Large model customized training system for recruitment industry

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a large model customization training system for the recruitment industry, which is used for solving the problem that in the prior art, an interview scoring sample is determined through attribute characteristics of job seekers and a corresponding interview attribute mapping table, and then a model is trained by utilizing the interview scoring sample; the problems that job seeker attribute features and position attribute features cannot be combined to form comprehensive features for training a model, accurate matching of positions and job seekers cannot be achieved, and the sense of participation and the satisfaction degree of the job seekers are reduced are solved. The job seeker feature vector and the position feature vector are extracted through the feature engineering module, the job seeker feature vector and the position feature vector are combined to construct the feature matrix for training the model, and the model training module uses the feature matrix to train the logistic regression model for the recruitment industry. Accurate matching of the position and the job seeker can be realized, and the sense of participation and satisfaction of the job seeker are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to a large-model customized training system for the recruitment industry. Background Art

[0002] The current recruitment industry faces the demand for massive data processing and precise matching, and traditional methods are difficult to effectively cope with it. Although there are some AI-based recruitment assistance tools, most of these tools are based on general models and lack a deep understanding and customized processing of data specific to the recruitment industry, resulting in low matching efficiency and insufficient accuracy.

[0003] The patent application with reference publication number CN117131372A discloses a model training method, apparatus, computer equipment, storage medium and program product. The present invention uses a preset interview attribute mapping table including attribute characteristics of job applicants and weights corresponding to each attribute characteristic to determine each interview score sample, and then uses each interview score sample to train an initial interview result classification model to obtain an intermediate interview result classification model. Finally, based on the intermediate interview result classification model and historical interview results corresponding to the historical interview scores of each historical job applicant, a target interview result classification model is determined. The target interview result classification model can be used to evaluate job applicants to obtain evaluation results of the job applicants. The use of this method can improve the accuracy of the evaluation results obtained by evaluating job applicants.

[0004] However, the existing technology determines the interview scoring samples through the attribute characteristics of the job seeker and the corresponding interview attribute mapping table, and then uses the interview scoring samples to train the model. It is unable to combine the attribute characteristics of the job seeker and the attribute characteristics of the position to form a comprehensive feature for training the model, and cannot achieve accurate matching between the position and the job seeker, which reduces the job seeker's sense of participation and satisfaction; at the same time, the existing technology evaluates the job seeker through the model and obtains the evaluation results of the job seeker. It only evaluates the job seeker based on the interview situation and cannot recommend positions with a high matching probability to the job seeker, resulting in the job seeker being unable to find a satisfactory job and the recruitment team being unable to recruit suitable personnel.

[0005] To this end, we propose a large-scale customized training system for the recruitment industry to address the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide a large-model customized training system for the recruitment industry, which solves the problem that the existing technology determines interview scoring samples through the attribute characteristics of job seekers and the corresponding interview attribute mapping table, and then uses the interview scoring samples to train the model. It is unable to combine the attribute characteristics of job seekers and the attribute characteristics of positions to form comprehensive characteristics for training the model, and cannot achieve accurate matching of positions and job seekers, which reduces the job seekers' sense of participation and satisfaction; at the same time, the existing technology evaluates job seekers through models and obtains the evaluation results of job seekers. It only evaluates job seekers based on the interview situation and cannot recommend positions with high matching probability to job seekers, resulting in the problem that job seekers cannot find satisfactory jobs and the recruitment team cannot recruit suitable personnel.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A large-scale customized model training system for the recruitment industry, including a cloud management platform, data collection and preprocessing modules, feature engineering modules, model training modules, job recommendation modules, interactive feedback modules, and report generation modules;

[0009] Data collection and preprocessing module, which is used to integrate text data from recruitment platforms, social media, and internal databases, and clean the text data before storing it;

[0010] Feature engineering module, which is used to extract and process features from text data and convert text data into numerical vectors;

[0011] Model training module, which uses the logistic regression algorithm to train a logistic regression model for the recruitment industry;

[0012] The job recommendation module uses the trained model to calculate the matching probability between job seekers and positions, and recommends suitable positions to job seekers based on the calculated matching probability;

[0013] Interactive feedback module, used to interact with job seekers in real time, collect feedback from job seekers and recruitment teams, and monitor user interaction data;

[0014] Report generation module, used to generate analytical reports of recruitment activities on a regular basis.

