Recruitment method and device

Abstract

The invention provides a recruitment method and device. The method comprises the steps of collecting video stream information and audio stream information of candidates of a to-be-recruitment post in an interview process and a structured text generated based on the audio stream information; based on the video stream information, the audio stream information and the structured text, determining a comprehensive emotion vector of the candidate; determining post skill vectors of the candidates based on the structured text; determining an emotion matching degree based on the post emotion demand vector of the to-be-recruited post and the comprehensive emotion vector; determining a skill matching degree based on the post skill demand vector and the post skill vector of the to-be-recruited post; and according to the emotion matching degree and the skill matching degree, determining the matching degree between the candidate and the to-be-recruited post. Through emotion and skill two-dimensional comprehensive evaluation, the intelligent level of the recruitment method is greatly improved, the recruitment decision is more scientific and accurate, and the quality and efficiency of enterprise talent selection are effectively improved.

Description

Technical Field

[0001] This application relates to the field of computerization technology, and in particular to a recruitment method and apparatus. Background Technology

[0002] Recruitment refers to the systematic process by which companies screen and select qualified candidates based on job requirements to meet the organization's human resource needs across different business areas. Recruitment is not only the foundation of a company's talent allocation but also crucial for ensuring its development and competitiveness. Through effective recruitment, companies can acquire employees with suitable skills and who fit the team culture, thereby improving work efficiency, enhancing innovation capabilities, and strengthening the company's overall competitiveness.

[0003] However, existing recruitment methods have many drawbacks, affecting the accuracy of talent matching and recruitment efficiency. First, many traditional recruitment systems rely solely on resume keyword matching or experience screening, ignoring the impact of candidates' emotional characteristics on job suitability. Although some systems attempt to obtain more information through text sentiment analysis, they remain limited to a single modality and cannot fully capture the deeper emotional signals such as candidates' tone of voice and facial micro-expressions during the interview process.

[0004] In summary, existing recruitment methods result in a lack of flexibility in the talent selection process and an inability to accurately assess the overall match between candidates and positions. Summary of the Invention

[0005] This application provides a recruitment method and apparatus to at least partially solve the above-mentioned problems.

[0006] Firstly, this application provides a recruitment method, including:

[0007] Determine the emotional needs vector and skill requirements vector for the positions to be recruited;

[0008] Collect video stream information, audio stream information, and structured text generated based on the audio stream information of the current candidate during the interview process for the position to be recruited;

[0009] Based on the video stream information, the audio stream information, and the structured text, determine the comprehensive sentiment vector of the current candidate;

[0010] Based on the structured text, determine the job skill vector of the current candidate;

[0011] Based on the job-related emotional needs vector and the comprehensive emotional vector, the emotional matching degree is determined;

[0012] Based on the job skill requirement vector and the job skill vector, the skill matching degree is determined;

[0013] The matching degree between the current candidate and the job opening is determined based on the emotional matching degree and the skill matching degree.

[0014] Preferably,

[0015] The determination of the job affective requirement vector and job skill requirement vector for the position to be recruited includes:

[0016] From the preset enterprise management information, at least one target emotional feature corresponding to the job to be recruited and the emotional intensity corresponding to each target emotional feature are determined. The enterprise management information includes the enterprise strategic document of the enterprise development plan within the target time period, the job responsibilities, skill requirements and emotional requirements of the job to be recruited, and historical recruitment data within the preset time period.

[0017] Normalize each of the target emotional features and the corresponding emotional intensity to construct the job emotional requirement vector for the job to be recruited.

[0018] From the enterprise management information, determine at least one target skill item and the corresponding level of mastery for the position to be recruited;

[0019] Based on the enterprise management information, determine the weight corresponding to each of the target skill items;

[0020] The mastery level of each objective and its corresponding weight are normalized to construct the job skill requirement vector for the position to be recruited.

[0021] Preferably,

[0022] The step of determining at least one target skill item and the normalized level of mastery of the target skill item corresponding to the job position to be recruited from the enterprise management information includes:

[0023] From the enterprise management information, extract at least one core skill keyword and the corresponding target mastery level for the job to be recruited, and use the core skill keyword as the target skill item.

[0024] Preferably,

[0025] The step of determining the weight corresponding to each target skill item based on the enterprise management information includes:

[0026] For each of the target skill items, when the enterprise management information is obtained, the weight corresponding to the target skill item is determined according to the following formula, and the weight in the enterprise management information is updated according to the determined weight:

[0027] W j =argmax P(W j |D)

[0028] Among them, W j Let P(W) represent the weight corresponding to the j-th target skill item, D represent the enterprise management information, and P(W) represent the weight corresponding to the j-th target skill item. j |D) The weight of the representation, given the enterprise management information, is W. j The posterior probability.

[0029] Preferably,

[0030] Determining the comprehensive sentiment vector of the current candidate based on the video stream information, the audio stream information, and the structured text includes:

[0031] In chronological order, the intensity value of at least one facial action unit of the current candidate is extracted from the video stream information, wherein the intensity value of the facial action unit represents the expression state of the corresponding facial organ of the current candidate.

[0032] Based on the temporal changes in the intensity values ​​of the facial motion units, a micro-expression fluctuation feature vector is generated;

[0033] Extract Mel-frequency cepstral coefficients from the audio stream information;

[0034] Based on the Mel frequency cepstral coefficients, the voice emotion state category of the current candidate is identified;

[0035] Based on the voice emotion state category, a voice emotion feature vector is determined to characterize the current candidate's emotional state;

[0036] Semantic analysis is performed on the structured text to identify the current candidate's sentiment polarity value and corresponding confidence level, wherein the sentiment polarity value represents the current candidate's emotional tendency.

[0037] Based on at least one sentiment keyword in the preset enterprise management information, the structured text is analyzed and the corresponding target sentiment keyword in the structured text is identified, wherein the sentiment keyword is a word that can reflect the personal feelings and attitudes of the current candidate;

[0038] Determine the weights corresponding to the target sentiment keywords from the enterprise management information;

[0039] The emotional polarity value, the confidence level, and the weights corresponding to each of the target emotional keywords are normalized to construct a text emotional feature vector.

[0040] Based on the micro-expression fluctuation feature vector, the voice emotion feature vector, and the text emotion feature vector, the comprehensive emotion vector of the current candidate is determined.

[0041] Preferably,

[0042] The generation of micro-expression fluctuation feature vectors based on the temporal changes in the intensity values ​​of the facial motion units includes:

[0043]

[0044] Where Fv represents the micro-expression fluctuation feature vector, n is the total number of facial action units, and AU i W represents the intensity value of the i-th facial action unit. i The weight of the i-th facial action unit in emotion judgment is represented by t, and the interview duration of the current candidate is represented by t.

