A global human resources service intelligent matching system based on big data
By introducing big data technology and gradient descent model into the human resource matching system, dynamically adjusting the weights, real-time human resource matching is achieved, solving the problem that traditional systems cannot adapt to market changes, and improving the accuracy and efficiency of matching.
Patent Information
- Application Number
- CN202410295958.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-15
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-03-15
AI Technical Summary
Due to the offline training static model, traditional human resources matching systems are difficult to adapt to changes in market and user needs. Relying on manual intervention leads to high costs and low efficiency, and cannot be adjusted in time to meet changes in market dynamics and user behavior.
Adopt the global human resources service intelligent matching system based on big data, and collect data flows in real time through the data acquisition module, and clean data by preprocessing module. The feature selection module dynamically selects features. The weight adjustment module introduces a gradient descent model and a dynamic weight adjustment mechanism. The model is The evaluation module automatically corrects model deviations, the calculation module calculates the skill matching scores of candidates and positions in real time, and the matching recommendation module generates a sorting table and recommends candidates.
It improves the real-time and accuracy of the matching system, reduces the demand for manual intervention, reduces labor costs, enhances recruitment efficiency and the accuracy of personalized matching, and can promptly reflect market changes and user behavior.
Smart Images

Figure CN118052530B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of human resource matching systems, and in particular to a global human resource service intelligent matching system based on big data. Background Art
[0002] With the development of globalization and technology, companies and individuals face more complex challenges in the recruitment and career development process. Traditional recruitment methods may no longer be sufficient to meet the rapidly changing market demands. The Internet era has brought a large amount of recruitment information and candidate data, which makes manual screening cumbersome and error-prone. The matching system can process this information more quickly and accurately, and improve recruitment efficiency. The matching system is a system that uses advanced information technology and artificial intelligence technology to improve the efficiency and accuracy of human resource management, and aims to better meet the recruitment, talent management and career development needs of companies and individuals.
[0003] The prior art has the following deficiencies:
[0004] Traditional matching systems have many limitations. First, the matching system uses a static model trained offline, which makes it difficult for the system to adapt to the ever-changing market and user needs. The model update lags and the matching accuracy is significantly affected. Second, traditional systems usually rely on manual intervention for data processing, model adjustment and feedback interpretation. This process is time-consuming and expensive, which increases costs while limiting the efficiency and flexibility of the system. Finally, the limited processing capacity of the matching system for real-time data makes it difficult for the traditional matching system to respond quickly to immediate changes in market dynamics and user behavior, which in turn causes the matching system to be unable to adjust in time to meet actual needs. Summary of the invention
[0005] The purpose of the present invention is to provide a global human resource service intelligent matching system based on big data to solve the shortcomings of the background technology.
[0006] In order to achieve the above-mentioned object, the present invention provides the following technical solutions: a global human resource service intelligent matching system based on big data, comprising a data collection module, a preprocessing module, a feature selection module, a weight adjustment module, a model evaluation module, a calculation module, and a matching recommendation module;
[0007] Data collection module: Integrate real-time data stream technology to collect data streams from large databases in real time. The data streams include recruitment and candidate data.
[0008] Preprocessing module: preprocess the data stream using the real-time processing engine;
[0009] Feature selection module: dynamically selects representative feature data in the data stream according to the changes in real-time data;
[0010] Weight adjustment module: learns and adapts to changing data streams in real time through the gradient descent model, and introduces a dynamic weight adjustment mechanism in the gradient descent model to dynamically adjust the weight of features according to the importance of real-time data;
[0011] Model evaluation module: After evaluating the performance of the gradient descent model, it automatically discovers and corrects the deviation of the gradient descent model based on the evaluation results;
[0012] Calculation module: For each newly posted position, the skill matching score between the candidate and the position is calculated in real time through the gradient descent model;
[0013] Matching recommendation module: Generates a ranking table based on the skill matching scores between candidates and positions, and recommends a corresponding number of candidates in the positive order of the ranking table to the enterprise.
[0014] In a preferred embodiment, the weight adjustment module introduces a dynamic weight adjustment mechanism in the gradient descent model to dynamically adjust the weight of the feature according to the importance of the real-time data;
[0015] Obtain the adaptive learning rate, momentum, and gradient change rate of real-time data, and calculate the weight of real-time data. The function expression is: In the formula, θnew is the current weight of the real-time data, θold is the previous weight of the real-time data, LR is the learning rate, Vnew is the momentum after the real-time data is updated, GCR is the gradient change rate, is the original weight value of gradient descent;
[0016] The current weight of real-time data is calculated regularly and updated to the gradient descent model for use.
