A person-job matching system based on artificial intelligence

Abstract

The present invention relates to the field of data processing technology, and in particular to an artificial intelligence-based person-job matching system, the system comprising: a collection module, a determination module, an evaluation module, a confirmation module, an adjustment module, a correction module, and a matching module. The present invention collects various key data, determines the positions of interest based on resume and recruitment keywords, combined with browsing frequency and browsing speed, uses a model to evaluate the recruitment index and resume index, and determines a candidate position list based on historical training frequency and historical promotion speed. Finally, the resume index or similarity threshold is dynamically adjusted according to the submission and interview success rate of the candidate position list, and the candidate position list is corrected according to the number of adjustments and the recommended positions are matched, so that the recommendation results are more in line with the target and position, thereby improving the accuracy and effectiveness of the matching and effectively solving the problems of inaccurate matching and slow response speed caused by incomplete data collection and over-reliance on static models.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an artificial intelligence-based person-job matching system. Background Art

[0002] In today's rapidly evolving labor market, companies face increasingly fierce competition and need to efficiently and accurately find the right talent to maintain their competitive advantage. At the same time, job seekers also expect to quickly find positions that match their skills and interests. Traditional job-person matching often relies on manual resume screening and interviews. This method is not only time-consuming and labor-intensive, but also susceptible to subjective factors, leading to inefficient matching and talent loss. Therefore, AI-based job-person matching systems have emerged, aiming to improve recruitment efficiency and quality through advanced technology, achieving precise matching between companies and job seekers.

[0003] Patent document with publication number CN116578933A discloses a talent resume matching method and system based on artificial intelligence big data. The method includes: S100, obtaining resume information and recruitment demand information, and collecting a first data set in the resume information and a second data set in the recruitment demand information book through a preset intelligent model; S200, performing big data collection based on the first data set and the second data set respectively, to obtain a first associated data set and a second associated data set; S300, inputting the first associated data set and the second associated data set into the training model respectively, to obtain a first label tree and a second label tree; and evaluating the matching degree between the resume information and the recruitment demand information based on the comparison of the first label tree and the second label tree.

[0004] It can be seen that the talent resume matching method based on artificial intelligence big data has the following problems: it only relies on static resume information and recruitment demand information, and cannot comprehensively evaluate the suitability of job seekers, resulting in inaccurate matching results; the static model causes the matching results to be unable to reflect market changes and changes in job seekers' behavior in a timely manner, reducing the system's response speed and matching accuracy; the matching results are too dependent on static information and cannot comprehensively evaluate the comprehensive ability and potential of job seekers, thereby affecting the accuracy and effectiveness of the matching. Summary of the Invention

[0005] To this end, the present invention provides an artificial intelligence-based job matching system, which is used to overcome the problems of inaccurate matching and slow response speed in the prior art caused by incomplete data collection and over-reliance on static models through multi-dimensional real-time data and dynamic adjustment mechanisms.

[0006] To achieve the above objectives, the present invention provides an artificial intelligence-based job matching system, comprising:

[0007] The collection module is used to collect the recruitment keywords, historical training frequency, historical promotion speed, target browsing frequency, browsing speed, delivery success rate and interview success rate of each matching position in the position database in real time after the target uploads a resume containing resume keywords.

[0008] a determination module, connected to the acquisition module, for determining a number of positions of interest based on the resume keywords, the recruitment keywords, a preset similarity threshold, the browsing frequency, and the browsing speed;

[0009] An evaluation module, connected to the acquisition module and the determination module, respectively, for evaluating a resume index based on the resume keywords based on a preset artificial intelligence model, and evaluating a recruitment index based on the recruitment keywords of each of the positions of interest;

[0010] a determination module, connected to the collection module and the evaluation module respectively, for determining a list of candidate positions based on the recruitment index, the resume index, the historical training frequency, and the historical promotion speed of each position of interest;

[0011] an adjustment module, connected to the collection module and the determination module respectively, for adjusting the resume index or the preset similarity threshold according to the candidate position list, the submission success rate, and the interview success rate;

[0012] a correction module connected to the adjustment module and the determination module, configured to correct the candidate job list according to the number of times the resume index is adjusted and the number of times the preset similarity threshold is adjusted within a preset correction time period, thereby forming a corrected job list;

[0013] A matching module is connected to the correction module and is used to match a plurality of recommended positions according to the corrected position list.

[0014] Furthermore, the determination module includes:

[0015] a vector conversion unit, configured to convert the recruitment keywords into vector form to obtain a plurality of recruitment key vectors, and to convert the resume keywords into vector form to obtain a resume key vector;

[0016] a temporary determination unit, connected to the vector conversion unit, for determining a number of temporary positions based on the resume key vector and each of the recruitment key vectors;

[0017] a browsing frequency determination unit, connected to the temporary determination unit, for determining a number of marked positions based on the browsing frequency of each temporary position;

[0018] The browsing speed determination unit is connected to the browsing frequency determination unit and is used to determine that the marked post is a focus post when the browsing speed of each marked post is less than a preset browsing speed threshold, so as to determine a number of focus posts.

[0019] Furthermore, the temporary determination unit includes:

[0020] A similarity calculation subunit, configured to calculate the cosine similarity between the resume key vector and each of the recruitment key vectors to obtain a plurality of similarities;

[0021] The temporary determination subunit is connected to the similarity calculation subunit and is used to determine that the position to be matched is a temporary position when the similarity is greater than a preset similarity threshold, so as to determine a number of temporary positions.

[0022] Furthermore, the browsing frequency determination unit includes:

[0023] A browsing frequency fluctuation calculation subunit, configured to calculate a standard deviation of the browsing frequency within a preset determination time period to obtain a browsing frequency fluctuation value;

[0024] a browsing frequency average calculation subunit, configured to calculate an average value of the browsing frequencies within the preset determination time period to obtain a browsing frequency average;

[0025] The browsing frequency determination subunit is connected to the browsing frequency fluctuation calculation subunit and the browsing frequency mean calculation subunit respectively, and is used to determine that the temporary position is a marked position when the browsing frequency fluctuation value is less than the preset browsing frequency fluctuation threshold and the browsing frequency mean is greater than the preset browsing frequency mean threshold, so as to determine a number of marked positions.

[0026] Furthermore, the determining module includes:

[0027] An index comparison unit, configured to compare the recruitment index and the resume index to obtain an index comparison result;

[0028] The candidate determination unit is connected to the index comparison unit and is used to determine the concerned position as a candidate position when the index comparison result shows that the resume index is greater than the recruitment index, so as to determine a number of candidate positions.

[0029] A ranking unit is connected to the candidate determination unit and is used to determine a candidate position list according to the historical training frequency and the historical promotion speed of each candidate position.

[0030] Furthermore, the sorting unit includes:

[0031] A training curve drawing subunit is used to draw a change curve of the historical training frequency within a preset sorting time to obtain a plurality of training curves;

[0032] A promotion curve drawing subunit is used to draw a change curve of the historical promotion speed within the preset sorting time to obtain a plurality of promotion curves;

[0033] a correlation calculation subunit, connected to the training curve drawing subunit and the promotion curve drawing subunit respectively, for calculating the correlation coefficients of the training curve and the promotion curve to obtain a plurality of correlation coefficient values;

[0034] The sorting subunit is connected to the correlation calculation subunit and is used to sort the correlation coefficient values ​​of all the candidate positions from high to low to obtain a candidate position list.

