Prediction method, device and equipment for position advertisement purchase and readable storage medium

Abstract

The invention discloses a prediction method and device for position advertisement purchase, equipment and a readable storage medium. The method comprises the steps of obtaining multiple pieces of feature data of a target position; inputting the plurality of feature data into a trained lightweight gradient lifting machine learning model to obtain a first prediction result of the target position; inputting the plurality of feature data into a trained deep neural network model to obtain a second prediction result of the target position; preliminarily determining whether the target position purchases the advertisement or not according to the first prediction result and the second prediction result; if the target position is preliminarily determined to purchase the advertisement, verifying whether first target feature data in the plurality of feature data is consistent with second target feature data of the purchased advertisement; the target feature is an important feature influencing the position advertisement permeability, and the second target feature data is feature data corresponding to the target feature of a sample position for purchasing the advertisement; and if yes, determining that the target position will purchase the advertisement. The method can improve the accuracy and efficiency of recommending and purchasing advertisements.

Description

Technical Field

[0001] The present application relates to the field of big data service technology, and in particular to a prediction method, device, equipment and readable storage medium for job purchase advertisements. Background Art

[0002] On recruitment platforms or recruitment websites, recruiters can increase the exposure of job openings by purchasing advertisements, making them more visible and attracting more job seekers to pay attention to and apply for them. In addition, for some positions that are urgently needed, purchasing advertisements can help recruiters obtain more applications in a short period of time and find suitable candidates as soon as possible.

[0003] However, it is currently difficult for users to know which positions will have a higher return after purchasing ads, resulting in a lower willingness of users to purchase ads, and / or a lower return rate for some positions that purchase ads, resulting in higher costs for users. In addition, for recommenders, in the process of recommending ads, there is currently a lack of accurate quantitative data guidance, and they do not know which positions are more likely to purchase ads, resulting in it being difficult to accurately find users who have a need to purchase ads during the recommendation process, which may disturb other users and result in a poor user experience. Summary of the invention

[0004] In view of this, embodiments of the present application provide a method, device, equipment and readable storage medium for predicting job purchase advertisements, which are used to solve at least one of the above technical problems.

[0005] In a first aspect, an embodiment of the present application provides a method for predicting job purchase advertisements, including: obtaining multiple feature data of a target job; inputting the multiple feature data into a trained lightweight gradient boosting machine learning model to obtain a first prediction result for the target job, the first prediction result being a prediction of whether the target job will purchase advertisements; inputting the multiple feature data into a trained deep neural network model to obtain a second prediction result for the target job, the second prediction result being a prediction of the probability of the target job purchasing advertisements; combining the first prediction result and the second prediction result to preliminarily determine whether the target job will purchase advertisements; if it is preliminarily determined that the target job will purchase advertisements, first target feature data corresponding to the target feature is screened out from the multiple feature data, and verifying whether the first target feature data is consistent with the second target feature data for purchasing advertisements; wherein the first target feature data corresponds one-to-one to the second target feature data, the target feature is an important feature affecting the penetration rate of job advertisements determined by the lightweight gradient boosting machine learning model based on the feature data of multiple sample jobs, and the second target feature data is feature data corresponding to the target feature of at least one sample job that purchases advertisements; if they are consistent, it is determined that the target job will purchase advertisements.

[0006] According to some embodiments of the present application, optionally, the prediction method for job purchase advertisements also includes: if it is preliminarily determined that the target position will not purchase advertisements, first target feature data corresponding to the target feature is screened out from multiple feature data, and whether the first target feature data is consistent with third target feature data of non-purchased advertisements; wherein the first target feature data corresponds one-to-one to the third target feature data, and the third target feature data is feature data corresponding to the target feature of at least one sample position that has not purchased advertisements; if they are consistent, it is determined that the target position will not purchase advertisements; if they are inconsistent, a prompt message is sent to the target terminal device.