[0045] Preferably,

[0046] The step of determining the voice emotion feature vector to represent the current candidate's emotional state based on the voice emotion state category includes:

[0047]

[0048] Where Fa represents the speech emotion feature vector, K is the total number of speech emotion categories, and w k The weight P(emotion) represents the weight of the k-th type of voice emotion state in the preset enterprise management information. k ) represents the predicted probability of the k-th speech emotion state category.

[0049] Preferably,

[0050] The determination of the comprehensive sentiment vector of the current candidate based on the micro-expression fluctuation feature vector, the voice sentiment feature vector, and the text sentiment feature vector includes:

[0051] The comprehensive sentiment vector of the current candidate is determined based on the following formula:

[0052] F fusion =α·Fv+β·Fa+γ·Ft

[0053] Among them, F fusion The comprehensive emotion vector of the current candidate is represented by Fv, the micro-expression fluctuation feature vector is represented by Fa, the voice emotion feature vector is represented by Ft, the text emotion feature vector is represented by α, the weight of the micro-expression fluctuation feature vector is represented by β, the weight of the voice emotion feature vector is represented by γ, and α+β+γ=1, and α, β and γ are all greater than 0.

[0054] Preferably,

[0055] Determining the job skill vector of the current candidate based on the structured text includes:

[0056] Identify at least one skill item of the current candidate from the current candidate's resume information and the structured text, and assess the mastery level of each skill item;

[0057] The weights corresponding to each skill item in the job skill requirement vector are normalized to construct the job skill vector.

[0058] Preferably,

[0059] The process of determining the emotional matching degree based on the job emotional needs vector and the comprehensive emotional vector includes:

[0060] Obtain at least one target sentiment feature and corresponding sentiment intensity for at least one low-performing employee corresponding to the job opening;

[0061] For each low-performing employee, the target emotional features and corresponding emotional intensities of the employee are normalized to construct a negative ideal job emotional vector.

[0062] Calculate the mean of the negative ideal job emotion vector for each of the aforementioned positions;

[0063] The emotional matching degree is determined based on the following formula and the mean of the negative ideal job emotional vector:

[0064]

[0065] Among them, Sim emotion F represents the degree of emotional matching. fusion Characterizing the comprehensive sentiment vector, The vector representing the emotional needs of the job position. The mean of the negative ideal job emotion vector is represented by i, where i represents the i-th emotion feature dimension and n represents the number of emotion feature dimensions.

[0066] Preferably,

[0067] The process of determining the skill matching degree based on the job skill requirement vector and the job skill vector includes:

[0068] The skill matching degree is determined based on the following formula:

[0069]

[0070] Among them, Sim skill S represents the skill matching degree. candi，i S represents the job skill vector. jobThe vector represents the skill requirement for the job position; m represents the total number of skills, and i represents the i-th skill.

[0071] Preferably,

[0072] Determining the match between the current candidate and the job opening based on the emotional compatibility and the skill compatibility includes:

[0073] The matching degree between the current candidate and the job opening is determined based on the following formula:

[0074] Score = λ·Sim skill +(1-λ)·Sim emotion

[0075] Wherein, Score represents the degree of match between the current candidate and the job opening, Sim skill Sim represents the skill matching degree. emotion The emotional matching degree is represented by λ, and the preset skill matching degree is represented by its weight in the job position to be recruited.

[0076] Secondly, this application provides a recruitment device, comprising:

[0077] The dynamic modeling module is used to determine the job affective requirement vector and job skill requirement vector for the position to be recruited.

[0078] The data acquisition module is used to collect video stream information, audio stream information, and structured text generated based on the audio stream information of the current candidate for the position to be recruited during the interview process.

[0079] The feature extraction module is used to determine the comprehensive sentiment vector of the current candidate based on the video stream information, the audio stream information, and the structured text collected by the data acquisition module; and to determine the job skill vector of the current candidate based on the structured text.

[0080] The matching decision module is used to determine the emotional matching degree based on the job emotional requirement vector determined by the dynamic modeling module and the comprehensive emotional vector determined by the feature extraction module; to determine the skill matching degree based on the job skill requirement vector and the job skill vector; and to determine the matching degree between the current candidate and the job to be recruited based on the emotional matching degree and the skill matching degree.

[0081] This application provides a recruitment method and apparatus. By comprehensively considering the multimodal emotional signals of candidates during the interview process (such as video streams, audio streams, and structured text), as well as the emotional and skill requirements of the position, a more comprehensive and accurate job matching can be achieved. Compared with traditional resume keyword matching or single sentiment analysis methods, this invention can capture and analyze the candidate's emotional state in real time, identify key emotional characteristics such as emotional stability and stress resistance, and effectively improve the accuracy of emotional fit assessment. Simultaneously, by combining voice and video information from the interview, the candidate's emotional tendencies can be assessed from multiple perspectives, avoiding the one-sidedness of emotional assessment. By dynamically assessing the matching degree between the emotional requirements of the position and the candidate's emotional vector, the recruitment process is ensured to be more flexible and personalized, providing companies with candidate selections that better meet the job requirements. Furthermore, the skill matching assessment based on structured text analysis further enhances the accurate matching of skill requirements and candidate professional abilities. This application, through a comprehensive assessment of both emotional and skill dimensions, greatly improves the intelligence level of the recruitment method, making recruitment decisions more scientific and accurate, thereby effectively improving the quality and efficiency of talent selection for enterprises. Attached Figure Description

[0082] Figure 1 This is a flowchart illustrating a recruitment method provided in one embodiment of this application;

[0083] Figure 2 This is a flowchart of another recruitment method provided in one embodiment of this application;

[0084] Figure 3 This is a schematic diagram of a recruitment device provided in one embodiment of this application. Detailed Implementation

[0085] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0086] Figure 1 A recruitment method provided in one embodiment of this application specifically includes the following steps:

[0087] Step 101: Determine the job affective needs vector and job skill needs vector for the position to be recruited;

[0088] Step 102: Collect video stream information, audio stream information, and structured text generated based on audio stream information of the current candidate for the position to be recruited during the interview process;

[0089] Step 103: Determine the comprehensive sentiment vector of the current candidate based on video stream information, audio stream information, and structured text;

[0090] Step 104: Determine the job skill vector of the current candidate based on the structured text;

[0091] Step 105: Determine the emotional matching degree based on the job emotional needs vector and the comprehensive emotional vector;

[0092] Step 106: Determine the skill matching degree based on the job skill requirement vector and the job skill vector;

[0093] Step 107: Determine the match between the current candidate and the position to be recruited based on the emotional match and skill match.