[0017] In a preferred embodiment, the weight adjustment module receives a real-time feature data stream, uses a gradient descent algorithm to perform real-time learning on a gradient descent model, calculates a loss function based on the received feature data and a corresponding target value, and uses a gradient descent algorithm to update the weights of the gradient descent model;
[0018] The function expression of the gradient descent model is: 预测 =θ*X+b, where Y 预测 is the model prediction value, θ is the weight of the feature data, X is the feature data, and b is the bias term;
[0019] The calculation expression of the loss function is:
[0020] Where SZ(θ,b) is the loss function, Y 预测 i is the model prediction value of the i-th sample, Y i is the target value of the i-th sample, and n is the number of samples.
[0021] In a preferred embodiment, the preprocessing module uses a real-time processing engine to clean the data stream, remove missing values and abnormal values, and normalize the data format;
[0022] According to business needs, the real-time processing engine is used to aggregate and count data in real time, calculate real-time statistical indicators of recruitment needs, including the number of recruitment positions and the frequency of skill requirements, and forward the preprocessed data stream to the feature selection module.
[0023] In a preferred embodiment, the updated momentum calculation expression is: Vnew = β*V old +(1-β), where Vold is the previous momentum of the real-time data, β is the hyperparameter of the momentum, and β is between [0,1);
[0024] The calculation expression of the gradient change rate is: In the formula, σ new is the gradient of the loss function of the current iteration with respect to the weight, σ old is the gradient of the loss function of the previous iteration with respect to the weight;
[0025] The learning rate calculation expression is: Where xcs is the initial learning rate, xsjz is the learning rate attenuation factor, and dcs is the number of iterations.
[0026] In a preferred embodiment, the calculation module receives each newly posted job information from the matching system, including job requirements and skill requirements, and also receives relevant information of the candidate, including skills and work experience;
[0027] The gradient descent model is used to calculate the skill matching score between the candidate and the position, the extracted features are input into the model, and the model output is obtained as the matching score;
[0028] The input features of the gradient descent model include the candidate experience vector, the candidate skill vector, and the job skill requirement vector. The function expression of the gradient descent model is:
[0029] pf z =θ0*experience+θ1*candidate+θ2*requirement; where, pf z is the matching score, experience is the candidate experience vector, candidate is the candidate skill vector, requirement is the position skill requirement vector, θ0, θ1, θ2 are the weight parameters of the candidate experience vector, candidate skill vector and position skill requirement vector respectively, and θ0+θ1+θ2=1.
[0030] In a preferred embodiment, the matching recommendation module receives a skill matching score between each candidate and the position from the calculation module;
[0031] Sort the received skill matching scores by scores to generate a sorting table, the data structure of which includes candidate ID and matching score information;
[0032] According to the ranking table, select the corresponding number of candidates with matching scores in positive order. Suppose there are s candidates in the ranking table, and the company needs to recruit p candidates, then select {1, 2, ..., p} candidates from the ranking table to join the company, and screen out {p, p+1, ..., s} candidates;
[0033] Recommend the selected candidate information to corporate users, including the candidate's basic information, skill characteristics, and matching scores. Implement security measures during the recommendation process and desensitize or encrypt information involving user privacy.
[0034] A global human resources service intelligent matching method based on big data, the matching method comprising the following steps:
[0035] Integrate real-time data stream technology to collect data streams in real time from large databases. The data streams include recruitment and candidate data to ensure that the system can reflect market changes in a timely manner. Use real-time processing engines to clean and preprocess data to ensure the real-time and accuracy of data. Dynamically select the most representative features in the data stream according to changes in real-time data to adapt to changing market demands. Use the gradient descent model to learn and adapt to changing data streams in real time, and introduce a dynamic weight adjustment mechanism in the gradient descent model to dynamically adjust the weights of features according to the importance of real-time data to improve the adaptability and robustness of the gradient descent model to real-time data. After evaluating the performance of the gradient descent model, the matching system automatically discovers and corrects the deviations of the gradient descent model in a timely manner based on the evaluation results, thereby improving the performance of the gradient descent model. For each newly released position, the gradient descent model calculates the skill matching score between the candidate and the position in real time, and generates a ranking table after sorting the candidates according to the skill matching score between the candidates and the position, and recommends the corresponding number of candidates in the positive order of the ranking table to the enterprise.
