Intelligent recruitment advertisement putting decision-making system based on big data model

Abstract

The invention relates to the technical field of recruitment advertisement putting, in particular to an intelligent recruitment advertisement putting decision-making system based on a big data model, which comprises an advertisement putting management center, a data acquisition module, a recruitment matching module, a putting precision module, a post division module, a putting effect module and a response management module. According to the method, based on multi-dimensional feature matching of cosine similarity, the recommendation list is dynamically adjusted in combination with job seeker behaviors, it is ensured that high-matching-degree job seekers are preferentially reached, through secondary verification and interaction score weighting, high-intention crowds are further screened out, invalid advertisement putting is reduced, the enterprise recruitment cost is reduced, and the enterprise recruitment efficiency is improved. And the post priority is converted from qualitative description to quantitative score, so that an objective basis is provided for advertisement budget distribution and putting strategy adjustment, and meanwhile, the pushing time period, channel and content are dynamically optimized through comparison between prediction and actual effects, so that a traditional experience-driven recruitment mode is promoted to be transformed into a data-driven intelligent decision-making mode.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of recruitment advertisement delivery, and in particular to an intelligent recruitment advertisement delivery decision system based on a big data model. BACKGROUND

[0002] In the digital era, the enterprise recruitment mode has shifted from traditional offline channels to online and offline integration. Online recruitment has become the core approach for enterprises to obtain talents due to its wide coverage and fast information transmission. The scientificity of the online recruitment advertisement delivery strategy directly affects the recruitment efficiency.

[0003] With the intensification of competition in the employment market and the rapid penetration of digital technology, enterprises have an increasingly urgent demand for efficient recruitment channels. However, there are still many technical bottlenecks and industry pain points in the current recruitment advertisement delivery field. Specifically, the current recruitment advertisement delivery method lacks precision, cannot fully utilize big data resources, and is prone to inaccurate delivery time, resulting in low recruitment efficiency and wasted recruitment costs for enterprises.

[0004] Therefore, in view of the above technical defects, a solution is proposed. SUMMARY

[0005] The purpose of the present application is to provide an intelligent recruitment advertisement delivery decision system based on a big data model to solve the above technical defects. Through data collection, intelligent matching, secondary verification, and dynamic optimization, the appropriate position can accurately reach the appropriate person.

[0006] The purpose of the present application can be achieved by the following technical solution: an intelligent recruitment advertisement delivery decision system based on a big data model, comprising an advertisement delivery management center, a data collection module, a recruitment matching module, a delivery precision module, a position division module, a delivery effect module, and a response management module.

[0007] The data collection module is used to collect job seeker data, recruitment position data, and market data. Based on the collected job seeker data, recruitment position data, and market data, an original data set is constructed, and the original data set is sent to the advertisement delivery management center for storage.

[0008] The recruitment matching module is used to analyze the feature label similarity of the original data set to obtain a matching score. According to the matching score, a post-high matching job seeker list is formed.

[0009] The delivery precision module is used to perform secondary verification and optimization adjustment analysis on the post-high matching job seeker, and to perform discriminant processing on the final matching score to obtain a post-high priority job seeker list.

[0010] The post division module is used for advertisement placement and post division matching analysis on the post-high priority job seeker list, discriminant processing on the obtained post priority score, and obtaining of emergency posts and ordinary posts.

[0011] The placement effect module is used for initial intention and delivery intention joint analysis on the collected predicted placement effect data and actual placement effect data of the placed advertisements, and obtaining of initial deviation signals or result deviation signals.

[0012] Preferably, the analysis process of the recruitment matching module is as follows:

[0013] S1: cleaning, integrating and standardizing the original data set to obtain an initial data set;

[0014] S2: feature extraction on the initial data set to obtain a job seeker feature label set and a recruitment post feature label set;

[0015] S3: generating an N-dimensional feature vector Vp for each job seeker, the vector dimension corresponding to the feature labels in the output job seeker feature label set;

[0016] generating an M-dimensional feature vector Vo for each recruitment post, the vector dimension being consistent with the job seeker feature vector;

[0017] standardizing each dimension in the vector, and converting non-numeric features into numeric form through word embedding in the prior art.