[0094] In this application embodiment, by comprehensively considering the multimodal emotional signals of candidates during the interview process (such as video streams, audio streams, and structured text), as well as the emotional and skill requirements of the position, a more comprehensive and accurate job matching can be achieved. Compared with traditional resume keyword matching or single sentiment analysis methods, this invention can capture and analyze the candidate's emotional state in real time, identify key emotional characteristics such as emotional stability and stress resistance, and effectively improve the accuracy of emotional fit assessment. Simultaneously, by combining voice and video information from the interview, the candidate's emotional tendencies can be assessed from multiple perspectives, avoiding the one-sidedness of emotional assessment. By dynamically assessing the matching degree between the emotional requirements of the position and the candidate's emotional vector, the recruitment process is ensured to be more flexible and personalized, providing companies with candidate selections that better meet the job requirements. Furthermore, the skill matching assessment based on structured text analysis further enhances the accurate matching of skill requirements and the candidate's professional abilities. This application, through a comprehensive assessment of both emotional and skill dimensions, greatly improves the intelligence level of the recruitment method, making recruitment decisions more scientific and accurate, thereby effectively improving the quality and efficiency of talent selection for enterprises.

[0095] In one embodiment of this application, determining the job affective requirement vector and job skill requirement vector for the position to be recruited includes:

[0096] From the preset enterprise management information, determine at least one target emotional characteristic and the emotional intensity corresponding to each target emotional characteristic for the job to be recruited. The enterprise management information includes the enterprise strategic document of the enterprise development plan within the target time period, the job responsibilities, skill requirements and emotional requirements of the job to be recruited, and historical recruitment data within the preset time period.

[0097] Normalize the emotional features and corresponding emotional intensities of each target to construct a job emotional requirement vector for the position to be recruited.

[0098] From the enterprise management information, identify at least one target skill item and the corresponding level of mastery for the position to be recruited;

[0099] Based on enterprise management information, determine the weight corresponding to each target skill item;

[0100] The mastery level of each objective and its corresponding weight are normalized to construct a job skill requirement vector for the position to be recruited.

[0101] In this embodiment, dynamic and intelligent modeling of job requirements is achieved by automatically extracting and constructing job emotional demand vectors and job skill demand vectors from enterprise management information. This method comprehensively utilizes multi-source information such as enterprise strategic documents, job responsibilities, and historical recruitment data to fully reflect the emotional tendencies and skill structure requirements of the enterprise at a specific development stage, thus avoiding the problems of relying on manual weight setting and outdated updates in traditional job modeling. Through normalization of target emotional characteristics and their corresponding intensities, the requirements for different emotional dimensions (such as teamwork, resilience, and innovation) can be quantified, achieving accurate expression of emotional needs. Simultaneously, dynamic modeling of job skill items and their weights can automatically optimize skill requirement configuration based on enterprise strategic adjustments or organizational structure changes, thereby improving the timeliness and scientific rigor of job descriptions. This method not only enhances the adaptability and intelligence of recruitment methods but also provides a more accurate computational foundation for subsequent candidate matching, significantly enhancing the reliability and practical value of recruitment matching results.

[0102] In one embodiment of this application, determining at least one target skill item and its normalized mastery level for the position to be recruited from enterprise management information includes:

[0103] Extract at least one core skill keyword and the corresponding level of mastery of the target skills for the job to be recruited from the enterprise management information, and use the core skill keyword as the target skill item.

[0104] In this embodiment, by extracting core skill keywords from enterprise management information and using them as target skill items, the core skills required for a position can be accurately identified and defined. This method normalizes the core skills and target mastery levels, ensuring the quantification and standardization of skill requirements, making the skill requirements of the position clearer and more objective. This keyword-based extraction method avoids matching biases caused by vague job descriptions or unclear skill requirements in traditional recruitment processes. It allows for dynamic adjustment of skill items according to the specific needs of the enterprise, ensuring a high degree of alignment between job requirements and corporate strategy. By accurately defining target skill items and mastery levels, the accuracy and reliability of the recruitment system in the skill matching process are greatly improved, providing a more scientific basis for candidate selection, thereby enhancing overall recruitment efficiency and talent matching quality.

[0105] In one embodiment of this application, the weight corresponding to each target skill item is determined based on enterprise management information, including:

[0106] For each target skill item, upon obtaining enterprise management information, the weight corresponding to the target skill item is determined according to the following formula, and the weight in the enterprise management information is updated based on the determined weight:

[0107] W j =argmax P(W j |D)

[0108] Among them, W j Let P(W) represent the weight corresponding to the j-th target skill item, D represent enterprise management information, and P(W) represent the weight corresponding to the j-th target skill item. j |D) The weight of the representation, given the known enterprise management information, is W. j The posterior probability.

[0109] In this embodiment, by introducing posterior probability calculation, the weight of each target skill item is dynamically determined, achieving more flexible and accurate modeling of job skill requirements. Specifically, the weight of each target skill item is updated based on enterprise management information (such as enterprise strategic planning, job responsibilities, etc.), effectively reflecting the actual importance the enterprise attaches to each skill. The weights of skill items can be adjusted in real time according to the actual needs of the enterprise, avoiding the biases that may arise from fixed or manually set weights in traditional recruitment methods. Furthermore, the posterior probability calculation is based on the latest data from enterprise management information, making the updates of skill weights more aligned with the enterprise's current strategic goals and development direction. Through dynamic optimization of the target skill item weights, the flexibility and accuracy in the job matching process are significantly improved, helping enterprises accurately identify and attract qualified candidates based on changing needs.

[0110] In one embodiment of this application, determining the comprehensive sentiment vector of the current candidate based on video stream information, audio stream information, and structured text includes:

[0111] In chronological order, the intensity value of at least one facial action unit of the current candidate is extracted from the video stream information, wherein the intensity value of the facial action unit represents the expression state of the corresponding facial organ of the current candidate.

[0112] Based on the temporal variation of the intensity values ​​of facial action units, a micro-expression fluctuation feature vector is generated;

[0113] Extract Mel-frequency cepstral coefficients from audio stream information;

[0114] Based on Mel frequency cepstral coefficients, the current candidate's voice emotion state category is identified;

[0115] Based on the voice emotion state category, determine the voice emotion feature vector used to represent the current candidate's emotional state;

[0116] Semantic analysis is performed on structured text to identify the current candidate's sentiment polarity value and corresponding confidence level, where the sentiment polarity value represents the current candidate's emotional tendency.

[0117] Based on at least one sentiment keyword in the pre-set enterprise management information, the structured text is analyzed and the corresponding target sentiment keyword in the structured text is identified. The sentiment keyword is a word that can reflect the personal feelings and attitudes of the current candidate.

[0118] Determine the weights corresponding to target sentiment keywords from enterprise management information;

[0119] The emotional polarity value, confidence level, and weights corresponding to each target emotional keyword are normalized to construct a text emotional feature vector;

[0120] Based on micro-expression fluctuation feature vectors, voice emotion feature vectors, and text emotion feature vectors, the comprehensive emotion vector of the current candidate is determined.