[0015] As a preferred embodiment of the present invention, the specific steps of data integration and preprocessing by the data collection and preprocessing module are as follows:

[0016] Automatically scrape text data from recruitment platforms, social media, and industry websites. This text data includes job data and job seeker data. Job data includes job title, job description, skill requirements, industry, and company information. Job seeker data includes name, contact information, education background, work experience, and skills. The automatically scraped text data is integrated and regularly updated using APIs.

[0017] Clean the integrated text data, remove duplicate records, process missing values and unify the data format;

[0018] The cleaned text data is stored in a structured database or an unstructured database. Structured data is stored in a structured database, and document data is stored in an unstructured database.

[0019] As a preferred embodiment of the present invention, the specific process of extracting and processing features by the feature engineering module is as follows:

[0020] Use web page parsing tools to extract text data content. The parsing library automatically processes tags to obtain clean text data. Regular expressions are used to remove special symbols in the text, convert all text to lowercase letters, use a stop word list to filter out common words that do not affect the meaning of the text analysis, and use Chinese and English word segmentation libraries to segment the text into multiple words or phrases.

[0021] The preprocessed text is converted into a word frequency representation. Each document is represented as a fixed-length vector whose elements represent the number of occurrences of each word in the vocabulary. The weight of the word in the text is calculated using TF-IDF. The expression formula is:

[0022] Where TF-IDF is term frequency-inverse document frequency, TF(t,d) is the frequency of word t in document d, N is the total number of documents, and DF(t) is the number of documents containing word t.

[0023] Extract combinations of n adjacent words to capture contextual information. Use the GloVe model to map words into low-dimensional vectors to capture the semantic relationship between words. When training the model, use the co-occurrence information of context words to generate vector representations that can reflect the meaning of similar words. Different features are horizontally merged and weighted according to the importance of the features.

[0024] As a preferred embodiment of the present invention, for categorical variables, label encoding is used to convert each category into an integer value, and each category is converted into a binary vector so that there is no order relationship between categories. The numerical features are processed so that their mean is 0 and their variance is 1. The features are scaled to a range of 0 to 1 to handle features with inconsistent dimensions. All features are given the same weight through linear transformation.

[0025] Extract the category features of the job title, use one-hot encoding to extract the TF-IDF features of the job description, extract the n-gram features of the related skills, merge all the extracted features to form the job feature vector J, which is expressed as follows: J = [j1, j2, ..., j m ], where each j m A numerical representation of a feature;

[0026] Extract the job applicant's skill features, use TF-IDF or word vectors to generate skill-related features, extract features related to education background and work experience, merge all features, and form the job applicant feature vector C, which is expressed as follows: C = [c1, c2, ..., c n ], where each c n A numerical representation of a feature;

[0027] The feature matrix X is constructed by combining the job position and the job applicant features, which can be expressed as: X = [C, J].

[0028] As a preferred embodiment of the present invention, the specific process of the model training module training the logistic regression model is as follows:

[0029] The constructed feature matrix X is used as the input of the logistic regression model, and the label vector y is used as the output of the logistic regression model. The label vector y represents the matching status between the job applicant and the position. The output of the label vector y is 0 or 1, 0 indicates that the job applicant and the position do not match, and 1 indicates that the job applicant and the position match. The label vector y is used as the prediction target, and the sum of the prediction errors of all training data is minimized as the training target. The logistic regression model is trained until the sum of the prediction errors reaches convergence and the training is stopped. The logistic regression model that predicts the label vector y is obtained.

[0030] As a preferred embodiment of the present invention, the formula of the logistic regression model is expressed as:

[0031]

[0032] P(y=1|X) represents the probability of a job seeker matching a job position given the input feature X, w represents the weight vector, b is the bias term, and w T represents the transpose of the weight vector;

[0033] The logistic regression model is evaluated and optimized by the cross entropy loss function. The cross entropy loss function in the logistic regression model is defined as:

[0034]

[0035] L(w,b) is the loss value, which indicates the gap between the prediction of the logistic regression model and the true label, N is the number of samples, y m is the true label of the mth sample, p m is the predicted probability of the logistic regression model for the mth sample.