[0035] Furthermore, the adjustment module includes:

[0036] a delivery deviation calculation unit, configured to calculate a relative deviation between the delivery success rate and a preset expected delivery success rate to obtain a delivery deviation;

[0037] An interview deviation calculation unit, used to calculate the relative deviation between the interview success rate and a preset expected interview success rate to obtain the interview deviation;

[0038] a deviation index calculation unit, connected to the delivery deviation calculation unit and the interview deviation calculation unit, respectively, for performing a weighted summation of the delivery deviation, the preset delivery weight, the interview deviation, and the preset interview weight to obtain a deviation index;

[0039] a deviation comparison unit connected to the deviation index calculation unit, for comparing the deviation index with a preset deviation index range to obtain a deviation comparison result;

[0040] An adjustment unit is connected to the deviation comparison unit and is used to adjust the resume index according to the deviation index when the deviation comparison result is that the deviation index is greater than the maximum value of the preset deviation index range, or to adjust the preset similarity threshold according to the delivery deviation when the deviation comparison result is that the deviation index is less than the minimum value of the preset deviation index range.

[0041] Furthermore, the adjustment unit includes:

[0042] The first adjustment subunit is used to adjust the resume index according to the relative deviation between the deviation index and the maximum value of the preset deviation index range and the preset first adjustment coefficient when the deviation index is greater than the maximum value of the preset deviation index range, so as to obtain an adjusted resume index, and the relative deviation is negatively correlated with the adjusted resume index.

[0043] The second adjustment subunit is used to adjust the preset similarity threshold according to the preset delivery deviation threshold and the relative deviation of the delivery deviation and the preset second adjustment coefficient when the deviation index is less than or equal to the minimum value of the preset deviation index range, so as to obtain the adjusted similarity threshold, and the relative deviation is positively correlated with the adjusted resume index.

[0044] Furthermore, the correction module includes:

[0045] A times acquisition unit is used to acquire the number of the adjusted resume index obtained within the preset correction time to form the index adjustment times, and to acquire the number of the adjusted similarity threshold obtained within the preset correction time to form the threshold adjustment times;

[0046] a number difference calculation unit, connected to the number acquisition unit, for calculating the absolute value of the difference between the index adjustment number and the threshold adjustment number to obtain a number difference value;

[0047] A correction unit is connected to the number difference calculation, and is used to correct the candidate position list according to the number difference value, the index adjustment number and the threshold adjustment number.

[0048] Furthermore, the correction unit includes:

[0049] A first correction subunit is configured to correct the candidate job list based on the adjusted resume index to form a corrected job list when the number difference value is greater than a preset number difference threshold and the number of index adjustments is greater than the threshold adjustment number, and to correct the candidate job list based on the adjusted similarity threshold to form a corrected job list when the number difference value is greater than the preset number difference threshold and the number of index adjustments is less than the threshold adjustment number;

[0050] an adjustment fluctuation calculation subunit, configured to calculate, when the number difference value is less than or equal to the preset number difference threshold, a standard deviation of the number of index adjustments within the preset correction time period to obtain an index adjustment fluctuation value, and to calculate a standard deviation of the number of threshold adjustments within the preset correction time period to obtain a threshold adjustment fluctuation value;

[0051] a fluctuation deviation calculation subunit, connected to the adjustment fluctuation calculation subunit, for calculating a relative deviation between the index adjustment fluctuation value and a preset index adjustment threshold value to obtain an index adjustment deviation, and for calculating a relative deviation between the threshold adjustment fluctuation value and a preset threshold adjustment threshold value to obtain a threshold adjustment deviation;

[0052] The second correction subunit is connected to the fluctuation deviation calculation subunit and is used to correct the candidate job list based on the adjusted resume index to form a corrected job list when the index adjustment deviation is greater than the threshold adjustment deviation, and to correct the candidate job list based on the adjusted similarity threshold to form a corrected job list when the index adjustment deviation is less than or equal to the threshold adjustment deviation.

[0053] Compared with the prior art, the beneficial effect of the present invention is that, by collecting the keywords of the target's uploaded resume and the relevant data of each position in the position database and the target's behavioral data, the positions of interest are determined based on the keywords of the resume and recruitment, combined with the browsing frequency and browsing speed, and the positions that meet the job requirements and reflect the candidate's real interests are screened out. Afterwards, the resume and the position of interest are evaluated using an artificial intelligence model to obtain a recruitment index and a resume index, providing a quantitative basis for job matching. The candidate position list is then determined based on the recruitment index, resume index, historical training frequency and historical promotion speed, and the attractiveness and development prospects of the position are comprehensively considered. The resume index or similarity threshold is dynamically adjusted according to the delivery and interview success rate of the candidate position list, and the candidate position list is re-determined according to the number of adjustments and the recommended positions are matched, so that the recommendation results are more in line with the target's job search situation and the actual matching degree of the position, thereby improving the accuracy and effectiveness of the matching, helping the target find a more suitable position, and effectively solving the problems of inaccurate matching and slow response speed caused by incomplete data collection and over-reliance on static models in the prior art.

[0054] Furthermore, through vector conversion, the originally abstract textual information can be quantified into concrete numerical vectors, providing an operational data foundation for subsequent matching calculations. The vector form can more intuitively reflect the similarity and correlation between keywords, facilitating precise matching using mathematical methods. Subsequently, temporary positions are identified based on the relationship between the recruitment key vector and the resume key vector, enabling more objective and accurate screening of positions with a high degree of match to the resume. Job seekers are screened for positions of genuine interest based on browsing frequency and browsing speed. Browsing frequency reflects the level of interest shown by job seekers. Browsing speed further reflects the duration of a job seeker's engagement with job information. Browsing speeds below a preset browsing speed threshold identify positions of interest to the job seeker, indicating that the job information is highly consistent with the job seeker's expectations. By first filtering out tagged positions based on browsing frequency and then further filtering out positions of interest based on browsing speed, job seekers can more accurately identify positions of genuine interest, avoiding the potential misjudgment caused by relying solely on single-dimensional data.

[0055] Furthermore, by calculating the cosine similarity between the resume key vector and each job posting key vector, a quantifiable basis for job matching is provided. Cosine similarity effectively measures the directional correlation between two vectors and is unaffected by vector length, thus more accurately reflecting the semantic association between keywords. Next, when the similarity exceeds a threshold, the matching position is determined to be temporary. This effectively selects positions with a high degree of match with the resume, avoids interference from less relevant positions, and improves the efficiency and accuracy of job screening.

[0056] Furthermore, by considering both the fluctuation value and the mean value, a more comprehensive assessment of a job seeker's genuine interest in a position can be achieved. The fluctuation value of browsing frequency reflects the stability of a job seeker's browsing behavior. A smaller fluctuation value indicates that the job seeker's attention to the position is relatively stable over a predetermined period of time, rather than occasional or random browsing. The mean browsing frequency, on the other hand, reflects the overall level of interest in the position. A higher mean value indicates a high level of interest. A position is marked only when the fluctuation value is less than a preset threshold and the mean value is greater than the preset threshold. This effectively identifies positions that are consistently and highly viewed by job seekers, avoiding misjudgments due to accidental factors.

[0057] Furthermore, identifying candidate positions using the Recruitment Index and Resume Index provides an intuitive measure of the match between job seekers and positions. The Recruitment Index reflects the job's requirements and expectations for the candidate, while the Resume Index reflects the candidate's abilities and qualifications. When the Resume Index is greater than the Recruitment Index, it indicates that the candidate's abilities and qualifications exceed the job requirements to a certain extent. Next, candidate positions are ranked using historical training frequency and historical promotion rates, reflecting the company's employee training and development opportunities. Higher training frequency and promotion rates indicate a company's greater focus on employee growth and development, and job seekers in such positions may have greater room for development and better career prospects. The final list of candidate positions not only meets the candidate's abilities and needs, but also provides them with better career development opportunities.

[0058] Furthermore, by plotting the changes in training frequency and promotion rate and calculating their correlation coefficient, we can quantitatively assess the relationship between training and development opportunities within a position. A higher correlation coefficient indicates a strong positive correlation between training opportunities and promotion rate. That is, the more training a company provides, the faster its employees are promoted, reflecting the company's commitment to providing better training and development opportunities for its employees. By ranking candidate positions based on the correlation coefficient, we can prioritize those that prioritize employee growth and development. This approach not only considers the current fit of a position but also the career development needs of the job seeker, making recommended positions more attractive and practical.