[0121] In this embodiment, by fusing video stream information, audio stream information, and structured text information, a comprehensive emotional vector of the candidate is accurately constructed, enabling a comprehensive and dynamic assessment of the candidate's emotional state. Specifically, the extraction and temporal variation analysis of facial action unit intensity values ​​from the video stream information helps capture the candidate's micro-expression fluctuations, reflecting their potential emotional changes. The extraction of Mel-frequency cepstral coefficients and the classification of speech emotion states from the audio stream information provide speech-level emotion analysis. Semantic analysis and emotion polarity value recognition of the structured text enable the system to understand the candidate's expressed emotional tendencies and confidence levels. Through the comprehensive analysis of these different emotional dimensions, a more comprehensive and accurate emotional feature vector can be constructed, avoiding the one-sidedness of analysis from a single data source and improving the accuracy of emotion recognition. This method not only improves the accuracy of capturing the candidate's emotional state but also provides a more scientific and comprehensive basis for subsequent candidate matching and evaluation.

[0122] It should be noted that the enterprise management information includes the enterprise strategy document for the enterprise development plan within the target time period, the job responsibilities, skill requirements and emotional requirements of the positions to be recruited, and historical recruitment data within the preset time period.

[0123] In one embodiment of this application, a micro-expression fluctuation feature vector is generated based on the temporal variation of the intensity values ​​of facial action units, including:

[0124]

[0125] Where Fv represents the micro-expression fluctuation feature vector, n is the total number of facial action units, and AU i W represents the intensity value of the i-th facial action unit. i The weight of the i-th facial action unit in emotion judgment is represented by t, and the interview duration of the current candidate is represented by t.

[0126] In this embodiment, by generating micro-expression fluctuation feature vectors based on the temporal changes in the intensity values ​​of facial action units, the emotional fluctuations of candidates during the interview process can be captured more accurately. Specifically, the micro-expression fluctuation feature vector is generated by weighting the intensity value of each facial action unit with its emotional weight, and then normalizing it by combining it with the interview duration. This ensures that emotional changes over different time periods are comprehensively reflected. This method effectively combines the candidate's facial expression dynamics with their emotional state, highlighting the influence of important action units through weighted averaging, avoiding over-interpretation of unimportant expressions, and thus improving the accuracy of emotional judgment. By generating micro-expression fluctuation feature vectors, the potential emotional responses of candidates can be better explored, enhancing the accuracy and reliability of the emotion recognition model and providing a more detailed and scientific basis for recruitment decisions.

[0127] In one embodiment of this application, a voice emotion feature vector is determined based on the voice emotion state category to represent the current candidate's emotional state, including:

[0128]

[0129] Where Fa represents the speech emotion feature vector, K is the total number of speech emotion categories, and w k The weight P(emotion) represents the weight of the k-th type of voice emotion state in the preset enterprise management information. k ) represents the predicted probability of the k-th speech emotion state category.

[0130] In this embodiment, by generating a voice emotion feature vector based on voice emotion state categories, the emotional state conveyed by a candidate through speech can be comprehensively and accurately assessed. Specifically, the voice emotion feature vector comprehensively considers the impact of different emotion categories on the overall emotional state by weighted averaging of the predicted probabilities and corresponding weights of various voice emotion state categories, thereby ensuring high accuracy and scientific rigor in emotion assessment. When analyzing voice emotion, not only the predicted emotion categories are considered, but also weighting factors from enterprise management information are introduced, making the emotion analysis more targeted and personalized, and allowing for dynamic adjustment of the degree of influence of voice emotion according to the actual needs of the enterprise. This approach improves the flexibility of voice emotion analysis, enhances the expressive power of the emotion feature vector, helps to more accurately identify the candidate's potential emotional state, and thus optimizes the recruitment decision-making process.

[0131] In one embodiment of this application, the comprehensive sentiment vector of the current candidate is determined based on micro-expression fluctuation feature vectors, speech sentiment feature vectors, and text sentiment feature vectors, including:

[0132] The overall sentiment vector of the current candidate is determined based on the following formula:

[0133] F fusion =α·Fv+β·Fa+γ·Ft

[0134] Among them, F fusion The comprehensive emotion vector representing the current candidate is defined by Fv, the micro-expression fluctuation feature vector, Fa, the speech emotion feature vector, Ft, the text emotion feature vector, α, β, and γ, where α+β+γ = 1 and α, β, and γ are all greater than 0.

[0135] In this embodiment, a comprehensive emotion vector for the current candidate is generated by weighted fusion of micro-expression fluctuation feature vectors, speech emotion feature vectors, and text emotion feature vectors, thus achieving a unified expression of multimodal emotion information. This fusion method uses weight parameters α, β, and γ to adjust the influence ratio of the three types of emotion features on the comprehensive emotion result, ensuring that the sum of the weights is 1, thereby achieving a balance between different emotion dimensions. By comprehensively analyzing the candidate's facial expression changes, speech emotion features, and text semantic emotion during the interview process, not only can the comprehensiveness and accuracy of emotion recognition be improved, but the information bias that may be caused by single-modal features can also be effectively overcome.

[0136] In one embodiment of this application, determining the job skill vector of the current candidate based on structured text includes:

[0137] Identify at least one skill item of the current candidate from the candidate's resume information and structured text, and assess the level of mastery of each skill item;

[0138] Normalize the weights corresponding to each skill item in the job skill requirement vector to construct the job skill vector.

[0139] In this embodiment, by extracting the candidate's skill items from structured text and evaluating the mastery of each skill item in conjunction with resume information, a job skill vector is constructed, which accurately reflects the candidate's skill matching degree for the position. By normalizing the weights of each skill item and its corresponding job skill requirement vector, the contribution of each skill to the overall job skill vector is ensured to be reasonable and balanced, effectively avoiding excessive or insufficient emphasis on any particular skill item. This method can comprehensively and systematically evaluate the candidate's mastery of various skills and compare them with job requirements, helping recruiters accurately identify the most suitable candidate for the position and improve job matching.

[0140] In one embodiment of this application, by combining the job emotional needs vector with the current candidate's comprehensive emotional vector, and employing a matching degree calculation method that incorporates a negative ideal job emotional vector, the emotional suitability of the current candidate and the job can be assessed more comprehensively and accurately. By analyzing the target emotional characteristics and their intensity of low-performing employees corresponding to the job to be recruited, and performing normalization processing to construct a negative ideal job emotional vector, the emotional patterns leading to poor performance in that job can be effectively characterized. By comparing the current candidate's comprehensive emotional vector with both the job's ideal emotional needs vector and the negative ideal job emotional vector, it is possible not only to measure the candidate's closeness to the ideal emotional needs but also to assess their distance from the negative ideal emotional pattern, thereby improving the discriminativeness and reliability of the matching degree assessment based on a two-way measurement. This approach enables companies to more accurately identify candidates whose emotional traits are more aligned with the job's success characteristics, significantly reducing the negative impact of emotional mismatch on job performance.