[0036] In the above technical solution, the technical effects and advantages provided by the present invention are:
[0037] The present invention integrates real-time data flow technology through a data acquisition module, collects data flow from a large database in real time, and the weight adjustment module learns and adapts to the changing data flow in real time through a gradient descent model, and introduces a dynamic weight adjustment mechanism in the gradient descent model, dynamically adjusts the weight of the feature according to the importance of real-time data, and after the model evaluation module evaluates the performance of the gradient descent model, the matching system automatically finds and corrects the deviation of the gradient descent model in time according to the evaluation result, improves the performance of the gradient descent model, and for each newly released position, the calculation module calculates the skill matching score between the candidate and the position in real time through the gradient descent model, and the matching recommendation module generates a ranking table after sorting the skill matching score between the candidate and the position, and recommends the corresponding number of candidates in the positive order of the ranking table to the enterprise. The matching system can reflect market changes and user behaviors more timely, greatly improves the real-time performance of the matching system, introduces real-time learning and dynamic weight adjustment mechanisms, enables the system to better adapt to changing market demands and user characteristics, improves the accuracy of personalized matching, reduces the need for manual intervention, reduces labor costs, and improves recruitment efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0039] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION
[0040] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] Example: See Figure 1 As shown, the global human resource service intelligent matching system based on big data described in this embodiment includes a data collection module, a preprocessing module, a feature selection module, a weight adjustment module, a model evaluation module, a calculation module, and a matching recommendation module;
[0042] Data collection module: Integrates real-time data stream technology to collect data streams from large databases in real time. The data streams include recruitment and candidate data to ensure that the system can reflect market changes in a timely manner. The data streams are sent to the pre-processing module;
[0043] Preprocessing module: Use the real-time processing engine to clean and preprocess the data stream to ensure the real-time and accuracy of the data. The preprocessed data stream is sent to the feature selection module;
[0044] Feature selection module: dynamically selects the most representative features in the data stream, such as skills and education, based on changes in real-time data to adapt to changing market demands, and sends feature data to the weight adjustment module;
[0045] Weight adjustment module: The gradient descent model is used to learn and adapt to the changing data stream in real time, and a dynamic weight adjustment mechanism is introduced into the gradient descent model to dynamically adjust the weight of the feature according to the importance of the real-time data, thereby improving the adaptability and robustness of the gradient descent model to the real-time data. The gradient descent model after weight adjustment is sent to the model evaluation module;
[0046] Model evaluation module: After evaluating the performance of the gradient descent model, the matching system automatically finds and corrects the deviation of the gradient descent model in a timely manner based on the evaluation results, improves the performance of the gradient descent model, and sends the performance-adjusted gradient descent model to the calculation module;
[0047] Calculation module: For each newly posted position, the skill matching score between the candidate and the position is calculated in real time through the gradient descent model, and the matching score is sent to the matching recommendation module;
[0048] Matching recommendation module: Generates a ranking table based on the skill matching scores between candidates and positions, and recommends a corresponding number of candidates in the positive order of the ranking table to the enterprise.
[0049] This application integrates real-time data flow technology through the data acquisition module, collects data flow from a large database in real time, and the weight adjustment module learns and adapts to the changing data flow in real time through the gradient descent model, and introduces a dynamic weight adjustment mechanism in the gradient descent model to dynamically adjust the weight of the feature according to the importance of real-time data. After the model evaluation module evaluates the performance of the gradient descent model, the matching system automatically finds and corrects the deviation of the gradient descent model in a timely manner based on the evaluation results, and improves the performance of the gradient descent model. For each newly released position, the calculation module calculates the skill matching score between the candidate and the position in real time through the gradient descent model. The matching recommendation module generates a ranking table after sorting the skill matching score between the candidate and the position, and recommends the corresponding number of candidates in the ranking table to the enterprise. The matching system can reflect market changes and user behavior more timely, greatly improves the real-time performance of the matching system, introduces real-time learning and dynamic weight adjustment mechanisms, enables the system to better adapt to changing market demands and user characteristics, improves the accuracy of personalized matching, reduces the need for manual intervention, reduces labor costs, and improves recruitment efficiency.