[0036] As a preferred embodiment of the present invention, the specific process of the job recommendation module calculating the matching probability between job seekers and jobs and recommending jobs is as follows:

[0037] Collect the characteristic data of new job seekers and form the characteristic vector X 新 , the matching probability between each position and the new job seeker is calculated by the trained logistic regression model. The calculation formula is as follows: P(y=1|X 新 )=σ(w T X 新 +b), set the recommendation threshold to 0.7, and compare the calculated matching probability of each position with the set recommendation threshold:

[0038] If P(y=1|X 新 )>0.7, then recommend the position to the new job seeker;

[0039] If P(y=1|X 新 )≤0.7, the position will not be recommended to new job seekers;

[0040] Among the positions recommended to new job seekers, the positions are ranked according to the calculated position matching probabilities, and the top five positions are recommended to the new job seekers from high to low.

[0041] As a preferred embodiment of the present invention, the specific process of the interaction and feedback processing performed by the interactive feedback module is as follows:

[0042] Build a platform that integrates web pages and mobile applications to provide an intuitive interactive interface for job seekers and recruitment teams. The platform provides job seekers with personalized job recommendations, real-time chat and online questionnaire filling functions, while recruitment teams can view job seekers' feedback, historical chat records and interaction statistics through the platform.

[0043] As a preferred embodiment of the present invention, the specific process of the report generation module generating the recruitment activity analysis report is as follows:

[0044] Collect recruitment data from recruitment systems, recruitment websites, and human resource management systems. This data includes job postings, number of applicants, interview feedback, and hiring status. Clean the collected recruitment data to remove duplicate and erroneous records, fill in missing values, and integrate data scattered across different systems into a unified data warehouse.

[0045] Conduct in-depth analysis of cleaned and integrated recruitment data using descriptive statistics, trend analysis, and comparative analysis methods;

[0046] Utilize report generation tools or custom scripts to automate the generation of analysis reports and automatically share the generated reports with relevant personnel via email or internal systems.

[0047] As a preferred embodiment of the present invention, the specific process of using descriptive statistics, trend analysis and comparative analysis methods to conduct in-depth analysis of the cleaned and integrated recruitment data is as follows:

[0048] When using descriptive statistical methods, we analyze recruitment data from the following aspects:

[0049] Calculate basic statistics, including the total number of job postings, the number of applicants for each position, the ratio of qualified to unqualified interviewees, the number of hires, and the hire rate;

[0050] Use data visualization to show data distribution, using histograms to show the number of applicants and the number of job postings, pie charts to show the proportion of interview feedback, and scatter plots to identify outliers in the number of applicants;

[0051] Define key performance indicators, including recruitment cycle time and job application and acceptance rates;

[0052] When applying trend analysis methods, we analyze the changes in job postings and the number of applicants through time series analysis of recruitment data, identify seasonal fluctuations, and use linear regression models to predict future trends while calculating moving averages to smooth out fluctuations.

[0053] When using comparative analysis methods, focus on recruitment data, such as the differences in acceptance rates and the number of applicants for different positions, changes in recruitment efficiency between different quarters and years, and the comparison of recruitment costs and job seeker satisfaction.

[0054] Compared with the prior art, the advantages of the present invention are:

[0055] (1) In the present invention, a feature engineering module is used to extract the job applicant feature vector and the position feature vector, and the job applicant feature vector and the position feature vector are combined to construct a feature matrix for model training. The model training module uses the feature matrix to train a logistic regression model for the recruitment industry, which can achieve accurate matching between positions and job applicants and improve the job applicants' sense of participation and satisfaction.

[0056] (2) In the present invention, the position recommendation module uses a trained model to calculate the matching probability between job seekers and positions, and recommends positions with high matching probabilities to job seekers based on the calculated position matching probabilities. This ensures that the positions are consistent with the background and skills of the job seekers, enhances the scientific nature of recruitment and the satisfaction of job seekers, and enables job seekers to find satisfactory jobs and the recruitment team to recruit suitable personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is a system block diagram of the present invention;

[0058] Figure 2 This is a system block diagram of Embodiment 1 of the present invention;

[0059] Figure 3 It is a logical flow diagram of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making creative work shall fall within the scope of protection of the present invention.