[0141] In this embodiment of the application, the skill matching degree is determined based on the job skill requirement vector and the job skill vector, including:

[0142] The skill matching degree is determined based on the following formula:

[0143]

[0144] Among them, Sim skill Characterizes skill matching degree, S candi，i S represents the job skill vector. job The vector represents the skill requirement for a job position; m represents the total number of skills, and i represents the i-th skill.

[0145] In one embodiment of this application, by calculating the inner product and Euclidean distance between the candidate's job skill vector and the job skill requirement vector, the degree of matching between the candidate's skills and job requirements can be comprehensively assessed. Compared to traditional single-skill matching methods, this assessment method is more accurate and scientific, comprehensively considering the relative importance and matching degree of various skills. It provides recruiters with a more efficient way to identify and screen high-quality candidates with skills that meet job requirements, thereby improving the accuracy of recruitment decisions, reducing the risk of substandard work performance due to skill mismatch, and improving recruitment efficiency and quality.

[0146] In this embodiment of the application, the matching degree between the current candidate and the job position is determined based on the emotional matching degree and the skill matching degree, including:

[0147] The matching degree between the current candidate and the position to be recruited is determined based on the following formula:

[0148] Score = λ·Sim skill+(1-λ)·Sim emotion

[0149] Wherein, the score represents the degree of match between the current candidate and the position to be recruited, and the Sim... skill Sim represents skill matching degree. emotion λ represents the emotional fit, and λ represents the weight of the pre-defined skill fit in the job opening.

[0150] In one embodiment of this application, by comprehensively considering both skills and emotional factors, this application can provide a more accurate matching assessment, helping companies to fully understand the degree to which the current candidate's abilities fit the job requirements. Compared with traditional single-skill assessment methods, this method not only considers whether the current candidate possesses the required professional skills, but also delves into the emotional fit between the current candidate and the position, thereby effectively avoiding problems such as low work efficiency or employee turnover caused by emotional mismatch. The introduction of this method not only improves the accuracy of recruitment decisions, but also significantly enhances employee job satisfaction and job adaptability, providing companies with a more scientific and comprehensive basis for recruitment.

[0151] like Figure 2 As shown, in order to more clearly illustrate the technical solution and advantages of this application, the following is a detailed description of a recruitment method provided by an embodiment of this application, which may include the following steps:

[0152] Step 201: From the enterprise management information, determine at least one target sentiment trait corresponding to the job position to be recruited and the sentiment intensity corresponding to each target sentiment trait.

[0153] Specifically, by utilizing a Dynamic Position Requirement Matching System (DPRM), companies can pre-set their management information and historical recruitment data for a predetermined time period. By extracting the target sentiment characteristics and sentiment intensity of the positions to be recruited from the company's management information, the matching assessment process can fully integrate the company's strategic planning with the sentiment requirements of the positions. This fundamentally ensures that the input data of the matching model is consistent with the company's development direction, thereby improving the relevance and reliability of the assessment results in actual recruitment.

[0154] The enterprise management information includes enterprise strategy documents outlining the enterprise's development plan for the target time period, as well as the job responsibilities, skills, and emotional requirements for the positions to be filled (e.g., a job-emotional characteristic association database. For example, a customer service position requires "high empathy" → AU12 intensity ≥ 0.7).

[0155] Step 202: Normalize the emotional features and corresponding emotional intensities of each target to construct the emotional requirement vector for the job to be recruited.

[0156] Specifically, by using a real-time job requirement matching system, the emotional characteristics and intensity are normalized to construct a job emotional requirement vector. This can eliminate the influence of different dimensions or numerical ranges, making subsequent matching calculations more fair and comparable, and improving the accuracy and stability of emotional matching calculations.

[0157] Step 203: Extract at least one core skill keyword and the corresponding target mastery level for the job to be recruited from the enterprise management information, and use the core skill keyword as the target skill item.

[0158] Specifically, by using a real-time job requirement matching system, core skill keywords and target mastery levels can be extracted and used as target skill items. This helps to accurately depict the actual skill requirements of a job, avoid misjudgments caused by vague skill descriptions, and thus improve the relevance of skill matching.

[0159] Step 204: Determine the weight of each target skill item based on the enterprise management information.

[0160] Specifically, utilizing a real-time job requirement matching system, for each target skill item, upon obtaining enterprise management information, the weight corresponding to the target skill item is determined according to the following formula, and the weight in the enterprise management information is updated based on the determined weight. By calculating the weight of the target skill item through posterior probability and dynamically updating the weight parameters in the enterprise management information, the importance of skill items can be automatically adjusted based on the enterprise's historical recruitment data and the latest strategy, achieving adaptive optimization of matching indicators and thus more accurately reflecting current recruitment needs.

[0161] W j =argmax P(W j |D)

[0162] Among them, W j Let P(W) represent the weight corresponding to the j-th target skill item, D represent enterprise management information, and P(W) represent the weight corresponding to the j-th target skill item. j |D) The weight of the representation, given the known enterprise management information, is W. j The posterior probability.

[0163] Step 205: Normalize the mastery level of each objective and its corresponding weight to construct a job skill requirement vector for the position to be recruited.

[0164] Specifically, by using a real-time job requirement matching system, the mastery level and weight of the target are normalized and a job skill requirement vector is constructed. This helps to eliminate calculation biases caused by uneven distribution of skill values ​​or different units, and ensures the scientificity and fairness of skill matching calculation.

[0165] Step 206: Collect video stream information, audio stream information, and structured text generated based on audio stream information of the current candidate for the position to be recruited during the interview process.

[0166] Specifically, a high-definition camera (1080P, 30fps) captures the candidate's facial micro-expressions in real time during the interview, and H.264 encoding is used for compression and transmission to obtain the candidate's video stream information. This compression method has a high compression ratio and excellent video quality, avoiding any impact on the quality of the transmitted video stream information. An audio stream is simultaneously recorded using a built-in noise-canceling microphone array (48kHz sampling rate). Based on ASR technology (using the Kaldi framework), the speech stream information is transcribed into structured text in real time. By simultaneously acquiring visual, auditory, and linguistic information sources from the video stream, audio stream, and transcript, complete input is provided for subsequent multimodal feature extraction, ensuring the comprehensiveness of the overall evaluation.

[0167] Step 207: Extract the intensity value of at least one facial action unit of the current candidate from the video stream information in chronological order.

[0168] Specifically, the intensity values ​​of facial motion units characterize the expression state of the corresponding facial organs of the current candidate. By using the improved OpenFace toolkit to extract the intensity values ​​of facial motion units from video stream information in temporal order, the dynamic changes of the current candidate's facial expressions can be finely depicted, capturing instantaneous emotional fluctuations, which helps to improve the sensitivity and accuracy of emotion recognition.

[0169] Step 208: Generate micro-expression fluctuation feature vectors based on the temporal changes in the intensity values ​​of facial action units.