[0018] Preferably, S4: calculating the cosine similarity of the job seeker vector Vp and the post vector Vo, and obtaining a matching score set in the corresponding preset cosine similarity interval based on the cosine similarity;

[0019] For each post, sorting according to the matching score, screening out TopG job seekers, G being greater than 30, and forming a post-high matching job seeker list.

[0020] Preferably, the analysis process of the placement precision module is as follows:

[0021] Based on the hard requirement label of the recruitment post, the TopG job seekers are subjected to secondary verification, and the job seekers who do not meet the hard requirement label are excluded from the post-high matching job seekers, obtaining candidate job seekers;

[0022] Based on the interactive behavior label of the candidate job seekers and the post, different scores are assigned to the parameters in the interactive behavior label: 1 point for click behavior, 3 points for delivery behavior, and 5 points for interview behavior;

[0023] Based on the interactive behavior label, the interactive scores of each candidate job seeker are obtained.

[0024] Preferably, the sum of the matching score of the candidate job seeker and the interaction score multiplied by the corresponding weight coefficient is set as the final matching score, and the final matching score is judged, the candidate job seeker corresponding to the final matching score greater than the preset final matching score threshold is retained, and is set as a high-priority job seeker, and a post-high-priority job seeker list is constructed.

[0025] Preferably, the analysis process of the post division module is as follows:

[0026] The basic state data of the advertisement to be put on the post is obtained based on the post-high-priority job seeker list, and the basic state data includes post urgency, importance and recruitment difficulty;

[0027] The preset weight coefficient corresponding to the post urgency, importance and recruitment difficulty is allocated, the sum of the post urgency, importance and recruitment difficulty multiplied by the corresponding preset weight coefficient is set as the post priority score, and the post priority score is judged to obtain an urgent post and a common post.

[0028] Preferably, the post urgency represents the sum of the vacancy duration of the post and the influence degree of the vacancy multiplied by the corresponding preset weight coefficient, and the influence degree of the vacancy represents the number of personnel who stop working; the importance represents the duration between the response replacement of the post and the completion time of the replacement; and the recruitment difficulty represents the sum of the market talent supply-demand ratio (the ratio of the number of job seekers to the number of enterprise needs) and the historical recruitment cycle (the average duration of the enterprise's past recruitment of the same post) multiplied by the corresponding preset weight coefficient.

[0029] Preferably, the analysis process of the delivery effect module is as follows:

[0030] The predicted delivery effect data and the actual delivery effect data of the delivered advertisement are obtained, the predicted delivery effect data includes the predicted click rate and the predicted delivery rate, the actual delivery effect data is judged with the predicted delivery effect data to obtain an initial deviation signal or a result deviation signal;

[0031] When the initial deviation signal is generated, the average duration from the delivery time of the delivered advertisement to the time when the advertisement is opened by clicking is set as the average response duration, and the average response duration is judged to obtain a push optimization signal or a replacement signal;

[0032] When the result deviation signal is generated, the average reading duration of the delivered advertisement is obtained, and the average reading duration is judged to obtain a replacement signal or a fine-tuning signal.

[0033] The beneficial effects of the present application are as follows:

[0034] (1) The application converts unordered data into structured features by feature label extraction and vector conversion, providing reliable input for subsequent analysis, and solving the low efficiency problem of matching caused by scattered data and chaotic format in traditional recruitment;

[0035] (2) The application realizes multi-dimensional feature matching based on cosine similarity, dynamically adjusts the recommendation list according to the behavior of job seekers, ensures that job seekers with high matching degree are preferentially reached, and further filters out high-intention crowds through secondary verification and interaction score weighting, reduces invalid advertisement placement, and reduces enterprise recruitment cost;

[0036] (3) The application converts the priority of the post from "qualitative description" to "quantitative score", provides an objective basis for advertisement budget allocation and placement strategy adjustment, and dynamically optimizes the push period, channel and content by comparing the prediction with the actual effect, promoting the transformation of the traditional experience-driven recruitment mode to the data-driven intelligent decision-making mode. BRIEF DESCRIPTION OF DRAWINGS

[0037] The application will be further described below with reference to the drawings;