[0007] According to some embodiments of the present application, optionally, before inputting multiple feature data into the trained lightweight gradient boosting machine learning model to obtain the first prediction result of the target position, the prediction method of the position purchase advertisement also includes: obtaining a training set and a test set, each of which includes multiple sample positions, and each sample position includes multiple feature data; training the lightweight gradient boosting machine learning model based on the training set, and when the accuracy of the lightweight gradient boosting machine learning model predicting the purchase advertisement of the training set is greater than or equal to a preset threshold, or when the number of iterations reaches a preset threshold, verifying the lightweight gradient boosting machine learning model based on the test set; when the accuracy of the lightweight gradient boosting machine learning model predicting the purchase advertisement of the test set is greater than or equal to the preset threshold, obtaining a trained lightweight gradient boosting machine learning model; outputting a feature importance sorting graph based on the trained lightweight gradient boosting machine learning model, and sorting the multiple features corresponding to the multiple feature data in the feature importance sorting graph according to importance; selecting the top N features in the sorting as target features, and selecting feature data corresponding to the target features of at least one sample position that purchases advertisements as second target feature data, where N is a positive integer.

[0008] According to some embodiments of the present application, optionally, the target feature includes at least a first target feature having a positive impact on the purchase of advertisements, and the first target feature is determined based on a honeycomb diagram output by a trained lightweight gradient boosting machine learning model.

[0009] According to some embodiments of the present application, optionally, the characteristic data of the sample positions in the test set are characteristic data of the sample positions on weekdays, the training set includes positive sample positions that purchased advertisements and negative sample positions that did not purchase advertisements, the negative sample positions are randomly selected from multiple negative sample positions, and the difference between the number of positive sample positions and the number of negative sample positions in the training set is less than a preset difference.

[0010] According to some embodiments of the present application, optionally, the multiple feature data of the target position include at least one of the following: recent behavioral feature data of the recruiter, feature data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the display-to-investment ratio of the target position, the average daily number of submissions of the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, the average daily number of exposures of the target position under the work location-function grid, the display-to-investment ratio of the work location-function grid, the advertising display-to-investment ratio of the work location-function grid, and the difference between the advertising display-to-investment ratio and the natural display-to-investment ratio of the work location-function grid; multiple Inputting feature data into a trained lightweight gradient boosting machine learning model to obtain a first prediction result for a target position includes: inputting a first part of feature data from multiple feature data into a trained lightweight gradient boosting machine learning model to obtain a first prediction result for the target position; wherein the first part of feature data includes at least one of the following: recent behavioral feature data of the recruiter, feature data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, and the average daily exposure of the target position under the work location-function grid.

[0011] According to some embodiments of the present application, optionally, multiple feature data are input into a trained deep neural network model to obtain a second prediction result for the target position, including: when the target position is an existing position in the recruitment platform, multiple feature data are input into the trained deep neural network model to obtain the second prediction result for the target position; when the target position is a new position, other feature data among the multiple feature data except the investment-to-expense ratio of the target position are input into the trained deep neural network model to obtain the second prediction result for the target position.

[0012] In a second aspect, an embodiment of the present application provides a prediction device for a job advertisement purchase, comprising: an acquisition module for acquiring multiple feature data of a target job; a first prediction module for inputting multiple feature data into a trained lightweight gradient boosting machine learning model to obtain a first prediction result for the target job, the first prediction result being a prediction of whether the target job will purchase advertisements; a second prediction module for inputting multiple feature data into a trained deep neural network model to obtain a second prediction result for the target job, the second prediction result being a prediction of the probability of the target job purchasing advertisements; a preliminary determination module for combining the first prediction result and the second prediction result to preliminarily determine whether the target job will purchase advertisements; a verification module for, if it is preliminarily determined that the target job will purchase advertisements, filtering out first target feature data corresponding to the target feature from multiple feature data, and verifying whether the first target feature data is consistent with the second target feature data for purchasing advertisements; wherein the first target feature data corresponds one-to-one to the second target feature data, the target feature is an important feature affecting the penetration rate of job advertisements determined by the lightweight gradient boosting machine learning model based on the feature data of multiple sample jobs, and the second target feature data is feature data corresponding to the target feature of at least one sample job that purchases advertisements; if they are consistent, it is determined that the target job will purchase advertisements.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the method for predicting job purchase advertisements as described above are implemented.