[0170] Specifically, by analyzing the temporal changes of AU through an LSTM network, and based on the following formula, a micro-expression fluctuation feature vector is generated based on the temporal changes of facial action unit intensity values. This can quantify subtle changes in facial expressions, reflect the candidate's true emotional response, and provide more objective and refined visual emotional features for emotion matching calculation.

[0171]

[0172] Where Fv represents the micro-expression fluctuation feature vector, n is the total number of facial action units, and AU i W represents the intensity value of the i-th facial action unit. i The weight of the i-th facial action unit in emotion judgment is represented by t, and the interview duration of the current candidate is represented by t.

[0173] Step 209: Extract Mel frequency cepstral coefficients from the audio stream information.

[0174] Specifically, by extracting Mel-frequency cepstral coefficients from audio stream information using the Librosa library, the speech characteristics and timbre features of candidates can be effectively characterized, laying a solid acoustic feature foundation for speech emotion recognition.

[0175] Step 210: Identify the current candidate's speech emotion state category based on Mel frequency cepstral coefficients.

[0176] Specifically, by using an SVM classifier to identify the candidate's emotional state category based on Mel frequency cepstral coefficients, the emotional expression characteristics of the candidate in the interview can be accurately determined, capturing their psychological state and attitude from auditory signals, and adding a reliable source of emotional information for matching assessment.

[0177] Step 211: Based on the voice emotion state category, determine the voice emotion feature vector used to represent the current candidate's emotional state.

[0178] Specifically, by using an SVM classifier, the voice emotion feature vector is calculated based on the voice emotion state category according to the following formula. Combined with the enterprise emotion weight, the voice emotion of the current candidate can be quantified into a numerical form that can be directly used in the matching calculation, which facilitates accurate comparison with the emotional requirements of the job.

[0179]

[0180] Where Fa represents the speech emotion feature vector, K is the total number of speech emotion categories, and w k The weight P(emotion) represents the weight corresponding to the k-th type of voice emotion state in enterprise management information. k ) represents the predicted probability of the k-th speech emotion state category.

[0181] Step 212: Perform semantic analysis on the structured text to identify the current candidate's sentiment polarity and corresponding confidence level.

[0182] Specifically, the sentiment polarity value represents the current candidate's emotional inclination. Based on the BERT-Base model, by performing semantic analysis on structured text, the sentiment polarity value and confidence level of the current candidate can be identified, and their emotional inclination can be extracted from the language content, making up for the emotional information that may be missed by purely visual or auditory data.

[0183] Step 213: Based on at least one sentiment keyword in the enterprise management information, analyze the structured text and identify the corresponding target sentiment keyword in the structured text.

[0184] Specifically, sentiment keywords are words that reflect the candidate's personal feelings and attitudes. Analyzing sentiment keywords in structured text based on enterprise management information analysis can deeply identify the implicit emotions and attitudes in language, enabling fine-grained mining of candidates' verbal emotions and improving the accuracy of text sentiment analysis.

[0185] Step 214: Determine the weights corresponding to the target sentiment keywords from the enterprise management information.

[0186] Specifically, determining the weights of target sentiment keywords from enterprise management information can differentiate the impact of different sentiment expressions on the matching degree calculation, thus better aligning with the actual focus of enterprises on sentiment characteristics.

[0187] Step 215: Normalize the sentiment polarity value, confidence level, and weights corresponding to each target sentiment keyword to construct a text sentiment feature vector.

[0188] Specifically, normalizing the polarity, confidence, and keyword weights in the text sentiment features and constructing a text sentiment feature vector ensures that the feature is comparable to other sentiment features and improves the computational accuracy during multimodal fusion.

[0189] Step 216: Determine the comprehensive sentiment vector of the current candidate based on the micro-expression fluctuation feature vector, the voice sentiment feature vector, and the text sentiment feature vector.

[0190] Specifically, based on the following formula, and by weighting and fusing micro-expression fluctuation feature vectors, voice emotion feature vectors, and text emotion feature vectors, a comprehensive emotion vector is generated. This enables the complementarity of multimodal emotion information, making the emotional characteristics of candidates more comprehensive and more consistent with their true psychological state, thereby improving the credibility of emotion matching calculation.

[0191] F fusion =α·Fv+β·Fa+γ·Ft

[0192] Among them, F fusion The comprehensive emotion vector representing the current candidate is defined by Fv, the micro-expression fluctuation feature vector, Fa, the speech emotion feature vector, Ft, the text emotion feature vector, α, β, and γ, where α+β+γ = 1 and α, β, and γ are all greater than 0.

[0193] Step 217: Identify at least one skill item of the current candidate from the candidate's resume information and structured text, and assess the level of mastery of each skill item.

[0194] Specifically, identifying and assessing a candidate's skills from resume information and structured texts can accurately reveal their actual skill set and level, avoiding misjudgments of skills due to missing information or biased interview performance.

[0195] Step 218: Normalize the weights corresponding to each skill item in the job skill requirement vector to construct the job skill vector.

[0196] Specifically, normalizing the current candidate's skill items and corresponding weights to construct a job skill vector ensures that the quantitative results of the current candidate's skill data are consistent with the job skill requirements in terms of dimensionality, thus guaranteeing the accuracy of skill matching calculation.

[0197] Step 219: Obtain at least one target sentiment trait and corresponding sentiment intensity for at least one low-performing employee corresponding to the position to be recruited.

[0198] Specifically, obtaining the emotional characteristics and intensity of low-performing employees can provide a "negative sample" reference for matching degree calculation, which can be used to identify emotional patterns that are not suitable for the job, thereby improving the discriminative ability of emotional matching degree assessment.

[0199] Step 220: For each low-performing employee, normalize the target emotional characteristics and corresponding emotional intensities of each low-performing employee, construct a negative ideal job emotional vector, and calculate the mean of each negative ideal job emotional vector.

[0200] Specifically, constructing a negative ideal job emotion vector and calculating its mean can provide a numerical benchmark for inappropriate emotions for the emotion matching formula, further improving the matching calculation's ability to identify potential low-performing candidates.

[0201] Step 221: Determine the emotional matching degree based on the mean of the negative ideal job emotional vector, the comprehensive emotional vector, and the job emotional demand vector.

[0202] Specifically, based on the following formula, the emotional matching degree is calculated using the comprehensive emotional vector, the job emotional demand vector, and the negative ideal job emotional vector. This allows for a precise quantitative comparison of the current candidate's emotional characteristics, effectively avoiding job adaptation problems caused by emotional mismatch.

[0203]

[0204] Among them, Sim emotion F represents the degree of emotional fit. fusion Representing the comprehensive sentiment vector, Characterizing the emotional needs vector of a job position The mean of the negative ideal job emotion vector is represented by i, where i represents the i-th emotion feature dimension and n represents the number of emotion feature dimensions.

[0205] Step 222: Determine the skill matching degree based on the job skill requirement vector and the job skill vector.