[0059] Furthermore, by calculating the relative deviations between the application and interview success rates and the expected values, we can quantitatively assess the gap between a candidate's actual performance and their expected goals during the job search process. Next, we multiply the application and interview deviations by preset weights and add them together to create a deviation index. This fully accounts for the varying importance of the application and interview stages in the job search process, balancing their impact through weighted allocation. Finally, based on the comparison of the deviation index with the preset deviation index range, we selectively adjust the resume index or similarity threshold, making the entire matching process more scientific, flexible, and adaptable, effectively improving the efficiency and quality of job matching.

[0060] Furthermore, when the deviation index exceeds the maximum value of the preset range, it indicates that the current matching effect is poor and the resume index overestimates the suitability of the candidate. By calculating the relative deviation of the deviation index from the maximum value and combining it with the preset first adjustment coefficient to reduce the resume index, the over-optimistic estimation in the matching process is effectively reduced, making the matching results closer to actual needs. On the contrary, when the deviation index is lower than the minimum value of the preset range, it indicates that the matching result is too loose, and the similarity threshold may be set too low. At this time, increasing the similarity threshold based on the relative deviation between the delivery deviation and the preset delivery deviation threshold and the preset second adjustment coefficient can tighten the matching criteria and avoid too many unsuitable candidates from entering the subsequent process. Not only can the matching strategy be flexibly adjusted according to the actual matching effect, but the adjustment range can also be accurately controlled in a quantitative manner to ensure the stability and adaptability of the system.

[0061] Furthermore, by counting the number of times the resume index and the similarity threshold are adjusted within the preset correction time, the absolute value of the difference between the two times is calculated to obtain the number difference value, which reflects the system's balance between the resume index adjustment and the similarity threshold adjustment. According to the number difference value and the specific number of adjustments, the candidate job list is corrected, and the matching results can be dynamically optimized according to the frequency differences of the adjustment behaviors.

[0062] Furthermore, by comparing the number of times difference value with the preset number of times difference threshold, when the number of times difference value is greater than the preset threshold, it indicates that there is a significant imbalance between the resume index and the similarity threshold in the adjustment behavior. At this time, the dimension with the larger number of times of adjustment is selected for correction, which can directly optimize the main problem of the system. When the number of times difference value is less than or equal to the preset threshold, it means that the adjustment behavior is relatively balanced in the two dimensions. At this time, it is necessary to further consider the stability of the adjustment. By calculating the standard deviation of the number of adjustments to obtain the fluctuation value, and further calculating the relative deviation of the fluctuation value from the preset threshold, the stability of the adjustment behavior can be quantified. Selecting the dimension with the smaller fluctuation deviation for correction can avoid matching deviation caused by the instability of the adjustment behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a schematic diagram of the artificial intelligence-based job matching system of this embodiment;

[0064] Figure 2 This is a decision logic diagram for the browsing speed decision unit of this embodiment to determine the posts of interest;

[0065] Figure 3 This is a decision logic diagram for determining a temporary position by the temporary determination subunit in this embodiment;

[0066] Figure 4 This is a decision logic diagram for the browsing frequency decision subunit to decide the marking position in this embodiment. DETAILED DESCRIPTION

[0067] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0068] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0069] See also Figure 1 As shown, it is a schematic diagram of the person-job matching system based on artificial intelligence in this embodiment;

[0070] This embodiment provides an artificial intelligence-based job matching system, including:

[0071] The collection module is used to collect the recruitment keywords, historical training frequency, historical promotion speed, target browsing frequency, browsing speed, delivery success rate and interview success rate of each matching position in the position database in real time after the target uploads a resume containing resume keywords.

[0072] a determination module, connected to the acquisition module, for determining a number of positions of interest based on the resume keywords, the recruitment keywords, a preset similarity threshold, the browsing frequency, and the browsing speed;

[0073] An evaluation module, connected to the acquisition module and the determination module, respectively, for evaluating a resume index based on the resume keywords based on a preset artificial intelligence model, and evaluating a recruitment index based on the recruitment keywords of each of the positions of interest;

[0074] a determination module, connected to the collection module and the evaluation module respectively, for determining a list of candidate positions based on the recruitment index, the resume index, the historical training frequency, and the historical promotion speed of each position of interest;

[0075] an adjustment module, connected to the collection module and the determination module respectively, for adjusting the resume index or the preset similarity threshold according to the candidate position list, the submission success rate, and the interview success rate;

[0076] a correction module connected to the adjustment module and the determination module, configured to correct the candidate job list according to the number of times the resume index is adjusted and the number of times the preset similarity threshold is adjusted within a preset correction time period, thereby forming a corrected job list;

[0077] A matching module is connected to the correction module and is used to match a plurality of recommended positions according to the corrected position list.

[0078] The target described in this embodiment refers to job seekers who use recruitment software to apply for jobs. The job database refers to the collection of all data stored in the recruitment software system, including but not limited to job information, company information, and historical data. Each job to be matched refers to all jobs in the job database that are currently open and need to recruit suitable candidates.

[0079] After the target uploads a resume containing resume keywords, the collection module collects multi-dimensional data from the job database in real time. Resume keywords refer to key information related to the job, such as skills, experience, and educational background, mentioned in the job applicant's resume. Recruitment keywords refer to key information related to the job requirements, such as skills, experience, and educational background, mentioned in the job description by the employer. Both can be extracted using natural language processing technology. Historical training frequency refers to the number of training sessions provided by the company to its employees over a certain period of time, reflecting the company's emphasis on employee development and its investment in training resources. Historical promotion rate refers to the average time between employee promotions within the company, reflecting the company's career development opportunities and the efficiency of its promotion mechanisms. Both can be calculated by analyzing the employer's human resources management system using SQL queries or data analysis tools. The target's browsing frequency refers to the frequency with which job applicants browse different job positions on the recruitment software, while browsing rate refers to the average time they spend viewing each job page. Both can be collected from the recruitment software's user behavior logs. The delivery success rate refers to the percentage of job applicants who receive interviews after submitting their resumes, and the interview success rate refers to the percentage of job applicants who are hired after interviewing. Both can be calculated from the job database using SQL queries or data warehouse tools.

[0080] The Resume Index is a quantitative indicator derived from a comprehensive assessment of various information in a job applicant's resume (such as skills, experience, and educational background) using a preset AI model. It reflects the applicant's overall capabilities and background. The Recruitment Index is a quantitative indicator derived from a comprehensive assessment of various requirements in a job description (such as skills, experience, and educational background) using a preset AI model. It reflects the ability requirements for the position.

[0081] The preset artificial intelligence model is a model built based on classification or regression based on machine learning, and is used to evaluate the resume index and recruitment index based on the input feature data (resume keywords, recruitment keywords).

[0082] 1. Model Overview

[0083] The pre-built AI model, based on the BERT architecture, processes text data and extracts semantic information for precise matching. By combining BERT's deep learning capabilities with multi-task learning, the model processes the text features of resumes and job descriptions separately, effectively capturing the semantic and contextual information within the text.

[0084] 2. Initial Model Architecture

[0085] 2.1 Resume Feature Extraction Module

[0086] The resume feature extraction module is based on the BERT architecture and is used to process the semantic features of resume text.

[0087] Input layer: receives the preprocessed resume text, whose dimension is (\text{batch_size}, \text{sequence_length}).

[0088] BERT layer:

[0089] Multi-head self-attention mechanism: used to capture long-range dependencies in text.

[0090] Feedforward neural network: used to further process features.

[0091] Residual connections and layer normalization: used to stabilize the training process.

[0092] Fully connected layer: maps the output of the BERT layer to a low-dimensional CV feature vector.

[0093] Output layer: Outputs the resume feature vector for subsequent analysis.