[0038] Fig. 1 is a system flow chart of the application;

[0039] Fig. 2 is a partial analysis diagram of embodiment one of the application;

[0040] Fig. 3 is a partial analysis diagram of embodiment two of the application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0042] In this paper, "embodiment" means that the specific features, structures or characteristics described in conjunction with the embodiment can be included in at least one embodiment of the application. The phrase appears at various places in the specification does not necessarily refer to the same embodiment, nor is it independent or alternative to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments;

[0043] Embodiment one:

[0044] Please refer to Figs. 1 to 3As shown, the application is an intelligent recruitment advertisement placement decision system based on a big data model, comprising an advertisement placement management center, a data collection module, a recruitment matching module, a precise placement module, a post division module, a placement effect module, and a response management module. The data collection module is in one-way communication connection with the advertisement placement management center. The advertisement placement management center is in two-way communication connection with the recruitment matching module and the precise placement module. The advertisement placement management center is in one-way communication connection with the post division module and the placement effect module. The post division module and the placement effect module are in one-way communication connection with the response management module.

[0045] The data collection module is used to collect job seeker data, recruitment post data, and market data. Based on the collected job seeker data, recruitment post data, and market data, an original data set is constructed, and the original data set is sent to the advertisement placement management center for storage.

[0046] The job seeker data includes but is not limited to the age, gender, education, work experience, skill certificate, job-seeking intention, browsing history, delivery record, etc. of the job seeker. The recruitment post data includes the post name, post responsibility, post requirement, salary, company background, etc. The market data includes the industry talent supply and demand situation, competitor recruitment strategy, and recruitment heat in different regions, etc. The module outputs the original data set.

[0047] The recruitment matching module is used to analyze the feature label similarity of the original data set. The specific feature label similarity analysis process is as follows:

[0048] S1: The original data set is cleaned, integrated, and standardized to obtain an initial data set. The repeated data, error data, and invalid data are removed. The data from different sources are unified in format and associated to provide high-quality data for subsequent analysis and modeling.

[0049] S2: Feature extraction is performed on the initial data set to obtain a job seeker feature label set and a recruitment post feature label set.

[0050] The job seeker feature label set includes basic attribute labels (such as age, education, work experience, etc.), behavior feature labels (such as "skill inclination" and "post preference" extracted based on browsing records and search keywords using the TF-IDF algorithm), etc.

[0051] The recruitment post feature label set includes hard requirement labels (such as education requirement (such as bachelor's degree or above) and work experience requirement (such as more than 3 years)), post value labels (such as "salary competitiveness: high" based on industry salary level and "post level: senior post" based on post responsibility complexity), etc.

[0052] Through deep processing and feature mining of multi-dimensional data, disordered and scattered raw data is converted into structured and reusable feature labels, providing high-quality input for subsequent analysis.

[0053] S3: Generate an N-dimensional feature vector Vp for each job seeker, with the vector dimensions corresponding to the feature labels in the output job seeker feature label set (such as age, education, skills, etc.);

[0054] Generate an M-dimensional feature vector Vo for each job posting, with the vector dimensions consistent with the job seeker feature vector, where N and M are both greater than zero;

[0055] Standardize each dimension in the vector, and convert non-numeric features into numeric form through word embedding in existing technology;

[0056] S4: Calculate the cosine similarity between the job seeker vector Vp and the job posting vector Vo, and obtain the matching score set in the corresponding preset cosine similarity interval based on the cosine similarity;

[0057] For each job posting, sort the matching scores and select the top G job seekers, where G is greater than 30, to form a job-high matching job seeker list, which is stored in the advertisement delivery management center;

[0058] Combine the job seeker's job-seeking behavior data (such as historical browsing job type, search keywords) to dynamically adjust the list (such as job seekers who frequently search for "remote work" in recent times, who are preferentially recommended for jobs containing this label).