[0061] Example 1: Figure 1 and Figure 2 As shown, the large-scale customized training system for the recruitment industry proposed in the present invention includes a cloud management platform, a data collection and preprocessing module, a feature engineering module, a model training module, a position recommendation module, an interactive feedback module, and a report generation module;

[0062] The cloud management platform has a two-way communication connection with the report generation module and the data collection and preprocessing module. The cloud management platform has a one-way communication connection with the job recommendation module. The data collection and preprocessing module has a one-way communication connection with the feature engineering module. The feature engineering module has a one-way communication connection with the model training module. The job recommendation module has a one-way communication connection with the interactive feedback module and the model training module. The interactive feedback module has a one-way communication connection with the model training module.

[0063] Data collection and preprocessing module, which is used to integrate text data from recruitment platforms, social media, and internal databases, and clean the text data before storing it;

[0064] The specific steps of data collection and preprocessing module for data integration and preprocessing are as follows:

[0065] Automatically scrape text data from recruitment platforms, social media, and industry websites. This text data includes job data and job seeker data. Job data includes job title, job description, skill requirements, industry, and company information. Job seeker data includes name, contact information, education background, work experience, and skills. The automatically scraped text data is integrated and regularly updated using APIs.

[0066] Clean the integrated text data, remove duplicate records, process missing values and unify the data format;

[0067] Store the cleaned text data into a structured database or an unstructured database, using the structured database to store structured data and the unstructured database to store document-type data;

[0068] The data collection and preprocessing module integrates text data from recruitment platforms, social media, and internal databases, and cleans and stores it. It has the following advantages: automated capture improves data collection efficiency, ensures the breadth and timeliness of information, and regular updates enhance data accuracy and reliability. The cleaning process removes duplicate records, handles missing values, and unifies the format to ensure data quality. Data is stored in structured and unstructured databases for efficient storage and retrieval to meet different processing requirements. This module significantly optimizes the data management process and provides a solid foundation for recruitment decisions.

[0069] Feature engineering module, which is used to extract and process features from text data and convert text data into numerical vectors;

[0070] The specific process of feature engineering module extracting and processing features is as follows:

[0071] Use web page parsing tools to extract text data content. The parsing library automatically processes tags to obtain clean text data. Regular expressions are used to remove special symbols in the text, convert all text to lowercase letters, use a stop word list to filter out common words that do not affect the meaning of the text analysis, and use Chinese and English word segmentation libraries to segment the text into multiple words or phrases.

[0072] The preprocessed text is converted into a word frequency representation. Each document is represented as a fixed-length vector whose elements represent the number of occurrences of each word in the vocabulary. The weight of the word in the text is calculated using TF-IDF. The expression formula is:

[0073] Where TF-IDF is term frequency-inverse document frequency, TF(t,d) is the frequency of word t in document d, N is the total number of documents, and DF(t) is the number of documents containing word t.

[0074] Extract combinations of n adjacent words to capture contextual information. Use the GloVe model to map words into low-dimensional vectors to capture the semantic relationships between words. When training the model, use the co-occurrence information of contextual words to generate vector representations that reflect the meaning of similar words. Different features are horizontally merged and weighted according to their importance.

[0075] For categorical variables, label encoding is used to convert each category into an integer value. Each category is converted into a binary vector so that there is no order relationship between categories. Numerical features are processed so that their mean is 0 and their variance is 1. Features are scaled to the range of 0 to 1 to handle features with inconsistent dimensions. All features are given the same weight through linear transformation.

[0076] Extract the category features of the job title, use one-hot encoding to extract the TF-IDF features of the job description, extract the n-gram features of the related skills, merge all the extracted features to form the job feature vector J, which is expressed as follows: J = [j1, j2, ..., j m ], where each j m A numerical representation of a feature;

[0077] Extract the job applicant's skill features, use TF-IDF or word vectors to generate skill-related features, extract features related to education background and work experience, merge all features, and form the job applicant feature vector C, which is expressed as follows: C = [c1, c2, ..., c n ], where each c n A numerical representation of a feature;

[0078] Combine the job position and applicant features to construct a feature matrix X, expressed as: X = [C, J];

[0079] The feature engineering module extracts and processes text data, converting it into numerical vectors. It offers the following advantages: First, it uses web parsing tools and regular expressions to clean the text and ensure data neatness. It then extracts key information through stopword filtering and word segmentation, combining the TF-IDF and GloVe models to quantify vocabulary importance and semantic relationships. The module uses label encoding and one-hot encoding to process categorical variables and standardizes numerical features, thereby improving model stability and convergence speed. Finally, it integrates the feature vectors of positions and applicants to construct a feature matrix, providing structured input for subsequent model training and improving matching accuracy and efficiency.