[0094] 2.2 Job Feature Extraction Module

[0095] The job feature extraction module is also based on the BERT architecture and is used to process the semantic features of job description text.

[0096] Input layer: Receives preprocessed job description text, whose dimension is (\text{batch_size}, \text{sequence_length}).

[0097] BERT layer:

[0098] Multi-head self-attention mechanism: used to capture long-range dependencies in text.

[0099] Feedforward neural network: used to further process features.

[0100] Residual connections and layer normalization: used to stabilize the training process.

[0101] Fully connected layer: maps the output of the BERT layer to a low-dimensional job feature vector.

[0102] Output layer: Output job feature vector for subsequent analysis.

[0103] 2.3 Matching Evaluation Module

[0104] The matching evaluation module is used to evaluate the matching degree between the resume feature vector and the job feature vector, and generate the resume index and recruitment index.

[0105] Input layer: receives resume feature vector and position feature vector.

[0106] Similarity calculation layer: Calculates the similarity between the resume feature vector and the job feature vector, usually using cosine similarity or dot product.

[0107] Fully connected layer: maps similarity to resume index and recruitment index.

[0108] Output layer: output resume index and recruitment index.

[0109] 3. Core parameters

[0110] 3.1 BERT Parameters

[0111] Input dimensions:

[0112] batch_size: The number of data samples in each training batch.

[0113] sequence_length: The length of the text sequence.

[0114] BERT layer parameters:

[0115] Multi-head self-attention mechanism:

[0116] num_heads: The number of attention heads used to process features from different subspaces in parallel.

[0117] hidden_dim: The hidden dimension of each attention head.

[0118] Feedforward Neural Network:

[0119] feed_forward_dim: The intermediate dimension of the feedforward network.

[0120] Activation function: ReLU or GELU is usually used.

[0121] Residual connections and layer normalization: used to stabilize the training process.

[0122] Output dimension: The dimension of the low-dimensional feature vector (such as 128 or 256).

[0123] 3.2 Matching Evaluation Parameters

[0124] Similarity calculation: cosine similarity or dot product.

[0125] Fully connected layer:

[0126] Output dimension: the dimension of resume index and recruitment index (such as 128 or 256).

[0127] 4. Training Process

[0128] 4.1 Input Data Preparation

[0129] Resume text: Preprocess the resume text, including word segmentation and encoding, to form an input dataset with a dimension of (\text{batch_size}, \text{sequence_length}).

[0130] Job description text: The job description text is preprocessed, including word segmentation and encoding, to form an input dataset with a dimension of (\text{batch_size}, \text{sequence_length}).

[0131] 4.2 Model Training

[0132] Loss function:

[0133] CV feature extraction: using mean squared error (MSE) loss.

[0134] Job feature extraction: Use mean square error (MSE) loss.

[0135] Matching evaluation: Use mean squared error (MSE) loss or cross entropy loss.

[0136] Optimizer: Adam optimizer is used, and the learning rate is set to 2×10-5.

[0137] Training process: The model parameters are updated through backpropagation. The training goal is to enable the model to accurately extract the features of resumes and job descriptions and evaluate their matching.

[0138] 4.3 Preset training duration

[0139] Data size: During training, the model needs to learn enough historical data to recognize patterns. The size of the dataset directly affects the training time.

[0140] Convergence: As the loss function decreases, the model gradually converges, and the training time is also affected by the model convergence speed.

[0141] 5. Trained Model

[0142] Model parameters:

[0143] BERT layer weights and biases.

[0144] Weights and biases of the fully connected layers.

[0145] Output features:

[0146] CV feature vector.

[0147] Position feature vector.

[0148] Resume Index and Recruitment Index.

[0149] 6. Use the trained model

[0150] Input new data: Input new resume text and job description text into the trained model.

[0151] Model inference: The model extracts features and outputs resume index and recruitment index for subsequent matching analysis.

[0152] The preset correction period is the time window used to collect statistics and analyze adjustment behaviors. It depends on the system's operating frequency, data update speed, and the dynamic nature of business needs, and is typically set between one week and one month. In this embodiment, it is set to two weeks to accumulate sufficient adjustment data to provide a reliable statistical basis for the correction module, while also adapting to the cyclical changes common in the recruitment market.

[0153] By collecting various key data in real time, we screen out a number of positions of interest based on the collected resume keywords, recruitment keywords, preset similarity thresholds, browsing frequency, and browsing speed. Next, based on a preset AI model, we use resume keywords to evaluate the resume index, and evaluate the recruitment index based on the recruitment keywords of the position of interest. We combine the recruitment index, resume index, historical training frequency, and historical promotion speed of the position of interest to determine a list of candidate positions. Based on the candidate position list, application success rate, and interview success rate, we adjust the resume index or preset similarity threshold. We revise the candidate position list based on the number of adjustments within the preset revision period. Based on the revised candidate position list, we match a number of recommended positions.

[0154] By collecting keywords from the target's uploaded resume, relevant data for each position in the job database, and the target's behavioral data, the system identifies positions of interest based on the keywords in the resume and job posting, combined with browsing frequency and speed. This system then screens for positions that both meet the job requirements and reflect the candidate's genuine interest. An AI model is then used to evaluate the resumes and positions of interest, generating recruitment and resume indices to provide a quantitative basis for job matching. A candidate list is then determined based on the recruitment index, resume index, historical training frequency, and historical promotion rate, taking into account the attractiveness and career prospects of the positions. The resume index or similarity threshold is dynamically adjusted based on the candidate list's submission and interview success rates. The candidate list is then redefined and recommended based on the number of adjustments. This ensures that recommendations are more closely aligned with the target's job search and the actual match between the positions, improving matching accuracy and effectiveness, and helping the target find more suitable positions. This effectively addresses the existing issues of inaccurate matching and slow response times, which are often caused by incomplete data collection and over-reliance on static models.

[0155] Please continue reading Figure 2 As shown, it is a decision logic diagram of the browsing speed decision unit of this embodiment for deciding the posts of interest;

[0156] The determination module includes:

[0157] a vector conversion unit, configured to convert the recruitment keywords into vector form to obtain a plurality of recruitment key vectors, and to convert the resume keywords into vector form to obtain a resume key vector;

[0158] a temporary determination unit connected to the vector conversion unit, configured to determine a number of temporary positions based on the resume key vector and each of the recruitment key vectors;

[0159] a browsing frequency determination unit, connected to the temporary determination unit, for determining a number of marked positions based on the browsing frequency of each temporary position;

[0160] The browsing speed determination unit is connected to the browsing frequency determination unit and is used to determine that the marked post is a focus post when the browsing speed of each marked post is less than a preset browsing speed threshold, so as to determine a number of focus posts.

[0161] The preset browsing speed threshold is a standard value used to measure browsing speed. It depends on the user behavior data of the recruitment software, the complexity of the position, the characteristics of the industry, and the average browsing habits of job seekers. It is usually set between 100 words per second and 300 words per second. In this embodiment, it is set to 200 words per second. This can effectively filter out positions that job seekers are truly interested in and eliminate those that are only quickly browsed without in-depth reading, thereby improving the quality and accuracy of the positions selected.

[0162] By converting recruitment keywords and resume keywords into vectors, we obtain recruitment key vectors and resume key vectors. Next, we identify several temporary positions based on these vectors. We then filter these temporary positions based on their browsing frequency to identify a number of tagged positions. Finally, if the browsing rate of these tagged positions is below a preset browsing rate threshold, they are identified as positions of interest, thereby identifying a number of positions of interest.