[0014] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, the steps of the method for predicting job purchase advertisements as described above are implemented.

[0015] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes computer program instructions, and when the computer program instructions are executed by a processor, the steps of the method for predicting job purchase advertisements as described above are implemented.

[0016] The prediction method, device, equipment and readable storage medium for job purchase advertisements provided by the embodiments of the present application are combined with the first prediction result of the lightweight gradient boosting machine learning model for the target job and the second prediction result of the deep neural network model for the target job. The two verify each other, and the accurate prediction of whether the target job will purchase advertisements can be achieved. On this basis, the first target feature data of the target job is verified by the second target feature data of the important feature that affects the job advertisement penetration rate, which is equivalent to further verifying the prediction result of whether the target job will purchase advertisements from the perspective of job advertisement penetration rate, and can further improve the accuracy of the prediction result of whether the target job will purchase advertisements. It can not only provide a reference for users who publish target jobs to purchase advertisements, which helps to improve the return rate of purchasing advertisements, but also provide a reference for recommenders, which helps to help recommenders accurately find users / jobs with the need to purchase advertisements, improve the accuracy and efficiency of recommending purchase advertisements, reduce the disturbance to other users, and improve the user experience. In addition, the first target feature data and the second target feature data of the target feature are conducive to providing data guidance for recommenders, helping recommenders to select suitable jobs for advertisement recommendation according to needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following briefly introduces the drawings in the embodiments of the present application.

[0018] Figure 1 A schematic diagram of a flow chart of a method for predicting job purchase advertisements provided in an embodiment of the present application.

[0019] Figure 2 Another flowchart of the method for predicting job purchase advertisements provided in an embodiment of the present application is provided.

[0020] Figure 3 A feature importance ranking diagram is schematically shown.

[0021] Figure 4 Schematic illustration of a honeycomb diagram of the output of a lightweight gradient boosting machine learning model.

[0022] Figure 5 A schematic diagram of the structure of a device for predicting job purchase advertisements provided in an embodiment of the present application.

[0023] Figure 6 A schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0025] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0026] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0027] It is obvious to those skilled in the art that various modifications and changes can be made in the present application without departing from the spirit or scope of the present application. Therefore, the present application is intended to cover modifications and changes of the present application that fall within the scope of the corresponding claims (technical solutions for protection) and their equivalents. It should be noted that the implementation methods provided in the embodiments of the present application can be combined with each other without contradiction.

[0028] Before describing the technical solutions provided by the embodiments of the present application, in order to facilitate the understanding of the embodiments of the present application, the present application first specifically describes the problems existing in the related art:

[0029] On recruitment platforms or recruitment websites, recruiters can increase the exposure of job openings by purchasing advertisements, making them more visible and attracting more job seekers to pay attention to and apply for them. In addition, for some positions that are urgently needed, purchasing advertisements can help recruiters obtain more applications in a short period of time and find suitable candidates as soon as possible.

[0030] However, it is currently difficult for users to know which positions will have a higher return after purchasing ads, resulting in a lower willingness of users to purchase ads, and / or a lower return rate for some positions that purchase ads, resulting in higher costs for users. In addition, for recommenders, in the process of recommending ads, there is currently a lack of accurate quantitative data guidance, and they do not know which positions are more likely to purchase ads, resulting in it being difficult to accurately find users who have a need to purchase ads during the recommendation process, which may disturb other users and result in a poor user experience.

[0031] In view of the above research findings of the inventors, the embodiments of the present application provide a method, device, equipment and readable storage medium for predicting job purchase advertisements, which can solve at least one of the above technical problems existing in the related art.

[0032] The following first introduces the prediction method of job purchase advertisements provided in the embodiment of the present application.