[0080] Model training module, which uses the logistic regression algorithm to train a logistic regression model for the recruitment industry;

[0081] The specific process of training the logistic regression model in the model training module is as follows:

[0082] The constructed feature matrix X is used as the input of the logistic regression model, and the label vector y is used as the output of the logistic regression model. The label vector y represents the matching status between the job applicant and the position. The output of the label vector y is 0 or 1, 0 indicates that the job applicant and the position do not match, and 1 indicates that the job applicant and the position match. The label vector y is used as the prediction target, and the sum of the prediction errors of all training data is minimized as the training target. The logistic regression model is trained until the sum of the prediction errors reaches convergence, and the training is stopped. The logistic regression model that predicts the label vector y is obtained.

[0083] The formula of the logistic regression model is expressed as:

[0084]

[0085] P(y=1|X) represents the probability of a job seeker matching a job position given the input feature X, w represents the weight vector, b is the bias term, and w T represents the transpose of the weight vector;

[0086] The logistic regression model is evaluated and optimized by the cross entropy loss function. The cross entropy loss function in the logistic regression model is defined as:

[0087]

[0088] L(w,b) is the loss value, which indicates the gap between the prediction of the logistic regression model and the true label, N is the number of samples, y m is the true label of the mth sample, p m is the predicted probability of the logistic regression model for the mth sample;

[0089] The model training module uses a logistic regression algorithm and focuses on matching problems in the recruitment industry. It has the following advantages: it takes a feature matrix as input and a label vector as output to optimize matching relationships. The simplicity and interpretability of logistic regression make it suitable for recruitment scenarios, quickly identifying matching probabilities and improving model accuracy and stability through the cross-entropy loss function. The module effectively learns the matching relationship between job seekers and positions, achieves accurate recommendations, and has good generalization capabilities to adapt to the changing needs of the recruitment market.

[0090] Embodiment 2: The technical solution of this embodiment of the present invention differs from that of Embodiment 1 in that:

[0091] like Figure 1 and Figure 3 As shown in the figure, the job recommendation module uses the trained model to calculate the matching probability between job seekers and jobs, and recommends suitable jobs to job seekers based on the calculated job matching probability;

[0092] The specific process of the job recommendation module calculating the probability of matching job seekers with jobs and recommending jobs is as follows:

[0093] Collect the characteristic data of new job seekers and form the characteristic vector X 新 , the matching probability between each position and the new job seeker is calculated by the trained logistic regression model. The calculation formula is as follows: P(y=1|X 新 )=σ(w T X 新 +b), set the recommendation threshold to 0.7, and compare the calculated matching probability of each position with the set recommendation threshold:

[0094] If P(y=1|X 新 )>0.7, then recommend the position to the new job seeker;

[0095] If P(y=1|X 新 )≤0.7, the position will not be recommended to new job seekers;

[0096] Among the positions recommended to new job seekers, the positions are ranked according to the calculated position matching probabilities, and the top five positions are recommended to the new job seekers from high to low;

[0097] The job recommendation module uses a trained logistic regression model to calculate the matching probability between job seekers and jobs in order to recommend suitable jobs. It collects the job seeker's characteristic data to form a feature vector, calculates the matching probability, sets a threshold of 0.7, and only recommends jobs above this threshold to ensure the accuracy and relevance of the recommendations. The recommended jobs are sorted by matching probability and the top five optimal choices are displayed, thereby improving recommendation efficiency, enhancing job seeker satisfaction and recruitment success rate.

[0098] Interactive feedback module, used to interact with job seekers in real time, collect feedback from job seekers and recruitment teams, and monitor user interaction data;

[0099] The specific process of interaction and feedback processing in the interactive feedback module is as follows:

[0100] Build a platform that integrates web pages and mobile applications to provide an intuitive interactive interface for job seekers and recruitment teams. The platform provides personalized job recommendations, real-time chat, and online questionnaire completion for job seekers. Recruitment teams can use the platform to view job seekers' feedback, historical chat records, and interaction statistics. The integrated chatbot can understand and automatically respond to job seekers' common questions. The platform also supports WebSocket-based instant chat, allowing recruiters to quickly respond to job seekers' inquiries.