[0163] Vector conversion quantifies previously abstract textual information into concrete numerical vectors, providing a viable data foundation for subsequent matching calculations. The vector form more intuitively reflects the similarity and correlation between keywords, facilitating precise matching using mathematical methods. Subsequently, temporary positions are identified based on the relationship between the recruitment key vector and the resume key vector, enabling more objective and accurate screening of positions with a high degree of match to the resume. Job seekers are screened for positions of genuine interest based on browsing frequency and browsing speed. Browsing frequency reflects the level of interest shown by job seekers. Browsing speed further reflects the duration of a job seeker's engagement with job information. Browsing speeds below a preset threshold identify positions of interest, indicating a high degree of alignment between the job seeker's expectations. By first filtering for tagged positions based on browsing frequency and then further filtering for positions of interest based on browsing speed, we can more accurately identify positions of genuine interest to job seekers, avoiding the potential misjudgment associated with relying solely on single-dimensional data.

[0164] Please continue reading Figure 3 As shown, it is a decision logic diagram of the temporary determination subunit determining a temporary position in this embodiment;

[0165] The temporary determination unit includes:

[0166] A similarity calculation subunit, configured to calculate the cosine similarity between the resume key vector and each of the recruitment key vectors to obtain a plurality of similarities;

[0167] The temporary determination subunit is connected to the similarity calculation subunit and is used to determine that the position to be matched is a temporary position when the similarity is greater than a preset similarity threshold, so as to determine a number of temporary positions.

[0168] The preset similarity threshold is a benchmark value used to determine the degree of similarity between the resume key vector and the job posting key vector. It depends on the precise matching requirements of the job posting, the differences in resume quality, and the diversity of the job postings, and is typically set between 0.5 and 0.8. In this embodiment, it is set to 0.7, which can effectively screen out jobs that are highly relevant to the resume, reduce false positives, and improve the quality and accuracy of temporary jobs.

[0169] By calculating the cosine similarity between the resume key vector and each job key vector, a number of similarity values ​​are obtained. These similarity values ​​are then compared with a preset similarity threshold. If the similarity of a matching position exceeds the preset similarity threshold, the position is determined to be temporary, and a number of temporary positions are determined.

[0170] By calculating the cosine similarity between the resume's key vector and each job posting's key vector, a quantifiable basis for job matching is provided. Cosine similarity effectively measures the directional correlation between two vectors and is unaffected by vector length, more accurately reflecting the semantic association between keywords. Next, when the similarity exceeds a threshold, the matching position is determined to be temporary. This effectively selects positions with a high degree of match to the resume, avoids interference from less relevant positions, and improves the efficiency and accuracy of job screening.

[0171] Please continue reading Figure 4 As shown, it is a decision logic diagram of the browsing frequency decision subunit in this embodiment for deciding the marked position;

[0172] The browsing frequency determination unit includes:

[0173] A browsing frequency fluctuation calculation subunit, configured to calculate a standard deviation of the browsing frequency within a preset determination time period to obtain a browsing frequency fluctuation value;

[0174] a browsing frequency average calculation subunit, configured to calculate an average value of the browsing frequencies within the preset determination time period to obtain a browsing frequency average;

[0175] The browsing frequency determination subunit is connected to the browsing frequency fluctuation calculation subunit and the browsing frequency mean calculation subunit respectively, and is used to determine that the temporary position is a marked position when the browsing frequency fluctuation value is less than the preset browsing frequency fluctuation threshold and the browsing frequency mean is greater than the preset browsing frequency mean threshold, so as to determine a number of marked positions.

[0176] The preset judgment period refers to the time interval used to calculate the browsing frequency fluctuation value and the browsing frequency mean. It depends on the periodicity of user behavior data on the recruitment platform and the fluctuation pattern of job popularity, and is usually set between 12 hours and 2 days. In this embodiment, it is set to 1 day to better balance the timeliness and stability of the data. This not only reflects the trend of job seekers' interest in positions in a relatively short period of time, but also avoids excessive data fluctuations that affect judgment accuracy due to a short period of time.

[0177] The preset browsing frequency fluctuation threshold is a standard deviation threshold used to measure browsing frequency stability. It depends on the popularity of the position and the stability of job seekers' attention, and is typically set between 0.1 and 0.3. In this embodiment, it is set to 0.2. This effectively filters out positions with relatively stable browsing frequencies, avoids misjudgments caused by occasional high or low browsing frequencies by job seekers, and thus improves the reliability of marked positions.

[0178] The preset average viewing frequency threshold is used to measure the average value of the overall job interest. It depends on the job's attractiveness and market demand, and is typically set between 5 and 10 times per day. In this example, it is set to 8 times per day to ensure that the selected jobs have high interest, while avoiding missing some potential jobs due to excessively high thresholds, thereby improving the quality and quantity of marked jobs.

[0179] By calculating the standard deviation of the browsing frequency, the browsing frequency fluctuation value is obtained; at the same time, the average value of the browsing frequency is calculated to obtain the browsing frequency mean; when the browsing frequency fluctuation value is less than the preset fluctuation threshold and the browsing frequency mean is greater than the preset mean threshold, the temporary position is determined to be a marked position.

[0180] By considering both the fluctuation value and the mean value, a more comprehensive assessment of a job seeker's genuine interest in a position can be achieved. The fluctuation value of browsing frequency reflects the stability of a job seeker's browsing behavior. A smaller fluctuation value indicates that the job seeker's interest in the position is relatively stable over a predetermined period of time, rather than occasional or random browsing. The mean browsing frequency, on the other hand, reflects the overall level of interest in the position. A higher mean value indicates a high level of interest. A position is marked only when the fluctuation value is less than the preset threshold and the mean value is greater than the preset threshold. This effectively identifies positions that are consistently and highly viewed by job seekers, avoiding misjudgments due to accidental factors.

[0181] Specifically, the determination module includes:

[0182] An index comparison unit, configured to compare the recruitment index and the resume index to obtain an index comparison result;

[0183] The candidate determination unit is connected to the index comparison unit and is used to determine the concerned position as a candidate position when the index comparison result shows that the resume index is greater than the recruitment index, so as to determine a number of candidate positions.

[0184] A ranking unit is connected to the candidate determination unit and is used to determine a candidate position list according to the historical training frequency and the historical promotion speed of each candidate position.

[0185] By comparing the recruitment index and the resume index, when the resume index is greater than the recruitment index, several candidate positions are determined, and based on the historical training frequency and historical promotion speed of these candidate positions, the candidate position list is further determined.

[0186] Determining candidate positions using the Recruitment Index and Resume Index provides an intuitive measure of the match between job seekers and positions. The Recruitment Index reflects the job's requirements and expectations for the candidate, while the Resume Index reflects the candidate's abilities and qualifications. When the Resume Index is greater than the Recruitment Index, it indicates that the candidate's abilities and qualifications exceed the job requirements to a certain extent. Next, candidate positions are ranked using historical training frequency and historical promotion rates, reflecting the company's employee training and development opportunities. Higher training frequency and promotion rates indicate a company's greater focus on employee growth and development, and job seekers in such positions may have greater room for development and better career prospects. The final list of candidate positions not only meets the candidate's abilities and needs, but also provides them with better career development opportunities.

[0187] Specifically, the sorting unit includes:

[0188] A training curve drawing subunit is used to draw a change curve of the historical training frequency within a preset sorting time to obtain a plurality of training curves;

[0189] A promotion curve drawing subunit is used to draw a change curve of the historical promotion speed within the preset sorting time to obtain a plurality of promotion curves;

[0190] a correlation calculation subunit, connected to the training curve drawing subunit and the promotion curve drawing subunit respectively, for calculating the correlation coefficients of the training curve and the promotion curve to obtain a plurality of correlation coefficient values;

[0191] The sorting subunit is connected to the correlation calculation subunit and is used to sort the correlation coefficient values ​​of all the candidate positions from high to low to obtain a candidate position list.

[0192] The preset sorting period is used to collect and analyze historical training frequency and promotion rate changes. It depends on factors such as the applicant's career development stage, the industry characteristics of the position, and the company's talent development cycle, and is usually set between one and five years. In this example, it is set to three years, which is consistent with the employee training and promotion cycles of most companies, making the sorting results more valuable for reference.