[0033] Figure 1 A flow chart of a method for predicting job purchase advertisements provided in an embodiment of the present application. Figure 1 As shown, the method for predicting job purchase advertisements provided in the embodiment of the present application may include the following steps S101 to S106.

[0034] S101: Acquire multiple feature data of a target position.

[0035] The target position can be any recruitment position. For example, the target position can be an existing position in the recruitment platform, that is, a published position, or a new position that has not been published before. In an embodiment of the present application, the target position can have multiple feature data. For example, the multiple feature data of the target position may include recent behavioral feature data of the recruiter (such as the recruiter who publishes the target position), such as the average number of HR logins to the recruitment platform in the last n days, the average number of HR chat replies in the last n days, and the average number of HR resume feedback in the last n days. The size of n can be flexibly adjusted according to actual conditions, for example, n can be equal to 1, 3, 7 or 30, etc., and this application does not limit this. Multiple feature data can also include basic feature data of the target position, such as the function, work place, salary and company of the target position. Of course, multiple feature data can also include other feature data. This will be described in detail below, so I will not go into details here.

[0036] S102: Input multiple feature data into the trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position, where the first prediction result is to predict whether the target position will purchase advertisements.

[0037] The Lightweight Gradient Boosting Machine Learning Model (LightGBM model for short) is a model built based on the Gradient Boosting Decision Tree (GBDT) algorithm, which is suitable for processing large-scale data. It uses a decision tree-based algorithm to achieve efficient training and prediction. Compared with other gradient boosting frameworks (such as XGBoost), LightGBM has faster training speed and lower memory consumption when processing large-scale data.

[0038] In the embodiment of the present application, the lightweight gradient boosting machine learning model can be used to process the binary classification problem of whether a position will purchase advertisements. In addition, the feature importance ranking diagram and / or honeycomb diagram output by the lightweight gradient boosting machine learning model can be used to analyze the impact of each feature on the penetration rate of job advertisements or the results of job advertisement purchases, and screen out the important features that affect the penetration rate of job advertisements, i.e., the target features, so as to facilitate the subsequent verification of the prediction results of whether the target position will purchase advertisements, and provide data guidance for the recommender.

[0039] The job advertisement penetration rate is an indicator to measure the effectiveness of job advertisements in the target audience. Specifically, in some examples, the job advertisement penetration rate may represent the ratio of the number of positions A that purchased advertisements for position A, such as software development engineer, to the total number of positions A. Position A may be any position.

[0040] The job advertisement penetration rate is positively correlated with the job advertisement purchase results. Generally, the higher the job advertisement penetration rate, the more likely the job is to purchase advertisements, that is, the higher the probability of purchasing advertisements. Therefore, the feature data corresponding to the target feature that has a greater impact on the job advertisement penetration rate can be used to verify the prediction results of whether the target job will purchase advertisements, thereby improving the accuracy of the prediction results of whether the target job will purchase advertisements.

[0041] In S102, a portion of the feature data among the multiple feature data may be input into the trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position output by the lightweight gradient boosting machine learning model. Of course, in order to improve the prediction accuracy, the multiple feature data may also be input into the trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position output by the lightweight gradient boosting machine learning model, and this application does not limit this.

[0042] The first prediction result of the target position may be a prediction of whether the target position will purchase advertisements. That is, the first prediction result may be a prediction that the target position will purchase advertisements or a prediction that the target position will not purchase advertisements.

[0043] S103: Input multiple feature data into the trained deep neural network model to obtain a second prediction result for the target position, where the second prediction result is a probability of purchasing an advertisement for the target position.

[0044] A deep neural network (DNN) is an artificial neural network architecture that learns and represents data through multiple layers of neurons. The basic structure of a deep neural network model can include an input layer, multiple hidden layers, and an output layer. The neurons (nodes) of each layer are connected by weights, and the weights between adjacent layers are updated through a back-propagation algorithm to minimize the error between the predicted value and the actual value. A deep neural network model can contain more than two hidden layers, which enables the deep neural network model to learn more complex features and patterns. Deep neural network models have good performance in handling many complex tasks.

