to claim 1, wherein The report module includes: Outputting separately according to the obtained skill potential assessment result, cultural fit assessment result and interview assessment result of the applicant to generate the comprehensive assessment result of the applicant; or, Conducting a comprehensive evaluation according to the obtained skill potential assessment result, cultural fit assessment result and interview assessment result of the applicant to obtain the comprehensive assessment result of the applicant.

Images