[0059] Embodiment two:

[0060] The delivery precision module is used to perform secondary verification, optimization, and adjustment analysis on the job-high matching job seekers. The specific process of secondary verification, optimization, and adjustment analysis is as follows:

[0061] Based on the hard requirement labels of the job posting, the top G job seekers are subjected to secondary verification, and job seekers who do not meet the hard requirement labels are excluded from the job-high matching job seekers, resulting in candidate job seekers;

[0062] Objective: To ensure that the final reached job seekers meet the basic recruitment threshold of the job, and to avoid ineffective advertisement delivery (such as job seekers who do not meet the education requirement and cannot deliver even if they click on the advertisement, wasting resources);

[0063] Based on the interaction behavior labels of the candidate job seekers and the job, different scores are assigned to the parameters in the interaction behavior labels: 1 point for click behavior, 3 points for delivery behavior, and 5 points for interview behavior;

[0064] Based on the interaction behavior labels, the interaction scores of each candidate job seeker are obtained;

[0065] The sum value of the candidate job seeker's matching score and the interaction score multiplied by the corresponding weight coefficient is set as the final matching score, and the final matching score is judged, the candidate job seeker corresponding to the final matching score greater than the preset final matching score threshold is retained, and is set as a high priority job seeker, a post-high priority job seeker list is constructed, and the post-high priority job seeker list is stored by the advertisement delivery management center, so that the job seekers with high intention and high matching are preferentially reached;

[0066] Even if the screening result always adapts to the real-time demand of the job seeker and the interaction degree of the post, the final realization of "precise reaching of suitable people by suitable post" reduces the recruitment cost and improves the advertisement conversion efficiency;

[0067] The application converts the traditional experience-dependent recruitment advertisement delivery into a quantifiable and optimized intelligent decision-making process through data-driven mode, and provides a systematic solution for enterprises to quickly locate target groups in a large number of job seekers and improve recruitment effect;

[0068] The post division module is used for advertisement delivery and post division matching analysis of the post-high priority job seeker list, and the specific advertisement delivery and post division matching analysis process is as follows:

[0069] Based on the post-high priority job seeker list, the basic state data of the post to be delivered advertisement is obtained, and the basic state data includes post urgency (the sum value of the post vacancy duration and the post vacancy influence degree multiplied by the corresponding preset weight coefficient, and the post vacancy influence degree represents the number of personnel who stop working), importance and recruitment difficulty;

[0070] And the preset weight coefficient corresponding to the post urgency, importance and recruitment difficulty is distributed, and the sum value of the post urgency, importance and recruitment difficulty multiplied by the corresponding preset weight coefficient is set as the post priority score, and the post priority score is judged:

[0071] If the post priority score is greater than or equal to the preset post priority score threshold, the corresponding post is set as an urgent post;

[0072] If the post priority score is less than the preset post priority score threshold, the corresponding post is set as an ordinary post, and the response management module is used for responding to the urgent post and the ordinary post, and displaying the urgent post and the ordinary post, so as to make targeted advertisement budget investment or advertisement delivery decision adjustment according to the different posts, so as to improve the pertinence and reliability of the advertisement delivery;

[0073] That is, the post priority is transformed from "qualitative description" to "quantitative score", which provides objective and executable basis for budget allocation, avoiding the waste of resources caused by traditional "head" allocation;

[0074] Among them, the importance represents the time length between the post response and the completion of the replacement, and the more difficult the replacement is, the higher the importance is.

[0075] The recruitment difficulty represents the sum of the market talent supply and demand ratio (the ratio of the number of job seekers to the number of enterprise needs) and the historical recruitment cycle (the average length of time for the enterprise to recruit similar posts in the past) multiplied by the corresponding preset weight coefficient.

[0076] Embodiment three:

[0077] The delivery effect module is used for initial intention and delivery intention joint analysis on the collected predicted delivery effect data and actual delivery effect data of the delivered advertisement, and the specific initial intention and delivery intention joint analysis process is as follows:

[0078] The predicted delivery effect data and actual delivery effect data of the delivered advertisement are obtained, and the predicted delivery effect data includes predicted click rate and predicted delivery rate;

[0079] The actual delivery effect data and the predicted delivery effect data are discriminated, if the predicted click rate is less than the preset predicted click rate threshold, an initial deviation signal is generated, and if the predicted delivery rate is less than the preset predicted delivery rate threshold, a result deviation signal is generated;