[0045] In an embodiment of the present application, a deep neural network model can be used to predict the probability of purchasing an advertisement for a target position, such as predicting that the probability of purchasing an advertisement for a target position is 10%, 30%, 50% or 90%, etc., and the present application does not limit this.

[0046] In S103, at least part of the feature data in the plurality of feature data may be input into a trained deep neural network model to obtain a second prediction result of the target position output by the deep neural network model. The second prediction result of the target position may be a probability of predicting the target position to purchase an advertisement.

[0047] S104: Combining the first prediction result and the second prediction result, preliminarily determining whether the target position will purchase an advertisement.

[0048] For example, when the first prediction result predicts that the target position will purchase advertisements, and the probability of predicting that the target position will purchase advertisements in the second prediction result is greater than or equal to the first preset threshold, it can be preliminarily determined that the target position will purchase advertisements. When the first prediction result predicts that the target position will not purchase advertisements, and the probability of predicting that the target position will purchase advertisements in the second prediction result is less than the second preset threshold, it can be preliminarily determined that the target position will not purchase advertisements. The first preset threshold and the second preset threshold can be flexibly adjusted according to actual conditions, and this application does not limit this. The second preset threshold can be less than or equal to the first preset threshold.

[0049] In this way, by combining the first prediction result of the lightweight gradient boosting machine learning model for the target position and the second prediction result of the deep neural network model for the target position, the two verify each other, which can improve the accuracy of the prediction result of whether the target position will purchase advertisements.

[0050] S105: If it is preliminarily determined that the target position will purchase advertisements, first target feature data corresponding to the target feature is screened out from the plurality of feature data, and it is verified whether the first target feature data is consistent with second target feature data for purchasing advertisements.

[0051] Among them, the target feature and the second target feature data can be predetermined. Specifically, the lightweight gradient boosting machine learning model can be trained based on the feature data of multiple sample positions to obtain a trained lightweight gradient boosting machine learning model. The target feature can be an important feature that affects the penetration rate of job advertisements determined by the trained lightweight gradient boosting machine learning model, such as whether the company that posted the job has purchased advertisements in the last 30 days, the advertising penetration rate of the industry in which the company that posted the job is located in the last 3 days, and the average number of resume feedbacks from HR in the last 30 days. The second target feature data can be feature data corresponding to the target feature of at least one sample position that purchased advertisements. For example, the second target feature data can be that the company that posted the job has purchased advertisements in the last 30 days, the advertising penetration rate of the industry in which the company that posted the job is located in the last 3 days is greater than or equal to a preset value, and the average number of resume feedbacks from HR in the last 30 days is greater than or equal to a preset number. The size of the preset value and the preset number of times can be flexibly adjusted according to actual conditions, and this application does not limit this.

[0052] The first target feature data is feature data corresponding to the target feature of the target position, and the first target feature data corresponds to the second target feature data one by one. For example, the first target feature data may be that the company that posted the target position has purchased advertisements in the last 30 days, the advertising penetration rate of the industry of the company that posted the target position in the last 3 days is x1, and the average number of resume feedbacks from HR in the last 30 days is y1.

[0053] Since the company that posted the target position has purchased advertisements in the last 30 days, if x1 is greater than the preset value and y1 is greater than or equal to the preset number of times, it means that the first target feature data is consistent with the second target feature data. Otherwise, it means that the first target feature data is inconsistent with the second target feature data.

[0054] S106: If they are consistent, it is determined that the target position will purchase advertisements.

[0055] If the first target feature data is consistent with the second target feature data, it can be finally determined that the target position will purchase advertisements.