[0050] The specific process is:
[0051] Integrate real-time data stream technology to collect data streams in real time from large databases. The data streams include recruitment and candidate data to ensure that the system can reflect market changes in a timely manner. Use real-time processing engines to clean and preprocess data to ensure the real-time and accuracy of data. Dynamically select the most representative features in the data stream, such as skills, education, etc., according to changes in real-time data to adapt to changing market demands. Use the gradient descent model to learn and adapt to changing data streams in real time, and introduce a dynamic weight adjustment mechanism in the gradient descent model to dynamically adjust the weights of features according to the importance of real-time data to improve the adaptability and robustness of the gradient descent model to real-time data. After evaluating the performance of the gradient descent model, the matching system automatically discovers and corrects the deviations of the gradient descent model in a timely manner based on the evaluation results, thereby improving the performance of the gradient descent model. For each newly released position, the gradient descent model calculates the skill matching score between the candidate and the position in real time, and generates a ranking table after sorting the candidates according to the skill matching score between the candidates and the position, and recommends the corresponding number of candidates in the positive order of the ranking table to the enterprise.
[0052] Data collection module: Integrates real-time data stream technology to collect data streams from large databases in real time. The data streams include recruitment and candidate data to ensure that the system can reflect market changes in a timely manner. The data streams are sent to the pre-processing module;
[0053] Real-time data streaming technology integration: Select appropriate real-time data streaming technology, such as Apache Kafka, Apache Flink, etc., for integration. Configure the data streaming technology to ensure that it can connect to the big database and obtain data updates in real time.
[0054] Connect to big databases: Establish connections with big databases to ensure that the data collection module can access recruitment and candidate data. Consider the security and stability of database connections and adopt appropriate authentication and encryption mechanisms.
[0055] Define data stream format: Determine the data stream format of recruitment and candidate data, such as JSON, Avro, etc. Define the fields included in the data stream and the data types of the fields to ensure the accuracy and completeness of the data.
[0056] Real-time data collection: Set up real-time data streaming technology to obtain recruitment and candidate data regularly or in real time. Use incremental data pull or push mechanisms to ensure that only changed data is transmitted, reducing the burden of data transmission.
[0057] Data flow monitoring and error handling: Implement monitoring mechanisms to promptly detect anomalies and errors in data flows. Set up error handling mechanisms and take appropriate measures for abnormal situations, such as logging and sending alerts.
[0058] Data flow quality control: Implement quality control measures to ensure the quality of the collected data. Irregular or erroneous data can be processed through data verification, cleaning and conversion.
[0059] Real-time data transmission to the preprocessing module: The collected real-time data stream is transmitted to the preprocessing module. Ensure that the format of the transmitted data stream matches the interface of the preprocessing module for subsequent processing.
[0060] Preprocessing module: Use the real-time processing engine to clean and preprocess the data stream to ensure the real-time and accuracy of the data. The preprocessed data stream is sent to the feature selection module;
[0061] Real-time processing engine selection: Select a suitable real-time processing engine, such as Apache Flink, Apache Storm, etc., to support real-time processing and streaming computing. Ensure that the real-time processing engine can meet the system's requirements for data processing speed and fault tolerance.
[0062] Data cleaning and format normalization: Use the real-time processing engine to clean the data stream and remove irregular data such as missing values and outliers. Perform data format normalization to ensure that each field in the data stream meets the system specifications.
[0063] Real-time aggregation and statistics: Based on business needs, the real-time processing engine is used to aggregate and count data in real time. Real-time statistical indicators of recruitment needs can be calculated, such as the number of recruitment positions, the frequency of skill requirements, etc.
[0064] Feature extraction: Extract meaningful features from the cleaned and normalized data. Features can include skill requirements, work location, work experience, etc. These features will be used in subsequent matching models.
[0065] Data quality monitoring: Set up a data quality monitoring mechanism to promptly detect abnormalities in the preprocessing process. For low-quality data, appropriate measures can be taken, such as logging and issuing alarms.
[0066] Real-time data stream forwarding: forward the preprocessed data stream to the feature selection module. To ensure the real-time nature of the data, the real-time processing engine needs to have the characteristics of high throughput and low latency.