[0080] When the initial deviation signal is generated, the average time length between the delivery time of the delivered advertisement and the clicked opening time is obtained, and the average time length between the delivery time of the delivered advertisement and the clicked opening time is set as the average response time, and the average response time is discriminated, if the average response time is less than the preset average response time threshold, a push optimization signal is generated, and if the average response time is greater than or equal to the preset average response time threshold, a replacement signal is generated, and the response management module is used to respond to the push optimization signal or the replacement signal, then the preset warning operation corresponding to the push optimization signal or the replacement signal is immediately made, for example: the preset warning operation corresponding to the push optimization signal: adjusting the push period or replacing the delivery channel, and the preset warning operation corresponding to the replacement signal: replacing the delivery scheme or optimizing the post-high priority job seeker list, so as to improve the initial intention of the job seeker and provide data support for subsequent delivery;

[0081] When the result deviation signal is generated, the average reading dwell time of the placed advertisement is obtained, and the average reading dwell time is discriminated. If the average reading dwell time is less than the preset average reading dwell time threshold, a replacement signal is generated. If the average reading dwell time is greater than or equal to the preset average reading dwell time threshold, a fine-tuning signal is generated. The response management module is used to respond to the fine-tuning signal or the replacement signal, and immediately make the preset warning operation corresponding to the fine-tuning signal or the replacement signal. For example, the preset warning operation corresponding to the replacement signal: adjusting the priority in the high-priority job seeker list, completing high-sensitivity information (salary benefits, work mode, job responsibilities, etc.), and the preset warning operation corresponding to the fine-tuning signal: reminding delivery, simplifying the delivery process, etc. That is, the user's behavior is adjusted to eliminate the user's concerns and guide action, so that the potential intention of "looking" is truly converted into actual action of "delivering", and the final conversion efficiency of the recruitment advertisement is improved.

[0082] In summary, by feature label extraction and vector conversion, unordered data is converted into structured features, providing reliable input for subsequent analysis, solving the low matching efficiency problem caused by scattered data and chaotic format in traditional recruitment. Based on the cosine similarity of multi-dimensional feature matching, the recommended list is dynamically adjusted according to the behavior of job seekers to ensure that high-matching-degree job seekers are preferentially reached. Through secondary verification and interaction score weighting, high-intention crowds are further screened out to reduce invalid advertisement delivery and reduce enterprise recruitment costs. The priority of the post is converted from "qualitative description" to "quantitative score", providing an objective basis for advertisement budget allocation and delivery strategy adjustment. At the same time, by comparing the prediction with the actual effect, the push period, channel and content are dynamically optimized to promote the transformation of the traditional experience-driven recruitment mode to data-driven intelligent decision-making.

[0083] The threshold is set for result comparison analysis to determine whether it is good or bad. The size of the threshold is determined by combining large model analysis of sample data and artificial experience to set the input storage, which can also be appropriately adjusted by seasonal or rational influence conditions.

[0084] The size of the coefficient is a specific numerical value obtained by quantifying each parameter for subsequent comparison. The size of the coefficient depends on the amount of sample data and the corresponding running coefficient initially set by the person skilled in the art for each group of sample data.

[0085] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can make equivalent substitutions or changes to the technical solutions and inventive concepts of the present application within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A smart recruitment advertising placement decision system based on a big data model, characterized in that, It includes an advertising placement management center, a data collection module, a recruitment matching module, a precise placement module, a job classification module, a placement performance module, and a response management module; The data acquisition module is used to collect job seeker data, job posting data, and market data. Based on the collected job seeker data, job posting data, and market data, it constructs a raw dataset and sends the raw dataset to the advertising placement management center for storage. The job matching module is used to process and analyze the feature label similarity of the original dataset, obtain matching scores, sort the matching scores, and form a job posting-highly matched job seeker list. The precise targeting module is used to perform secondary verification, optimization, and adjustment analysis on job postings and highly matched job seekers, and to process the final matching score to obtain a list of high-priority job seekers for the job postings. The job classification module is used to perform advertising and job classification matching analysis on the job-high priority job applicant list, and to process the obtained job priority scores to identify urgent and ordinary jobs. The campaign performance module is used to perform joint analysis of the initial intent and delivery intent of the collected predicted campaign performance data and actual campaign performance data to obtain the initial deviation signal or the result deviation signal.