[0056] The prediction method for job purchase advertisements provided in the embodiment of the present application combines the first prediction result of the target job by the lightweight gradient boosting machine learning model and the second prediction result of the target job by the deep neural network model. The two verify each other, and can realize the accurate prediction of whether the target job will purchase advertisements. On this basis, the first target feature data of the target job is verified by the second target feature data of the important feature that affects the penetration rate of job advertisements, which is equivalent to further verifying the prediction result of whether the target job will purchase advertisements from the perspective of the penetration rate of job advertisements, and can further improve the accuracy of the prediction result of whether the target job will purchase advertisements. It can not only provide a reference for users who publish target jobs to purchase advertisements, which helps to improve the return rate of purchasing advertisements, but also provide a reference for recommenders, which helps to help recommenders accurately find users / jobs with the need to purchase advertisements, improve the accuracy and efficiency of recommending purchase advertisements, reduce the disturbance to other users, and improve the user experience. In addition, the first target feature data and the second target feature data of the target feature are conducive to providing data guidance for recommenders, helping recommenders to select suitable jobs for advertisement recommendation according to needs.

[0057] According to some embodiments of the present application, optionally, the method for predicting job purchase advertisements provided by the embodiments of the present application may further include the following steps 1 and 2.

[0058] Step 1: If it is preliminarily determined that the target position will not purchase advertisements, first target feature data corresponding to the target feature is screened out from multiple feature data, and it is verified whether the first target feature data is consistent with third target feature data that has not purchased advertisements.

[0059] The third target feature data may be feature data corresponding to the target feature of at least one sample position that has not purchased an advertisement. For example, the third target feature data may be that the company that posted the position has not purchased an advertisement in the past 30 days, the advertising penetration rate of the industry in which the company that posted the position is located in the past three days is less than a preset value, and the average number of resume feedbacks from HR in the past 30 days is less than a preset number.

[0060] The first target characteristic data is characteristic data corresponding to the target characteristic of the target position, and the first target characteristic data corresponds to the second target characteristic data one by one. The first target characteristic data and the third target characteristic data can be compared one by one.

[0061] For example, if the first target feature data is that the company that posted the target position has not purchased advertising in the last 30 days, the advertising penetration rate of the industry in which the company that posted the target position is located in the last 3 days is x2, and the average number of resume feedbacks from HR in the last 30 days is y2, since the company that posted the target position has not purchased advertising in the last 30 days, if x2 is less than the preset value and y2 is less than or equal to the preset number, it means that the first target feature data is consistent with the third target feature data. Otherwise, it means that the first target feature data is inconsistent with the third target feature data.

[0062] Step 2: If they are consistent, it is determined that the target position will not purchase advertisements; if they are inconsistent, a prompt message is sent to the target terminal device.

[0063] If the first target feature data is consistent with the third target feature data, it can be determined that the target position will not purchase advertisements. If the first target feature data is inconsistent with the third target feature data, a prompt message can be sent to the target terminal device to prompt the user that the prediction result output by the lightweight gradient boosting machine learning model and / or the deep neural network model that the target position will not purchase advertisements may be misjudged. The target terminal device can be a terminal device used by developers or recommenders.

[0064] According to some embodiments of the present application, optionally, before inputting multiple feature data of the target position into the trained lightweight gradient boosting machine learning model and deep neural network model, the lightweight gradient boosting machine learning model and the deep neural network model can also be trained.

[0065] Figure 2 Another flow chart of the method for predicting job purchase advertisements provided in the embodiment of the present application is shown below. Figure 2 As shown, in some embodiments, before S102: inputting multiple feature data into a trained lightweight gradient boosting machine learning model to obtain a first prediction result for the target position, the prediction method for job purchase advertisements provided in the embodiments of the present application may also include the following steps S201 to S205.

[0066] S201: Obtain a training set and a test set, each of which includes a plurality of sample positions, and each sample position includes a plurality of feature data.

[0067] The sample positions can be existing positions in the recruitment platform, that is, positions that have been posted, or they can be new positions that have not been posted before. Both the training set and the test set can include feature data of multiple sample positions, and each sample position can include multiple feature data. Exemplarily, the multiple feature data of the sample position include at least one of the following: recent behavioral feature data of the recruiter who posted the sample position, feature data of the company where the sample position is located, the function of the sample position, the work place of the sample position, the average salary level of the sample position, the minimum education level of the sample position, the exhibition-to-investment ratio of the sample position, the average daily number of submissions of the sample position, the function advertising penetration rate, the work place advertising penetration rate, the company industry advertising penetration rate, the average daily exposure number of the sample position under the work place-function grid, the exhibition-to-investment ratio of the work place-function grid, the advertising exhibition-to-investment ratio of the work place-function grid, and the difference between the advertising exhibition-to-investment ratio and the natural exhibition-to-investment ratio of the work place-function grid.