[0193] By plotting the historical training frequency change curves and the historical promotion speed change curves for each position within a preset sorting period, several training curves and promotion curves are obtained; then, the correlation coefficients between the training curves and the promotion curves are calculated to obtain several correlation coefficient values; all candidate positions are sorted from high to low according to the correlation coefficient values ​​to obtain a list of candidate positions.

[0194] By plotting the frequency of training and the rate of promotion and calculating their correlation coefficient, we can quantitatively assess the relationship between training and development opportunities within a position. A higher correlation coefficient indicates a strong positive correlation between training opportunities and promotion rate. In other words, the more training a company provides, the faster its employees are promoted, reflecting the company's commitment to providing better training and development opportunities for its employees. By ranking candidate positions based on the correlation coefficient, we can prioritize those that prioritize employee growth and development. This approach not only considers the current fit of a position but also the career development needs of the job seeker, making recommended positions more attractive and practical.

[0195] Specifically, the adjustment module includes:

[0196] a delivery deviation calculation unit, configured to calculate a relative deviation between the delivery success rate and a preset expected delivery success rate to obtain a delivery deviation;

[0197] An interview deviation calculation unit, used to calculate the relative deviation between the interview success rate and a preset expected interview success rate to obtain the interview deviation;

[0198] a deviation index calculation unit, connected to the delivery deviation calculation unit and the interview deviation calculation unit, respectively, for performing a weighted summation of the delivery deviation, the preset delivery weight, the interview deviation, and the preset interview weight to obtain a deviation index;

[0199] a deviation comparison unit connected to the deviation index calculation unit, for comparing the deviation index with a preset deviation index range to obtain a deviation comparison result;

[0200] An adjustment unit is connected to the deviation comparison unit and is used to adjust the resume index according to the deviation index when the deviation comparison result is that the deviation index is greater than the maximum value of the preset deviation index range, or to adjust the preset similarity threshold according to the delivery deviation when the deviation comparison result is that the deviation index is less than the minimum value of the preset deviation index range.

[0201] The preset expected success rate is the ideal percentage of applicants receiving interviews after submitting a resume, set based on industry averages or historical data. It depends on the level of competition in the industry, the demand for the position, and the overall quality of the applicant, and is typically set between 30% and 70%. In this embodiment, it is set at 40%, which is a good reflection of the average performance for most positions and is neither too high nor too low, thus helping to reasonably assess the effectiveness of applicants' application results.

[0202] The preset expected interview success rate is a set interview pass rate based on industry averages or historical data. It depends on the intensity of competition for the position, the company's recruitment standards, and the quality of the applicant, and is typically set between 20% and 40%. In this embodiment, it is set at 30%, which effectively balances the expectations of both the applicant and the company, being neither too lenient nor too strict, and thus facilitating a reasonable assessment of the applicant's interview performance.

[0203] The preset delivery weight refers to the weight proportion of the delivery deviation when calculating the deviation index. It depends on the importance of the delivery success rate to the job seeker's overall job search results and is usually set between 0.3 and 0.7. In this embodiment, it is set to 0.4, which can reasonably reflect the importance of the delivery success rate in the job search process and is neither too high nor too low.

[0204] The preset interview weight refers to the weighting ratio of interview bias when calculating the bias index. It depends on the importance of the interview success rate to the job seeker's final job search outcome and is typically set between 0.3 and 0.7. In this example, it is set to 0.6, which reasonably reflects the critical role of the interview success rate in the job search process. It is higher than the submission weight, reflecting the decisive role of the interview in the job search process, helping to more accurately adjust the resume index and improve the job seeker's matching degree.

[0205] The preset deviation index range is a numerical range used to evaluate whether the deviation index is within a reasonable range. It depends on the business objectives of the system, statistical analysis of historical data, and expectations of matching results, and is usually set between 0 and 1. In this embodiment, it is set between 0.3 and 0.7, which can effectively identify situations with large deviations while avoiding frequent misjudgments caused by overly strict thresholds, thereby balancing the stability and flexibility of the system, ensuring timely adjustments when the deviation is large, and maintaining system stability when the deviation is small.

[0206] The delivery deviation is obtained by calculating the relative deviation between the delivery success rate and the preset expected delivery success rate. The interview deviation is also obtained by calculating the relative deviation between the interview success rate and the preset expected interview success rate. Next, the delivery deviation and interview deviation are multiplied by the preset delivery weight and interview weight respectively, and then weighted summed to obtain the deviation index. Then, the deviation index is compared with the preset deviation index range. If the deviation index is greater than the maximum value of the preset deviation index range, it means that the delivery and interview results are significantly lower than expected. At this time, the resume index is adjusted according to the deviation index. If the deviation index is less than the minimum value of the preset deviation index range, it means that the delivery and interview results are significantly higher than expected. At this time, the preset similarity threshold is adjusted according to the delivery deviation.

[0207] When the Deviation Index is higher than the maximum value within the preset range, it indicates that the current matching effect is significantly different from expectations. This may be due to inaccurate resume evaluation. By adjusting the resume index, you can re-evaluate the candidate's match and thus improve the accuracy of the match. When the Deviation Index is lower than the minimum value within the preset range, it indicates that the current matching effect is far better than expected. This may be due to the similarity threshold being set too low. By adjusting the similarity threshold, you can increase the strictness of the match and avoid over-matching.

[0208] By calculating the relative deviations between the application and interview success rates and the expected values, we can quantitatively assess the gap between a candidate's actual performance and their expected goals during the job search process. Next, we multiply the application and interview deviations by preset weights and sum them to create a deviation index. This fully accounts for the varying importance of the application and interview stages in the job search process, balancing their impact through weighted allocation. Finally, based on the comparison of the deviation index with the preset deviation index range, we selectively adjust the resume index or similarity threshold, making the entire matching process more scientific, flexible, and adaptable, effectively improving the efficiency and quality of job matching.

[0209] Specifically, the adjustment unit includes:

[0210] The first adjustment subunit is used to adjust the resume index according to the relative deviation between the deviation index and the maximum value of the preset deviation index range and the preset first adjustment coefficient when the deviation index is greater than the maximum value of the preset deviation index range, so as to obtain an adjusted resume index, and the relative deviation is negatively correlated with the adjusted resume index.

[0211] The second adjustment subunit is used to adjust the preset similarity threshold according to the preset delivery deviation threshold and the relative deviation of the delivery deviation and the preset second adjustment coefficient when the deviation index is less than or equal to the minimum value of the preset deviation index range, so as to obtain the adjusted similarity threshold, and the relative deviation is positively correlated with the adjusted resume index.

[0212] The preset first adjustment coefficient is used to adjust the resume index. It depends on the applicant's performance stability during the application and interview process, as well as the company's expectations of the applicant's abilities. It is usually set between 0.5 and 0.9. In this embodiment, it is set to 0.7, which makes the resume index more sensitive to changes in the applicant's performance during the application and interview process, thereby more accurately matching them to positions.

[0213] The preset delivery deviation threshold measures the maximum acceptable deviation between the delivery success rate and the expected delivery success rate. It is determined by the analysis of historical data, the nature of the recruitment position, and the stringency of the recruitment goals, and is typically set between 5% and 15%. In this embodiment, it is set to 10%, which effectively balances the strictness and flexibility of the matching process, avoiding inaccurate matching results due to excessive deviation, while also avoiding excessively strict matching results due to too small a deviation.

[0214] The preset second adjustment coefficient is used to adjust the preset similarity threshold. It depends on the system's sensitivity to delivery bias and the similarity threshold adjustment strategy, and is typically set between 0.1 and 0.5. In this embodiment, it is set to 0.3 to ensure that the adjustment of the similarity threshold to delivery bias is appropriate, effectively reflecting the impact of delivery bias without over-adjusting the similarity threshold, thereby maintaining system stability and accuracy.