2. The intelligent recruitment advertising placement decision system based on a big data model according to claim 1, characterized in that, The analysis process of the recruitment matching module is as follows: S1: The original dataset is cleaned, integrated, and standardized to obtain the initial dataset; S2: Extract features from the initial dataset to obtain a set of feature labels for job seekers and a set of feature labels for job postings; S3: Generate an N-dimensional feature vector Vp for each job seeker, where the vector dimension corresponds to the feature label in the output job seeker feature label set; For each job posting, generate an M-dimensional feature vector Vo, with the vector dimension consistent with the job seeker's feature vector. Each dimension in the vector is standardized, and non-numerical features are converted into numerical form using word embeddings in existing techniques.

3. The intelligent recruitment advertising placement decision system based on a big data model according to claim 2, characterized in that, It also includes S4: Calculate the cosine similarity between the job seeker vector Vp and the job vector Vo, and obtain the matching score set in the corresponding preset cosine similarity interval based on the cosine similarity; For each job posting, candidates are ranked according to their matching scores, and TopG candidates (G score greater than 30) are selected to form a job posting-highly matched candidate list.

4. The intelligent recruitment advertising placement decision system based on a big data model according to claim 1, characterized in that, The analysis process of the precise delivery module is as follows: Based on the hard requirement tags of the job posting, TopG job seekers are further validated, and those who are highly matched to the job posting are removed as those who do not meet the hard requirement tags, thus obtaining the candidate job seekers. Based on the interaction behavior tags between job seekers and positions, different scores are assigned to the parameters in the interaction behavior tags: 1 point for clicking behavior, 3 points for submitting applications, and 5 points for interviewing behavior; Interaction scores for each job candidate were obtained based on interactive behavior tags.

5. The intelligent recruitment advertising placement decision system based on a big data model according to claim 4, characterized in that, The final matching score is set by multiplying the candidate's matching score and interaction score by their respective weight coefficients. The final matching score is then processed to retain candidates whose final matching score is greater than the preset final matching score threshold and set them as high-priority candidates, thus constructing a job-high-priority candidate list.

6. The intelligent recruitment advertising placement decision system based on a big data model according to claim 1, characterized in that, The analysis process of the job classification module is as follows: Based on the job postings - high-priority job seeker list, the basic status data of the job postings to be advertised is obtained. The basic status data includes the job posting urgency, importance and recruitment difficulty. The system assigns preset weight coefficients to the urgency, importance, and recruitment difficulty of a position. The sum of the product of the urgency, importance, and recruitment difficulty with the assigned preset weight coefficients is set as the position priority score. The position priority score is then processed to distinguish between urgent and ordinary positions.

7. The intelligent recruitment advertising placement decision system based on a big data model according to claim 6, characterized in that, The job urgency is the sum of the product of the job vacancy duration and the job vacancy impact, each multiplied by a pre-set weighting coefficient. The job vacancy impact represents the number of employees whose work is stalled. The importance is the time between the response to a job replacement and the completion of the replacement. The recruitment difficulty is the sum of the product of the market talent supply-demand ratio (the ratio of the number of job seekers to the number of job openings) and the historical recruitment cycle (the average time for the company to recruit similar positions in the past), each multiplied by a pre-set weighting coefficient.

8. The intelligent recruitment advertising placement decision system based on a big data model according to claim 1, characterized in that, The analysis process of the delivery performance module is as follows: The system obtains the predicted and actual performance data of the ads that have been placed. The predicted performance data includes the predicted click-through rate and the predicted delivery rate. The actual performance data is compared with the predicted performance data to obtain the initial deviation signal or the result deviation signal. When the initial deviation signal is generated, the average duration between the time when the ad is placed and the time when it is clicked and opened is set as the average response time. The average response time is then processed to obtain the push optimization signal or the replacement signal. When a result deviation signal is generated, the average reading dwell time of the placed advertisement is obtained, and the average reading dwell time is processed to obtain a replacement signal or a fine-tuning signal.

Images