[0068] The exposure ratio of a sample position may be the ratio of the number of clicks to the number of impressions of the sample position. The functional advertising penetration rate may be the ratio of the number of positions that purchased advertisements under a certain function to the total number of positions under the function. The workplace advertising penetration rate may be the ratio of the number of positions that purchased advertisements at a certain workplace to the total number of positions at the workplace. The company industry advertising penetration rate may be the ratio of the number of positions that purchased advertisements in a certain industry to the total number of positions in the industry. In an embodiment of the present application, a certain workplace and a certain function are taken as a grid, and the average daily exposure number, the advertising exposure ratio, and the difference between the advertising exposure ratio and the natural exposure ratio of the sample positions under the grid are counted. It can be understood that for the sample position, the grid is a grid composed of the workplace where the sample position is located and the function corresponding to the sample position. For the target position, the grid is a grid composed of the workplace where the target position is located and the function corresponding to the target position.

[0069] The display-to-investment ratio of the work location-function grid may be the sum of the advertising display-to-investment ratio of the work location-function grid and the natural display-to-investment ratio (i.e., non-advertising display-to-investment ratio) of the work location-function grid. The advertising display-to-investment ratio of the work location-function grid may be the ratio of the number of clicks on the positions under the work location-function grid to the number of advertising impressions. The natural display-to-investment ratio of the work location-function grid may be the ratio of the number of clicks on the positions under the work location-function grid to the number of non-advertising impressions.

[0070] Table 1 schematically shows multiple characteristic data of sample positions and target positions.

[0071] Table 1

[0072]

[0073]

[0074]

[0075] As shown in Table 1, in some examples, the recent behavioral feature data of the recruiter who posted the sample position may include at least one of the following: the average number of HR logins in the last n days, the average number of HR chat replies in the last n days, the average number of HR active chats in the last n days, the average number of HR resume feedback in the last n days, the average number of HR-position chat replies in the last n days, the average number of HR-position active chats in the last n days, and the average number of HR-position resume feedback in the last n days. n is a positive integer, and the size of n can be flexibly adjusted according to actual conditions. For example, n can be equal to 1, 3, 7 or 30, etc., and this application does not limit this. Among them, for the sample position, the average number of HR-position chat replies in the last n days can be the average number of chat replies by HR for the sample position.

[0076] The company characteristic data of the sample position may include at least one of the size of the company, the nature of the company, the industry of the company, and whether the company has purchased advertisements in the past month.

[0077] In some specific embodiments, when training a lightweight gradient boosting machine learning model, for example, the following characteristic data of sample positions can be selected to train the lightweight gradient boosting machine learning model, including: the average daily number of HR logins in the last n days, the average daily number of HR chat replies in the last n days, the average daily number of HR active chats in the last n days, the average daily number of HR resume feedback in the last n days, the size of the company where the sample position is located, the nature of the company where the sample position is located, the industry of the company where the sample position is located, whether the company where the sample position is located has purchased advertisements in the past month, the function of the sample position, the work location of the sample position, the average salary level of the sample position, the minimum educational level of the sample position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, and the average daily exposure of the sample position under the work location-function grid.

[0078] In some specific embodiments, when training a deep neural network model, the following feature data of sample positions can be added to the feature data used by the above-mentioned lightweight gradient boosting machine learning model to train the deep neural network model, including: the average daily number of chat replies for HR-positions in the last n days, the average daily number of active chats for HR-positions in the last n days, the average daily number of resume feedback for HR-positions in the last n days, the display-to-investment ratio of sample positions, the average daily number of submissions for sample positions, the display-to-investment ratio of the work location-function grid, the advertising display-to-investment ratio of the work location-function grid, and the difference between the advertising display-to-investment ratio and the natural display-to-investment ratio of the work location-function grid. It should be noted that when the sample position is a new position, other feature data other than the display-to-investment ratio of the sample position can be selected. That is, when the sample position is a new position, the feature data of the display-to-investment ratio of the sample position can be omitted.