[0215] When the deviation index exceeds the maximum value of the preset deviation index range, the resume index is adjusted according to the relative deviation between the deviation index and the maximum value, and the preset first adjustment coefficient; when the deviation index is less than or equal to the minimum value of the preset deviation index range, the preset similarity threshold is adjusted according to the relative deviation between the preset delivery deviation threshold and the actual delivery deviation, and the preset second adjustment coefficient.

[0216] When the deviation index exceeds the maximum value of the preset range, it means that the current matching effect is poor and the resume index overestimates the suitability of the candidate. By calculating the relative deviation of the deviation index from the maximum value and combining it with the preset first adjustment coefficient to reduce the resume index, the over-optimistic estimation in the matching process is effectively reduced, making the matching results closer to actual needs. On the contrary, when the deviation index is lower than the minimum value of the preset range, it indicates that the matching result is too loose, and the similarity threshold may be set too low. At this time, increasing the similarity threshold based on the relative deviation between the delivery deviation and the preset delivery deviation threshold and the preset second adjustment coefficient can tighten the matching criteria and avoid too many unsuitable candidates from entering the subsequent process. It can not only flexibly adjust the matching strategy according to the actual matching effect, but also accurately control the adjustment range in a quantitative manner to ensure the stability and adaptability of the system.

[0217] Specifically, the correction module includes:

[0218] A times acquisition unit is used to acquire the number of the adjusted resume index obtained within the preset correction time to form the index adjustment times, and to acquire the number of the adjusted similarity threshold obtained within the preset correction time to form the threshold adjustment times;

[0219] a number difference calculation unit, connected to the number acquisition unit, for calculating the absolute value of the difference between the index adjustment number and the threshold adjustment number to obtain a number difference value;

[0220] A correction unit is connected to the number difference calculation, and is used to correct the candidate position list according to the number difference value, the index adjustment number and the threshold adjustment number.

[0221] The number of index adjustments is calculated by taking the number of resume index adjustments, and the number of threshold similarity adjustments is calculated by taking the number of threshold adjustments. The absolute difference between the index and threshold adjustments is then calculated to obtain the number difference. Finally, the candidate job list is modified based on the number difference and the specific number of index and threshold adjustments.

[0222] By counting the number of times the resume index and the similarity threshold are adjusted within the preset correction time, the absolute value of the difference between the two times is calculated to obtain the number difference value, which reflects the system's balance between resume index adjustment and similarity threshold adjustment. According to the number difference value and the specific number of adjustments, the candidate job list is corrected, and the matching results can be dynamically optimized according to the frequency differences of the adjustment behaviors.

[0223] Specifically, the correction unit includes:

[0224] A first correction subunit is configured to correct the candidate job list based on the adjusted resume index to form a corrected job list when the number difference value is greater than a preset number difference threshold and the number of index adjustments is greater than the threshold adjustment number, and to correct the candidate job list based on the adjusted similarity threshold to form a corrected job list when the number difference value is greater than the preset number difference threshold and the number of index adjustments is less than the threshold adjustment number;

[0225] an adjustment fluctuation calculation subunit, configured to calculate, when the number difference value is less than or equal to the preset number difference threshold, a standard deviation of the number of index adjustments within the preset correction time period to obtain an index adjustment fluctuation value, and to calculate a standard deviation of the number of threshold adjustments within the preset correction time period to obtain a threshold adjustment fluctuation value;

[0226] a fluctuation deviation calculation subunit, connected to the adjustment fluctuation calculation subunit, for calculating a relative deviation between the index adjustment fluctuation value and a preset index adjustment threshold value to obtain an index adjustment deviation, and for calculating a relative deviation between the threshold adjustment fluctuation value and a preset threshold adjustment threshold value to obtain a threshold adjustment deviation;

[0227] The second correction subunit is connected to the fluctuation deviation calculation subunit and is used to correct the candidate job list based on the adjusted resume index to form a corrected job list when the index adjustment deviation is greater than the threshold adjustment deviation, and to correct the candidate job list based on the adjusted similarity threshold to form a corrected job list when the index adjustment deviation is less than or equal to the threshold adjustment deviation.

[0228] The preset number difference threshold is a reference value used to determine whether the difference between the index adjustment number and the threshold adjustment number is significant. It depends on the system's adjustment frequency, business needs, and the fluctuation of historical data, and is usually set between 0 and 10. In this embodiment, it is set to 5, which can ensure the flexibility of system adjustment while avoiding unnecessary corrections due to minor differences, thereby improving the stability and adaptability of the system.

[0229] The preset index adjustment threshold is a reference value used to determine whether the index adjustment fluctuation value is within a reasonable range. It depends on the historical fluctuation of the resume index adjustment and the business requirements for matching accuracy, and is usually set between 0.1 and 0.5. In this embodiment, it is set to 0.3, which can effectively filter out misjudgments caused by accidental fluctuations, ensure the stability of the index adjustment, and at the same time take into account matching accuracy.

[0230] The preset threshold adjustment threshold is a reference value used to determine whether the fluctuation value of the similarity threshold adjustment is within a reasonable range. It depends on the historical fluctuation of the similarity threshold adjustment and the business requirements for matching strictness, and is usually set between 0.1 and 0.5. In this embodiment, it is set to 0.3, which can effectively avoid unstable matching results caused by frequent similarity threshold adjustments while ensuring the rationality of the matching criteria.

[0231] When the number difference value is greater than the preset number difference threshold, if the index adjustment number is greater than the threshold adjustment number, the candidate position list will be corrected based on the adjusted resume index, otherwise, the candidate position list will be corrected based on the adjusted similarity threshold. If the number difference value is less than or equal to the preset number difference threshold, the standard deviation of the index adjustment number and the standard deviation of the threshold adjustment number within the preset correction time are further calculated to obtain the index adjustment fluctuation value and the threshold adjustment fluctuation value. Then, the relative deviation between the index adjustment fluctuation value and the preset index adjustment threshold is calculated to obtain the index adjustment deviation, and the relative deviation between the threshold adjustment fluctuation value and the preset threshold adjustment threshold is calculated to obtain the threshold adjustment deviation. Finally, when the index adjustment deviation is greater than the threshold adjustment deviation, the candidate position list is corrected based on the adjusted resume index to form a corrected position list. Otherwise, the candidate position list is corrected based on the adjusted similarity threshold to form a corrected position list. After selecting the adjustment object, the adjustment object is adjusted according to the adjustment method of the adjustment module, and the candidate position list is regenerated to obtain a corrected position list.

[0232] By comparing the number of times difference value with the preset number of times difference threshold, when the number of times difference value is greater than the preset threshold, it indicates that the adjustment behavior is significantly unbalanced between the resume index and the similarity threshold. In this case, the dimension with the larger number of times of adjustment is selected for correction, which can directly optimize the main problem of the system. When the number of times difference value is less than or equal to the preset threshold, it means that the adjustment behavior is relatively balanced in both dimensions. At this time, it is necessary to further consider the stability of the adjustment. By calculating the standard deviation of the number of adjustments to obtain the fluctuation value, and further calculating the relative deviation of the fluctuation value from the preset threshold, the stability of the adjustment behavior can be quantified. Selecting the dimension with the smaller fluctuation deviation for correction can avoid matching deviation caused by the instability of the adjustment behavior.