[0206] Specifically, based on the following formula, the skill matching degree is calculated by comparing the job skill requirement vector with the current candidate's skill vector. This can scientifically measure the fit between the current candidate's skill level and the job requirements, and improve the objectivity of skill assessment.

[0207]

[0208] Among them, Sim skill Characterizes skill matching degree, S candi，i S represents the job skill vector. job The vector represents the skill requirements for a job position. m represents the total number of skills, and i represents the i-th skill.

[0209] Step 223: Determine the match between the current candidate and the position to be recruited based on the emotional match and skill match.

[0210] Specifically, based on the following formula, combining skill matching and emotional matching, and calculating the comprehensive matching degree according to weights, it can simultaneously take into account the ability fit and emotional fit between the current candidate and the position, and achieve a comprehensive and accurate assessment of the overall suitability of the current candidate, which helps to improve the recruitment success rate and the subsequent performance of employees.

[0211] Score = λ·Sim skill +(1-λ)·Sim emotion

[0212] Wherein, the score represents the degree of match between the current candidate and the position to be recruited, and the Sim... skill Sim represents skill matching degree. emotion λ represents the emotional fit, and λ represents the weight of the pre-defined skill fit in the job opening.

[0213] like Figure 3 As shown, this application provides a recruitment device, including:

[0214] The dynamic modeling module 301 is used to determine the job emotional needs vector and job skill needs vector for the positions to be recruited.

[0215] The data acquisition module 302 is used to collect video stream information, audio stream information, and structured text generated based on the audio stream information of the current candidate for the position to be recruited during the interview process.

[0216] The feature extraction module 303 is used to determine the comprehensive sentiment vector of the current candidate based on the video stream information, audio stream information, and structured text collected by the data acquisition module 302. Based on the structured text, the module determines the job skill vector of the current candidate.

[0217] The matching decision module 304 determines the emotional matching degree based on the job emotional requirement vector determined by the dynamic modeling module 301 and the comprehensive emotional vector determined by the feature extraction module 303. It also determines the skill matching degree based on the job skill requirement vector and the job skill vector. Finally, based on the emotional matching degree and the skill matching degree, it determines the matching degree between the current candidate and the job to be recruited.

[0218] In this application embodiment, by comprehensively considering the multimodal emotional signals of candidates during the interview process (such as video streams, audio streams, and structured text), as well as the emotional and skill requirements of the position, a more comprehensive and accurate job matching can be achieved. Compared with traditional resume keyword matching or single sentiment analysis methods, this invention can capture and analyze the candidate's emotional state in real time, identify key emotional characteristics such as emotional stability and stress resistance, and effectively improve the accuracy of emotional fit assessment. Simultaneously, by combining voice and video information from the interview, the candidate's emotional tendencies can be assessed from multiple perspectives, avoiding the one-sidedness of emotional assessment. By dynamically assessing the matching degree between the emotional requirements of the position and the candidate's emotional vector, the recruitment process is ensured to be more flexible and personalized, providing companies with candidate selections that better meet the job requirements. Furthermore, the skill matching assessment based on structured text analysis further enhances the accurate matching of skill requirements and the candidate's professional abilities. This application, through a comprehensive assessment of both emotional and skill dimensions, greatly improves the intelligence level of the recruitment method, making recruitment decisions more scientific and accurate, thereby effectively improving the quality and efficiency of talent selection for enterprises.

[0219] This application provides an electronic device, but the specific implementation of the electronic device is not limited by the specific embodiments of this application.

[0220] This electronic device may include: a processor, a communications interface, memory, and a communication bus. Among these:

[0221] The processor, communication interface, and memory communicate with each other via a communication bus.

[0222] A communication interface is used to communicate with other electronic devices or servers.

[0223] The processor is used to execute programs, specifically the relevant steps in the above method embodiments.

[0224] Specifically, the program may include program code, which includes computer operation instructions.

[0225] The processor may be a CPU, a GPU (Graphics Processing Unit), an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0226] Memory is used to store programs. Memory may include high-speed RAM, and may also include non-volatile memory, such as at least one disk drive.

[0227] The program may include multiple computer instructions, and the program may specifically cause the processor to execute the operation corresponding to any of the methods described in the foregoing multiple method embodiments through multiple computer instructions.

[0228] The specific implementation of each step in the program can be found in the corresponding steps and units described in the above method embodiments, and has corresponding beneficial effects, which will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0229] This application also provides a computer storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in any of the foregoing method embodiments. The computer storage medium includes, but is not limited to, compact disc read-only memory (CD-ROM), random access memory (RAM), floppy disk, hard disk, or magneto-optical disk.

[0230] This application also provides a computer program product, including computer instructions that instruct a computing device to perform an operation corresponding to any of the methods in the above-described multiple method embodiments.

[0231] Furthermore, it should be noted that the user-related information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to sample data used for training the model, data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0232] It should be noted that, depending on the implementation needs, the various components / steps described in the embodiments of this application can be broken down into more components / steps, or two or more components / steps or parts of the operation of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of this application.

[0233] The methods described in the embodiments of this application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code downloaded over a network that is originally stored in a remote recording medium or a non-transitory machine-readable medium and will be stored in a local recording medium. Thus, the methods described herein can be stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA)). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., Random Access Memory (RAM), Read-Only Memory (ROM), Flash Memory, etc.) capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the methods described herein are implemented. Furthermore, when a general-purpose computer accesses the code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the methods shown herein.

[0234] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for specific applications, but such implementations should not be considered beyond the scope of the embodiments of this application.

[0235] The above embodiments are only used to illustrate the embodiments of this application, and are not intended to limit the embodiments of this application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of this application. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of this application, and the patent protection scope of the embodiments of this application should be defined by the claims.

Claims

1. A recruitment method, characterized in that, include: Determine the emotional needs vector and skill requirements vector for the positions to be recruited; Collect video stream information, audio stream information, and structured text generated based on the audio stream information of the current candidate during the interview process for the position to be recruited; Based on the video stream information, the audio stream information, and the structured text, determine the comprehensive sentiment vector of the current candidate; Based on the structured text, determine the job skill vector of the current candidate; Based on the job-related emotional needs vector and the comprehensive emotional vector, the emotional matching degree is determined; Based on the job skill requirement vector and the job skill vector, the skill matching degree is determined; The matching degree between the current candidate and the job opening is determined based on the emotional matching degree and the skill matching degree.

2. The method according to claim 1, characterized in that, The determination of the job affective requirement vector and job skill requirement vector for the position to be recruited includes: From the preset enterprise management information, at least one target emotional feature corresponding to the job to be recruited and the emotional intensity corresponding to each target emotional feature are determined. The enterprise management information includes the enterprise strategic document of the enterprise development plan within the target time period, the job responsibilities, skill requirements and emotional requirements of the job to be recruited, and historical recruitment data within the preset time period. Normalize each of the target emotional features and the corresponding emotional intensity to construct the job emotional requirement vector for the job to be recruited. From the enterprise management information, determine at least one target skill item and the corresponding level of mastery for the position to be recruited; Based on the enterprise management information, determine the weight corresponding to each of the target skill items; The mastery level of each objective and its corresponding weight are normalized to construct the job skill requirement vector for the position to be recruited.