[0101] Design a comprehensive feedback mechanism so that job seekers can quickly provide feedback on their initial impressions after receiving a recommended position by easily selecting "satisfied," "average," or "unsatisfied." After completing their application or interview, the system automatically sends a detailed Likert-type feedback questionnaire to collect their opinions on the recommended position and the recruitment process. This feedback data will be stored in a relational or non-relational database, and user behavior analytics tools will be used to track page visits, clickthroughs, and feedback submissions. In addition, data visualization tools will be used to create real-time analytics dashboards displaying key indicators such as the number of online job seekers and the status of real-time chats.

[0102] The interactive feedback module enables real-time interaction between job seekers and the recruitment team through web pages and mobile applications, improving communication efficiency. Job seekers can obtain personalized job recommendations, participate in chats and fill out questionnaires to enhance satisfaction. The recruitment team can instantly view feedback and adjust strategies. The integrated chatbot automatically replies to common questions, and WebSocket ensures a quick response. After the job seeker provides feedback, the system automatically sends a Likert scale questionnaire to gain a deeper understanding of needs. Through data storage and visual analysis, the module optimizes the recruitment process and improves overall results.

[0103] Report generation module, used to generate analytical reports on recruitment activities on a regular basis;

[0104] The specific process of generating a recruitment activity analysis report by the report generation module is as follows:

[0105] Collect recruitment data from recruitment systems, recruitment websites, and human resource management systems. This data includes job postings, number of applicants, interview feedback, and hiring status. Clean the collected recruitment data to remove duplicate and erroneous records, fill in missing values, and integrate data scattered across different systems into a unified data warehouse.

[0106] Conduct in-depth analysis of cleaned and integrated recruitment data using descriptive statistics, trend analysis, and comparative analysis methods;

[0107] Use report generation tools or custom scripts to automatically generate analysis reports and automatically share the generated reports with relevant personnel via email or internal systems;

[0108] The specific process of using descriptive statistics, trend analysis, and comparative analysis methods to conduct in-depth analysis of the cleaned and integrated recruitment data is as follows:

[0109] When using descriptive statistical methods, we analyze recruitment data from the following aspects:

[0110] Calculate basic statistics, including the total number of job postings, the number of applicants for each position, the ratio of qualified to unqualified interviewees, the number of hires, and the hire rate;

[0111] Use data visualization to show data distribution, using histograms to show the number of applicants and the number of job postings, pie charts to show the proportion of interview feedback, and scatter plots to identify outliers in the number of applicants;

[0112] Define key performance indicators, including recruitment cycle time and job application and acceptance rates;

[0113] When applying trend analysis methods, we analyze the changes in job postings and the number of applicants through time series analysis of recruitment data, identify seasonal fluctuations, and use linear regression models to predict future trends while calculating moving averages to smooth out fluctuations.

[0114] When using comparative analysis, focus on recruitment data, focusing on differences in acceptance rates and the number of applicants for different positions, changes in recruitment efficiency between quarters and years, and comparisons between recruitment costs and job seeker satisfaction.

[0115] The report generation module regularly generates recruitment activity analysis reports, which has significant advantages. It collects data on job postings, number of applicants, interview feedback and hiring status from various recruitment systems to ensure data quality. It deeply mines data through descriptive statistics, trend analysis and comparative analysis, calculates basic statistics, identifies changing trends and compares the hiring rates and recruitment efficiency of different positions. Finally, the module automatically generates and shares analysis reports, realizes efficient information transmission, improves decision-making efficiency, comprehensively evaluates recruitment results and optimizes the recruitment process.

[0116] The working principle of the present invention: When in use, the present invention extracts the job applicant feature vector and the position feature vector through the feature engineering module, combines the job applicant feature vector and the position feature vector to construct a feature matrix for training the model, and uses the feature matrix to train the logistic regression model for the recruitment industry through the model training module, which can achieve accurate matching of positions and job seekers, improve the job seekers' sense of participation and satisfaction, and use the trained model through the position recommendation module to calculate the matching probability between job seekers and positions, and recommend positions with high matching probability to job seekers based on the calculated position matching probability, which can ensure that the position is consistent with the job seeker's background and skills, enhance the scientific nature of recruitment and job seeker satisfaction, and enable job seekers to find satisfactory jobs and recruitment teams to recruit suitable personnel.