[0067] Exception handling and fault tolerance mechanism: Implement an exception handling mechanism to capture and handle errors that may occur during preprocessing. Consider the fault tolerance characteristics of the real-time processing engine to ensure that the system can maintain stable operation in the face of abnormal situations.
[0068] Security assurance: Implement corresponding security measures during data processing, such as data encryption and identity authentication, to ensure the security of data processing. Desensitize or encrypt data involving user privacy to protect user privacy.
[0069] Feature selection module: dynamically selects the most representative features in the data stream, such as skills and education, based on changes in real-time data to adapt to changing market demands, and sends feature data to the weight adjustment module;
[0070] Real-time feature extraction: Receive real-time feature data stream from the preprocessing module. Use the real-time processing engine or corresponding algorithm to extract various features in the data stream, such as skills, education, work experience, etc.
[0071] Feature importance assessment: Use feature selection algorithms or models to evaluate the importance of each feature in real time. Common methods include feature importance of information gain, etc.
[0072] Dynamic feature selection: Dynamically select the most representative features based on real-time data changes. Consider the ever-changing market demand and ensure that the selected features can better reflect the current recruitment trends and candidate market.
[0073] Feature optimization and conversion: Optimize and convert the selected features to improve the performance of the matching model. Feature scaling and normalization can be performed to ensure that the features are on the same scale and improve the stability of the model.
[0074] Pass feature data to the weight adjustment module: Pass the selected and optimized feature data to the weight adjustment module. Make sure the format of the passed feature data matches the interface of the weight adjustment module for subsequent processing.
[0075] Weight adjustment module: The gradient descent model is used to learn and adapt to the changing data stream in real time, and a dynamic weight adjustment mechanism is introduced into the gradient descent model to dynamically adjust the weight of the feature according to the importance of the real-time data, thereby improving the adaptability and robustness of the gradient descent model to the real-time data. The gradient descent model after weight adjustment is sent to the model evaluation module;
[0076] Receive feature data stream: Receive real-time feature data stream and ensure that the format of the feature data matches the input of the model.
[0077] Real-time learning and weight update: Use the gradient descent algorithm to learn the gradient descent model in real time. According to the received feature data and the corresponding target value, calculate the loss function, and use the gradient descent algorithm to update the weight of the gradient descent model;
[0078] The function expression of the gradient descent model is: 预测 =θ*X+b, where Y 预测 is the model prediction value, θ is the weight of the feature data, X is the feature data, and b is the bias term;
[0079] The calculation expression of the loss function is:
[0080] Where SZ(θ,b) is the loss function, Y 预测 i is the model prediction value of the i-th sample, Yi is the target value of the i-th sample, and n is the number of samples.
[0081] Dynamic weight adjustment mechanism: Introduce a dynamic weight adjustment mechanism to dynamically adjust the weight of features based on the importance of real-time data. Use learning rate, momentum, and gradient change rate to ensure that weights can flexibly adapt to data changes;
[0082] The function expression of the bias term is: In the formula, bold is the historical bias, LR is the learning rate, is the partial derivative of the bias term, is the partial derivative of the loss function.
[0083] The establishment of a dynamic weight adjustment mechanism includes the following steps:
[0084] Obtain the adaptive learning rate, momentum, and gradient change rate of real-time data, and calculate the weight of real-time data. The function expression is: In the formula, θnew is the current weight of the real-time data, θold is the previous weight of the real-time data, LR is the learning rate, Vnew is the momentum after the real-time data is updated, GCR is the gradient change rate, is the original weight value of gradient descent;
[0085] Regularly calculate the current weight of real-time data and update it to the gradient descent model for use;
[0086] The calculation expression of the updated momentum is: Vnew = β*Vold + (1-β), where Vold is the previous momentum of the real-time data, β is the hyperparameter of the momentum, and β is between [0,1). When β is large, the past gradient information accounts for a larger proportion of the current momentum, and the new gradient information has a relatively small impact on the momentum. This helps to consider past trends more smoothly when updating weights, improve the acceleration of the model in the gradient direction, and slow down the fluctuation of the model in the non-gradient direction. When β is small, the past gradient information accounts for a smaller proportion of the current momentum, and the new gradient information has a relatively large impact on the momentum. This makes the model follow the current gradient more sensitively and is more sensitive to changes in the gradient. It is suitable for situations where the gradient changes greatly, which can improve the stability of the model and avoid overly rapid adjustments. It is suitable for relatively stable gradient situations and helps to accelerate convergence.