3. The method according to claim 2, characterized in that, The step of determining at least one target skill item and the normalized level of mastery of the target skill item corresponding to the job position to be recruited from the enterprise management information includes: From the enterprise management information, extract at least one core skill keyword and the corresponding target mastery level for the job to be recruited, and use the core skill keyword as the target skill item.

4. The method according to claim 2, characterized in that, The step of determining the weight corresponding to each target skill item based on the enterprise management information includes: For each of the target skill items, when the enterprise management information is obtained, the weight corresponding to the target skill item is determined according to the following formula, and the weight in the enterprise management information is updated according to the determined weight: W j =argmax P(W j |D) Among them, W j Let P(W) represent the weight corresponding to the j-th target skill item, D represent the enterprise management information, and P(W) represent the weight corresponding to the j-th target skill item. j |D) The weight of the representation, given the enterprise management information, is W. j The posterior probability.

5. The method according to claim 1, characterized in that, Determining the comprehensive sentiment vector of the current candidate based on the video stream information, the audio stream information, and the structured text includes: In chronological order, the intensity value of at least one facial action unit of the current candidate is extracted from the video stream information, wherein the intensity value of the facial action unit represents the expression state of the corresponding facial organ of the current candidate. Based on the temporal changes in the intensity values ​​of the facial motion units, a micro-expression fluctuation feature vector is generated; Extract Mel-frequency cepstral coefficients from the audio stream information; Based on the Mel frequency cepstral coefficients, the voice emotion state category of the current candidate is identified; Based on the voice emotion state category, a voice emotion feature vector is determined to characterize the current candidate's emotional state; Semantic analysis is performed on the structured text to identify the current candidate's sentiment polarity value and corresponding confidence level, wherein the sentiment polarity value represents the current candidate's emotional tendency. Based on at least one sentiment keyword in the preset enterprise management information, the structured text is analyzed and the corresponding target sentiment keyword in the structured text is identified, wherein the sentiment keyword is a word that can reflect the personal feelings and attitudes of the current candidate; Determine the weights corresponding to the target sentiment keywords from the enterprise management information; The emotional polarity value, the confidence level, and the weights corresponding to each of the target emotional keywords are normalized to construct a text emotional feature vector. Based on the micro-expression fluctuation feature vector, the voice emotion feature vector, and the text emotion feature vector, the comprehensive emotion vector of the current candidate is determined.

6. The method according to claim 5, characterized in that, The generation of micro-expression fluctuation feature vectors based on the temporal changes in the intensity values ​​of the facial motion units includes: Where Fv represents the micro-expression fluctuation feature vector, n is the total number of facial action units, and AU i W represents the intensity value of the i-th facial action unit. i The weight of the i-th facial action unit in emotion judgment is represented by t, and the interview duration of the current candidate is represented by t.

7. The method according to claim 5, characterized in that, The step of determining the voice emotion feature vector to represent the current candidate's emotional state based on the voice emotion state category includes: Where Fa represents the speech emotion feature vector, K is the total number of speech emotion categories, and w k The weight P(emotion) represents the weight of the k-th type of voice emotion state in the preset enterprise management information. k ) represents the predicted probability of the k-th speech emotion state category.

8. The method according to claim 5, characterized in that, The determination of the comprehensive sentiment vector of the current candidate based on the micro-expression fluctuation feature vector, the voice sentiment feature vector, and the text sentiment feature vector includes: The comprehensive sentiment vector of the current candidate is determined based on the following formula: F fusion =α·Fv+β·Fa+γ·Ft Among them, F fusion The comprehensive emotion vector of the current candidate is represented by Fv, the micro-expression fluctuation feature vector is represented by Fa, the voice emotion feature vector is represented by Ft, the text emotion feature vector is represented by α, the weight of the micro-expression fluctuation feature vector is represented by β, the weight of the voice emotion feature vector is represented by γ, and α+β+γ=1, and α, β and γ are all greater than 0.

9. The method according to any one of claims 1 to 8, characterized in that, Determining the job skill vector of the current candidate based on the structured text includes: Identify at least one skill item of the current candidate from the current candidate's resume information and the structured text, and assess the mastery level of each skill item; Normalize the weights corresponding to each skill item in the job skill requirement vector to construct a job skill vector. And / or, The process of determining the emotional matching degree based on the job emotional needs vector and the comprehensive emotional vector includes: Obtain at least one target sentiment feature and corresponding sentiment intensity for at least one low-performing employee corresponding to the job opening; For each low-performing employee, the target emotional features and corresponding emotional intensities of the employee are normalized to construct a negative ideal job emotional vector. Calculate the mean of the negative ideal job emotion vector for each of the aforementioned positions; The emotional matching degree is determined based on the following formula and the mean of the negative ideal job emotional vector: Among them, Sim emotion F represents the degree of emotional matching. fusion Characterizing the comprehensive sentiment vector, The vector representing the emotional needs of the job position. The mean of the negative ideal job emotion vector is represented by i, where i represents the i-th emotion feature dimension and n represents the number of emotion feature dimensions. And / or, The process of determining the skill matching degree based on the job skill requirement vector and the job skill vector includes: The skill matching degree is determined based on the following formula: Among them, Sim skill S represents the skill matching degree. candi，i S represents the job skill vector. job The vector representing the skill requirements for the job position; m represents the total number of skills, and i represents the i-th skill; And / or, Determining the match between the current candidate and the job opening based on the emotional compatibility and the skill compatibility includes: The matching degree between the current candidate and the job opening is determined based on the following formula: Score=λ·Sim skill +(1-λ)·Sim emotion Wherein, Score represents the degree of match between the current candidate and the job opening, Sim skill Sim represents the skill matching degree. emotion The emotional matching degree is represented by λ, and the preset skill matching degree is represented by its weight in the job position to be recruited.

10. A recruitment device, characterized in that, include: The dynamic modeling module is used to determine the job affective requirement vector and job skill requirement vector for the position to be recruited. The data acquisition module is used to collect video stream information, audio stream information, and structured text generated based on the audio stream information of the current candidate for the position to be recruited during the interview process. The feature extraction module is used to determine the comprehensive sentiment vector of the current candidate based on the video stream information, the audio stream information, and the structured text collected by the data acquisition module. Based on the structured text, determine the job skill vector of the current candidate; The matching decision module is used to determine the emotional matching degree based on the job emotional demand vector determined by the dynamic modeling module and the comprehensive emotional vector determined by the feature extraction module. Based on the job skill requirement vector and the job skill vector, the skill matching degree is determined; based on the emotional matching degree and the skill matching degree, the matching degree between the current candidate and the job to be recruited is determined.

Images