[0233] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based job matching system, characterized in that: include: A collection module is used to collect, in real time, the recruitment keywords, historical training frequency, historical promotion speed, browsing frequency, browsing speed, submission success rate, and interview success rate of each matching position in the job database after the target uploads a resume containing resume keywords, wherein the target refers to a job seeker who uses the recruitment software to apply for a job; a determination module, connected to the acquisition module, for determining a number of positions of interest based on the resume keywords, the recruitment keywords, a preset similarity threshold, the browsing frequency, and the browsing speed; an evaluation module, connected to the acquisition module and the determination module, respectively, for evaluating resume information according to the resume keywords based on a preset artificial intelligence model to obtain a resume index, and for evaluating job information according to the recruitment keywords of each of the positions of interest to obtain a recruitment index, wherein the resume index refers to an indicator used to evaluate the information in the job applicant's resume through the artificial intelligence model to reflect the job applicant's overall ability and background, and the recruitment index refers to an indicator used to evaluate the various requirements in the job description through the artificial intelligence model to reflect the ability requirements of the position; a determination module, connected to the collection module and the evaluation module respectively, for determining a list of candidate positions based on the recruitment index, the resume index, the historical training frequency, and the historical promotion speed of each position of interest; an adjustment module, connected to the collection module and the determination module respectively, for adjusting the resume index or the preset similarity threshold according to the candidate position list, the submission success rate, and the interview success rate; a correction module connected to the adjustment module and the determination module, configured to correct the candidate job list according to the number of times the resume index is adjusted and the number of times the preset similarity threshold is adjusted within a preset correction time period, thereby forming a corrected job list; a matching module, connected to the correction module, for matching a plurality of recommended positions according to the corrected position list; The adjustment module includes: a delivery deviation calculation unit, configured to calculate a relative deviation between the delivery success rate and a preset expected delivery success rate to obtain a delivery deviation; An interview deviation calculation unit, used to calculate the relative deviation between the interview success rate and a preset expected interview success rate to obtain the interview deviation; a deviation index calculation unit, connected to the delivery deviation calculation unit and the interview deviation calculation unit, respectively, for performing a weighted summation of the delivery deviation, the preset delivery weight, the interview deviation, and the preset interview weight to obtain a deviation index; a deviation comparison unit connected to the deviation index calculation unit, for comparing the deviation index with a preset deviation index range to obtain a deviation comparison result; an adjustment unit connected to the deviation comparison unit, configured to adjust the resume index according to the deviation index when the deviation comparison result shows that the deviation index is greater than the maximum value of the preset deviation index range, or to adjust the preset similarity threshold according to the delivery deviation when the deviation comparison result shows that the deviation index is less than the minimum value of the preset deviation index range; The correction module includes: A times acquisition unit is used to acquire the number of the adjusted resume index obtained within the preset correction time to form the index adjustment times, and to acquire the number of the adjusted similarity threshold obtained within the preset correction time to form the threshold adjustment times; a number difference calculation unit, connected to the number acquisition unit, for calculating the absolute value of the difference between the index adjustment number and the threshold adjustment number to obtain a number difference value; A correction unit is connected to the number difference calculation, and is used to correct the candidate position list according to the number difference value, the index adjustment number and the threshold adjustment number.

2. The artificial intelligence-based job matching system according to claim 1 is characterized in that: The determination module includes: a vector conversion unit, configured to convert the recruitment keywords into vector form to obtain a plurality of recruitment key vectors, and to convert the resume keywords into vector form to obtain a resume key vector; a temporary determination unit connected to the vector conversion unit, configured to determine a number of temporary positions based on the resume key vector and each of the recruitment key vectors; a browsing frequency determination unit, connected to the temporary determination unit, for determining a number of marked positions based on the browsing frequency of each temporary position; The browsing speed determination unit is connected to the browsing frequency determination unit and is used to determine that the marked post is a focus post when the browsing speed of each marked post is less than a preset browsing speed threshold, so as to determine a number of focus posts.

3. The artificial intelligence-based job matching system according to claim 2 is characterized in that: The temporary determination unit includes: A similarity calculation subunit, configured to calculate the cosine similarity between the resume key vector and each of the recruitment key vectors to obtain a plurality of similarities; The temporary determination subunit is connected to the similarity calculation subunit and is used to determine that the position to be matched is a temporary position when the similarity is greater than a preset similarity threshold, so as to determine a number of temporary positions.

4. The artificial intelligence-based job matching system according to claim 3 is characterized in that: The browsing frequency determination unit includes: A browsing frequency fluctuation calculation subunit, configured to calculate a standard deviation of the browsing frequency within a preset determination time period to obtain a browsing frequency fluctuation value; a browsing frequency average calculation subunit, configured to calculate an average value of the browsing frequencies within the preset determination time period to obtain a browsing frequency average; The browsing frequency determination subunit is connected to the browsing frequency fluctuation calculation subunit and the browsing frequency mean calculation subunit respectively, and is used to determine that the temporary position is a marked position when the browsing frequency fluctuation value is less than the preset browsing frequency fluctuation threshold and the browsing frequency mean is greater than the preset browsing frequency mean threshold, so as to determine a number of marked positions.

5. The artificial intelligence-based job matching system according to claim 4 is characterized in that: The determination module includes: An index comparison unit, configured to compare the recruitment index and the resume index to obtain an index comparison result; a candidate determination unit connected to the index comparison unit, configured to determine the position of interest as a candidate position when the index comparison result shows that the resume index is greater than the recruitment index, thereby determining a number of candidate positions; A ranking unit is connected to the candidate determination unit and is used to determine a candidate position list according to the historical training frequency and the historical promotion speed of each candidate position.

6. The artificial intelligence-based job matching system according to claim 5, characterized in that: The sorting unit includes: A training curve drawing subunit is used to draw a change curve of the historical training frequency within a preset sorting time to obtain a plurality of training curves; A promotion curve drawing subunit is used to draw a change curve of the historical promotion speed within the preset sorting time to obtain a plurality of promotion curves; a correlation calculation subunit, connected to the training curve drawing subunit and the promotion curve drawing subunit respectively, for calculating the correlation coefficients of the training curve and the promotion curve to obtain a plurality of correlation coefficient values; The sorting subunit is connected to the correlation calculation subunit and is used to sort the correlation coefficient values ​​of all the candidate positions from high to low to obtain a candidate position list.

7. The artificial intelligence-based job matching system according to claim 6 is characterized in that: The adjustment unit includes: a first adjustment subunit, configured to adjust the resume index according to a relative deviation between the deviation index and the maximum value of the preset deviation index range and a preset first adjustment coefficient when the deviation index is greater than a maximum value of the preset deviation index range, to obtain an adjusted resume index, wherein the relative deviation is negatively correlated with the adjusted resume index; The second adjustment subunit is used to adjust the preset similarity threshold according to the preset delivery deviation threshold and the relative deviation of the delivery deviation and the preset second adjustment coefficient when the deviation index is less than or equal to the minimum value of the preset deviation index range, so as to obtain the adjusted similarity threshold, and the relative deviation is positively correlated with the adjusted resume index.

8. The artificial intelligence-based job matching system according to claim 7, characterized in that: The correction unit includes: A first correction subunit is configured to correct the candidate job list based on the adjusted resume index to form a corrected job list when the number difference value is greater than a preset number difference threshold and the number of index adjustments is greater than the threshold adjustment number, and to correct the candidate job list based on the adjusted similarity threshold to form a corrected job list when the number difference value is greater than the preset number difference threshold and the number of index adjustments is less than the threshold adjustment number; an adjustment fluctuation calculation subunit, configured to calculate, when the number difference value is less than or equal to the preset number difference threshold, a standard deviation of the number of index adjustments within the preset correction time period to obtain an index adjustment fluctuation value, and to calculate a standard deviation of the number of threshold adjustments within the preset correction time period to obtain a threshold adjustment fluctuation value; a fluctuation deviation calculation subunit, connected to the adjustment fluctuation calculation subunit, for calculating a relative deviation between the index adjustment fluctuation value and a preset index adjustment threshold value to obtain an index adjustment deviation, and for calculating a relative deviation between the threshold adjustment fluctuation value and a preset threshold adjustment threshold value to obtain a threshold adjustment deviation; The second correction subunit is connected to the fluctuation deviation calculation subunit and is used to correct the candidate job list based on the adjusted resume index to form a corrected job list when the index adjustment deviation is greater than the threshold adjustment deviation, and to correct the candidate job list based on the adjusted similarity threshold to form a corrected job list when the index adjustment deviation is less than or equal to the threshold adjustment deviation.

Images