[0087] The calculation expression of the gradient change rate is: Where σnew is the gradient of the loss function of the current iteration with respect to the weight, and σold is the gradient of the loss function of the previous iteration with respect to the weight.
[0088] The learning rate calculation expression is: Where xcs is the initial learning rate, xsjz is the learning rate attenuation factor, which is 0.95, and dcs is the number of iterations.
[0089] Real-time monitoring and performance evaluation: Implement real-time monitoring mechanism to monitor the performance of the model. According to the monitoring results, the parameters and configuration of the model are adjusted in real time to improve the accuracy and stability of the system.
[0090] Real-time data stream is passed to the model evaluation module: The feature data stream adjusted by the gradient descent model is passed to the model evaluation module. Make sure that the format of the passed data matches the interface of the model evaluation module for subsequent processing.
[0091] Exception handling and fault tolerance mechanism: Implement an exception handling mechanism to capture and handle possible weight adjustment errors.
[0092] Consider the fault-tolerance characteristics of the gradient descent model to ensure that the system can maintain stable operation in the face of abnormal situations.
[0093] Model evaluation module: After evaluating the performance of the gradient descent model, the matching system automatically finds and corrects the deviation of the gradient descent model in a timely manner based on the evaluation results, improves the performance of the gradient descent model, and sends the performance-adjusted gradient descent model to the calculation module;
[0094] Receive gradient descent model: Receive the model adjusted by the gradient descent algorithm from the previous module (e.g. weight adjustment module). Make sure the received model format matches the interface of the evaluation module.
[0095] Performance evaluation: Use evaluation data sets or real-time data streams to evaluate the performance of the gradient descent model. Consider evaluation indicators such as accuracy, recall, precision, etc. to comprehensively evaluate the matching effect of the model.
[0096] Discover and correct model deviations: Analyze the performance evaluation results to find possible deviations or deficiencies in the model. Based on the problems found, try to adopt appropriate strategies to correct them, such as adjusting parameters, adding features, optimizing algorithms, etc.
[0097] Dynamically adjust model configuration: Dynamically adjust the configuration of the gradient descent model according to the evaluation results, including learning rate, regularization parameters, etc. Implement adaptive adjustments to ensure that the model can adapt to changing data distribution and market demand.
[0098] Real-time monitoring and feedback: Implement a real-time monitoring mechanism to monitor the adjusted gradient descent model. Provide real-time feedback based on the monitoring results to further adjust and optimize the model.
[0099] Calculation module: For each newly posted position, the skill matching score between the candidate and the position is calculated in real time through the gradient descent model, and the matching score is sent to the matching recommendation module;
[0100] Receive newly posted job information and candidate data: Receive each newly posted job information from the matching system, including job requirements, skill requirements, etc. Also receive relevant information about the candidate, including skills, work experience, etc.
[0101] Gradient descent model to calculate matching score: Use the gradient descent model to calculate the skill matching score between the candidate and the position. Input the extracted features into the model and obtain the model output as the matching score;
[0102] The input features of the gradient descent model include the candidate experience vector, the candidate skill vector, and the job skill requirement vector. The function expression of the gradient descent model is:
[0103] pf z =θ0*experience+θ1*candidate+θ2*requirement+b; where, pfz is the matching score, experience is the candidate experience vector, candidate is the candidate skill vector, requirement is the job skill requirement vector, θ0, θ1, θ2 are the weight parameters of the candidate experience vector, candidate skill vector and job skill requirement vector respectively, and θ0, θ1, θ2 are all greater than 0, and b is the bias term;
[0104] The logic of obtaining the candidate experience vector is to extract data related to work experience from the candidate's resume, work experience or other relevant information. By quantifying work experience into numerical values, the experience vector can be constructed using information such as years of work experience and position level.
[0105] The logic for obtaining the candidate's skill vector is to extract skill information from the candidate's resume, skill certificates, training records, etc. Each skill is quantified into a numerical value, and the skill vector can be constructed using information such as professional skills and programming language proficiency.