[0117] The above are only preferred specific implementation methods of the present invention; however, the protection scope of the present invention is not limited thereto; any technician familiar with the technical field within the technical scope disclosed by the present invention; any equivalent replacement or change based on the technical solution and improved conception of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A large-scale customized training system for the recruitment industry, characterized by: It includes cloud management platform, data collection and preprocessing module, feature engineering module, model training module, job recommendation module, interactive feedback module and report generation module; Data collection and preprocessing module, which is used to integrate text data from recruitment platforms, social media, and internal databases, and clean the text data before storing it; Feature engineering module, which is used to extract and process features from text data and convert text data into numerical vectors; Model training module, which uses the logistic regression algorithm to train a logistic regression model for the recruitment industry; The job recommendation module uses the trained model to calculate the matching probability between job seekers and positions, and recommends suitable positions to job seekers based on the calculated matching probability; Interactive feedback module, used to interact with job seekers in real time, collect feedback from job seekers and recruitment teams, and monitor user interaction data; Report generation module, used to generate analytical reports of recruitment activities on a regular basis.

2. The large-scale customized training system for the recruitment industry according to claim 1 is characterized in that: The specific steps of data integration and preprocessing in the data collection and preprocessing module are as follows: Automatically scrape text data from recruitment platforms, social media, and industry websites. This text data includes job data and job seeker data. Job data includes job title, job description, skill requirements, industry, and company information. Job seeker data includes name, contact information, education background, work experience, and skills. The automatically scraped text data is integrated and regularly updated using APIs. Clean the integrated text data, remove duplicate records, process missing values and unify the data format; The cleaned text data is stored in a structured database or an unstructured database. Structured data is stored in a structured database, and document data is stored in an unstructured database.

3. The large-scale customized training system for the recruitment industry according to claim 2 is characterized in that: The specific process of extracting and processing features by the feature engineering module is as follows: Use web page parsing tools to extract text data content. The parsing library automatically processes tags to obtain clean text data. Regular expressions are used to remove special symbols in the text, convert all text to lowercase letters, use a stop word list to filter out common words that do not affect the meaning of the text analysis, and use Chinese and English word segmentation libraries to segment the text into multiple words or phrases. The preprocessed text is converted into a word frequency representation. Each document is represented as a fixed-length vector whose elements represent the number of occurrences of each word in the vocabulary. The weight of the word in the text is calculated using TF-IDF. The expression formula is: Where TF-IDF is term frequency-inverse document frequency, TF(t,d) is the frequency of word t in document d, N is the total number of documents, and DF(t) is the number of documents containing word t. Extract combinations of n adjacent words to capture contextual information. Use the GloVe model to map words into low-dimensional vectors to capture the semantic relationship between words. When training the model, use the co-occurrence information of context words to generate vector representations that can reflect the meaning of similar words. Different features are horizontally merged and weighted according to the importance of the features.

4. The large-scale customized training system for the recruitment industry according to claim 3 is characterized in that: For categorical variables, label encoding is used to convert each category into an integer value. Each category is converted into a binary vector so that there is no order relationship between categories. Numerical features are processed so that their mean is 0 and their variance is 1. Features are scaled to the range of 0 to 1 to handle features with inconsistent dimensions. All features are given the same weight through linear transformation. Extract the category features of the job title, use one-hot encoding to extract the TF-IDF features of the job description, extract the n-gram features of the related skills, merge all the extracted features to form the job feature vector J, which is expressed as follows: J = [j1, j2, ..., j m ], where each j m A numerical representation of a feature; Extract the job applicant's skill features, use TF-IDF or word vectors to generate skill-related features, extract features related to education background and work experience, merge all features, and form the job applicant feature vector C, which is expressed as follows: C = [c1, c2, ..., c n ], where each c n A numerical representation of a feature; The feature matrix X is constructed by combining the job position and the job applicant features, which can be expressed as: X = [C, J].