[0106] The logic of obtaining the job skill requirement vector is as follows: extract skill-related data from job descriptions, skill requirements, job qualifications, etc. Quantify the job requirements for each skill into numerical values and construct the job skill requirement vector.
[0107] Matching score sent to matching recommendation module: Send the calculated skill matching score to the matching recommendation module. Make sure the sent data format complies with the interface of the matching recommendation module for subsequent recommendation processing.
[0108] Specifically: Encode each skill or experience item into a binary vector whose length is equal to the total number of skill or experience items, where 1 indicates that the item is possessed and 0 indicates that the item is not possessed. This method is simple and intuitive, but may result in a high dimension.
[0109] For text data, TF-IDF can be used for feature representation, where the importance of each word is determined by its frequency in the document and its inverse document frequency in the entire corpus.
[0110] Directly convert certain features (such as years of work experience) into numerical values as part of the experience vector.
[0111] For example, suppose candidate A has 3 years of work experience and is proficient in Python and Java, while the position requires at least 2 years of relevant experience, proficiency in Python, and knowledge of Java. The following vector can be constructed:
[0112] Candidate A experience vector: 3;
[0113] Candidate A’s skill vector: 0, 1, 1 (assuming 0 means no skill and 1 means skill);
[0114] Job skill requirement vector: 2,1,1.
[0115] Matching recommendation module: sorts candidates and positions according to their skill matching scores, generates a ranking table, and recommends a corresponding number of candidates in the ranking table to the enterprise;
[0116] Receive skill matching scores: Receive the skill matching scores between each candidate and position from the calculation module. Ensure that the format of the received data matches the interface of the matching recommendation module.
[0117] Generate a ranking table: Sort the received skill matching scores by scores to generate a ranking table. Consider the data structure of the ranking table, which usually includes information such as candidate ID and matching score.
[0118] Select the corresponding number of candidates in positive order: According to the ranking table, select the corresponding number of candidates in positive order of matching scores. The number of recommendations can be adjusted according to the needs of the enterprise or system settings. Suppose there are s candidates in the ranking table, and the enterprise needs to recruit p candidates, then select {1, 2, ..., p} candidates from the ranking table to join the company, and screen out {p, p+1, ..., s} candidates.
[0119] Recommend candidates to enterprises: Recommend the selected candidate information to enterprise users. This may include the candidate's basic information, skill characteristics, matching score, etc.
[0120] Real-time monitoring and feedback: Implement a real-time monitoring mechanism to monitor the performance of the recommended modules and provide real-time feedback for performance adjustment and optimization.
[0121] Exception handling and fault tolerance mechanism: Implement an exception handling mechanism to capture and handle possible recommendation errors. Ensure that the system can maintain stable operation in the face of abnormal situations.
[0122] Security assurance: Implement security measures during the recommendation process to ensure the confidentiality and integrity of candidate information. For information that may involve user privacy, perform appropriate desensitization or encryption.
[0123] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.
[0124] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0125] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. A global human resources service intelligent matching system based on big data, characterized by: It includes data acquisition module, preprocessing module, feature selection module, weight adjustment module, model evaluation module, calculation module and matching recommendation module; Data collection module: Integrate real-time data stream technology to collect data streams from large databases in real time. The data streams include recruitment and candidate data. Preprocessing module: preprocess the data stream using the real-time processing engine; Feature selection module: dynamically selects representative feature data in the data stream according to the changes in real-time data; Weight adjustment module: learns and adapts to changing data streams in real time through the gradient descent model, and introduces a dynamic weight adjustment mechanism in the gradient descent model to dynamically adjust the weight of features according to the importance of real-time data; Model evaluation module: After evaluating the performance of the gradient descent model, it automatically discovers and corrects the deviation of the gradient descent model based on the evaluation results; Calculation module: For each newly posted position, the skill matching score between the candidate and the position is calculated in real time through the gradient descent model; Matching recommendation module: sorts candidates and positions according to their skill matching scores, generates a ranking table, and recommends a corresponding number of candidates in the ranking table to the enterprise; The weight adjustment module introduces a dynamic weight adjustment mechanism into the gradient descent model to dynamically adjust the weight of features according to the importance of real-time data; Obtain the adaptive learning rate, momentum, and gradient change rate of real-time data, and calculate the weight of real-time data. The function expression is: , where is the current weight of real-time data, is the last weight of real-time data, is the learning rate, is the momentum after real-time data update, is the gradient change rate, is the original weight value of gradient descent; Regularly calculate the current weight of real-time data and update it to the gradient descent model for use; The weight adjustment module receives a real-time feature data stream, uses a gradient descent algorithm to perform real-time learning on a gradient descent model, calculates a loss function based on the received feature data and a corresponding target value, and uses a gradient descent algorithm to update the weight of the gradient descent model; The function expression of the gradient descent model is: , where is the model prediction value, is the weight of the feature data, is the characteristic data, is the bias term; The calculation expression of the loss function is: , where is the loss function, is the model prediction value of the i-th sample, is the target value of the i-th sample, is the sample size; The preprocessing module uses a real-time processing engine to clean the data stream, remove missing values and abnormal values, and normalize the data format; According to business needs, the real-time processing engine is used to aggregate and count data in real time, calculate real-time statistical indicators of recruitment needs, including the number of recruitment positions and the frequency of skill requirements, and forward the pre-processed data stream to the feature selection module; The updated momentum calculation expression is: , where is the last momentum of real-time data, is the hyperparameter of momentum, The value is between [0,1); The calculation expression of the gradient change rate is: , where is the gradient of the loss function of the current iteration with respect to the weight, is the gradient of the loss function of the previous iteration with respect to the weight; The learning rate calculation expression is: , where is the initial learning rate, is the learning rate decay factor, is the number of iterations; The computing module receives each newly posted job information from the matching system, including job requirements and skill requirements, and also receives relevant information of the candidate, including skills and work experience; The gradient descent model is used to calculate the skill matching score between the candidate and the position, the extracted features are input into the model, and the model output is obtained as the matching score; The input features of the gradient descent model include the candidate experience vector, the candidate skill vector, and the job skill requirement vector. The function expression of the gradient descent model is: ; In the formula, is the matching score, is the candidate experience vector, is the candidate skill vector, is the job skill requirement vector, , , are the weight parameters of candidate experience vector, candidate skill vector and job skill requirement vector respectively, and ; The logic of obtaining the candidate experience vector is as follows: extracting data related to work experience from the candidate's resume and work experience, and quantifying the work experience into numerical values; The logic for obtaining the candidate's skill vector is as follows: extract skill information from the candidate's resume, skill certificate, and training record, and quantify each skill into a numerical value; The logic of obtaining the job skill requirement vector is as follows: extract the skill-related data from the job description, skill requirements, and qualification information, quantify the job requirements for each skill into numerical values, and construct the job skill requirement vector; The matching recommendation module receives the skill matching score between each candidate and the position from the calculation module; Sort the received skill matching scores by scores to generate a sorting table, the data structure of which includes candidate ID and matching score information; According to the ranking table, select the corresponding number of candidates with matching scores in positive order. Suppose there are s candidates in the ranking table, and the company needs to recruit p candidates, then select {1, 2, ..., p} candidates from the ranking table to join the company, and screen out {p, p+1, ..., s} candidates; Recommend the selected candidate information to corporate users, including the candidate's basic information, skill characteristics, and matching scores. Implement security measures during the recommendation process and desensitize or encrypt information involving user privacy.
2. A global human resource service intelligent matching method based on big data, implemented by the matching system according to claim 1, characterized in that: The matching method comprises the following steps: Integrate real-time data stream technology to collect data streams in real time from large databases. The data streams include recruitment and candidate data to ensure that the system can reflect market changes in a timely manner. Use real-time processing engines to clean and preprocess data to ensure the real-time and accuracy of data. Dynamically select the most representative features in the data stream according to changes in real-time data to adapt to changing market demands. Use the gradient descent model to learn and adapt to changing data streams in real time, and introduce a dynamic weight adjustment mechanism in the gradient descent model to dynamically adjust the weights of features according to the importance of real-time data to improve the adaptability and robustness of the gradient descent model to real-time data. After evaluating the performance of the gradient descent model, the matching system automatically discovers and corrects the deviations of the gradient descent model in a timely manner based on the evaluation results, thereby improving the performance of the gradient descent model. For each newly released position, the gradient descent model calculates the skill matching score between the candidate and the position in real time, and generates a ranking table after sorting the candidates according to the skill matching score between the candidates and the position, and recommends the corresponding number of candidates in the positive order of the ranking table to the enterprise.