5. The large-scale customized training system for the recruitment industry according to claim 4 is characterized in that: The specific process of training the logistic regression model by the model training module is as follows: The constructed feature matrix X is used as the input of the logistic regression model, and the label vector y is used as the output of the logistic regression model. The label vector y represents the matching status between the job applicant and the position. The output of the label vector y is 0 or 1, 0 indicates that the job applicant and the position do not match, and 1 indicates that the job applicant and the position match. The label vector y is used as the prediction target, and the training goal is to minimize the sum of the prediction errors of all training data. The logistic regression model is trained until the sum of the prediction errors reaches convergence and the training is stopped. The logistic regression model that predicts the label vector y is obtained.

6. The large-scale customized training system for the recruitment industry according to claim 5 is characterized in that: The formula of the logistic regression model is expressed as: P(y=1|X) represents the probability of a job seeker matching a job position given the input feature X, w represents the weight vector, b is the bias term, and w T represents the transpose of the weight vector; The logistic regression model is evaluated and optimized by the cross entropy loss function. The cross entropy loss function in the logistic regression model is defined as: L(w,b) is the loss value, which indicates the gap between the prediction of the logistic regression model and the true label, N is the number of samples, y m is the true label of the mth sample, p m is the predicted probability of the logistic regression model for the mth sample.

7. The large-scale customized training system for the recruitment industry according to claim 6 is characterized in that: The specific process of the job recommendation module calculating the matching probability between job seekers and jobs and recommending jobs is as follows: Collect the characteristic data of new job seekers and form the characteristic vector X 新 , the matching probability between each position and the new job seeker is calculated by the trained logistic regression model. The calculation formula is as follows: P(y=1|X 新 )=σ(w T X 新 +b), set the recommendation threshold to 0.7, and compare the calculated matching probability of each position with the set recommendation threshold: If P(y=1|X 新 )>0.7, then recommend the position to the new job seeker; If P(y=1|X 新 )≤0.7, the position will not be recommended to new job seekers; Among the positions recommended to new job seekers, the positions are ranked according to the calculated position matching probabilities, and the top five positions are recommended to the new job seekers from high to low.

8. The large-scale customized training system for the recruitment industry according to claim 1 is characterized in that: The specific process of interaction and feedback processing by the interactive feedback module is as follows: Build a platform that integrates web pages and mobile applications to provide an intuitive interactive interface for job seekers and recruitment teams. The platform provides job seekers with personalized job recommendations, real-time chat and online questionnaire filling functions, while recruitment teams can view job seekers' feedback, historical chat records and interaction statistics through the platform.

9. The large-scale customized training system for the recruitment industry according to claim 1 is characterized in that: The specific process of the report generation module generating a recruitment activity analysis report is as follows: Collect recruitment data from recruitment systems, recruitment websites, and human resource management systems. This data includes job postings, number of applicants, interview feedback, and hiring status. Clean the collected recruitment data to remove duplicate and erroneous records, fill in missing values, and integrate data scattered across different systems into a unified data warehouse. Conduct in-depth analysis of cleaned and integrated recruitment data using descriptive statistics, trend analysis, and comparative analysis methods; Utilize report generation tools or custom scripts to automate the generation of analysis reports and automatically share the generated reports with relevant personnel via email or internal systems.

10. The large-scale customized training system for the recruitment industry according to claim 9, characterized in that: The specific process of using descriptive statistics, trend analysis, and comparative analysis methods to conduct in-depth analysis of the cleaned and integrated recruitment data is as follows: When using descriptive statistical methods, we analyze recruitment data from the following aspects: Calculate basic statistics, including the total number of job postings, the number of applicants for each position, the ratio of qualified to unqualified interviewees, the number of hires, and the hire rate; Use data visualization to show data distribution, using histograms to show the number of applicants and the number of job postings, pie charts to show the proportion of interview feedback, and scatter plots to identify outliers in the number of applicants; Define key performance indicators, including recruitment cycle time and job application and acceptance rates; When applying trend analysis methods, we analyze the changes in job postings and the number of applicants through time series analysis of recruitment data, identify seasonal fluctuations, and use linear regression models to predict future trends while calculating moving averages to smooth out fluctuations. When using comparative analysis methods, focus on recruitment data, such as the differences in acceptance rates and the number of applicants for different positions, changes in recruitment efficiency between different quarters and years, and the comparison of recruitment costs and job seeker satisfaction.

Images