[0079] In this way, on the basis of the feature data used by the above-mentioned lightweight gradient boosting machine learning model, adding features related to the sample positions and features related to the exhibition-to-investment ratio to train the deep neural network model can improve the accuracy of the deep neural network model in predicting the probability of job purchase advertisements.

[0080] According to some embodiments of the present application, optionally, the training set may include positive sample positions that purchased advertisements and negative sample positions that did not purchase advertisements. Considering that the number of negative sample positions that did not purchase advertisements in the recruitment platform may be large, the negative sample positions in the training set can be randomly selected from multiple negative sample positions in the recruitment platform. That is, a portion of negative sample positions are randomly selected as negative sample positions in the training set. The difference between the number of positive sample positions and the number of negative sample positions in the training set is less than the preset difference, so that the number of positive sample positions in the training set is the same as or similar to the number of negative sample positions, thereby solving the problem of imbalance between the number of positive sample positions and the number of negative sample positions, and improving the accuracy of prediction of the lightweight gradient boosting machine learning model. The size of the preset difference can be flexibly adjusted according to actual conditions, and this application does not limit this.

[0081] According to some embodiments of the present application, the feature data of the sample positions in the test set may optionally be the feature data of the sample positions on weekdays, such as the feature data of the sample positions on the most recent non-weekend day. In this way, the impact of too few positive sample positions on weekends on the prediction results can be reduced, and the accuracy of the prediction of the lightweight gradient boosting machine learning model can be improved.

[0082] S202: Train the lightweight gradient boosting machine learning model based on the training set, and verify the lightweight gradient boosting machine learning model based on the test set when the accuracy of the lightweight gradient boosting machine learning model in predicting the purchase of advertisements in the training set is greater than or equal to a preset threshold, or when the number of iterations reaches a preset iteration threshold.

[0083] During the training phase, the training set can be used to fit the lightweight gradient boosting machine learning model. By feeding the training data, the lightweight gradient boosting machine learning model adjusts its model parameters to minimize the error between the predicted value and the true value. When the accuracy of the lightweight gradient boosting machine learning model predicting the purchased ads in the training set is greater than or equal to the preset threshold, or when the number of iterations reaches the preset iteration threshold, the training ends, and then the lightweight gradient boosting machine learning model is verified based on the test set.

[0084] For example, in some embodiments, the lightweight gradient boosting machine learning model is trained by the training set, and the model parameters of the lightweight gradient boosting machine learning model are adjusted. The model parameters of the lightweight gradient boosting machine learning model finally obtained are as follows:

[0085] The preset iteration threshold (n_estimators) is 1000, the learning rate (learning_rate) is 0.05, the number of parallel multi-threads (n_jobs) is 16, the calculation method of feature importance (importance_type) is 'gain', the number of leaf nodes (num_leaves) is 56, the row sampling frequency (subsample_freq) is 1, the row sampling ratio (subsample) is 0.9, the L1 regularization parameter (reg_alpha) is 0.08, and the L2 regularization parameter (reg_lambda) is 0.01.

[0086] Among them, the learning rate (learning_rate) controls the amplitude of each iteration of the model. The larger the learning rate model, the faster the training speed, and the smaller the learning rate model, the more stable the training speed, and the two need to be balanced. The more parallel multithreading (n_jobs), the faster the training speed. The more leaf nodes (num_leaves), the deeper the split binary tree structure. The row sampling frequency (subsample_freq) indicates how many times the re-sampling is performed. The row sampling ratio (subsample) is between 0-1. Sampling a certain proportion of samples when training each tree can increase the training speed. The L1 regularization parameter reg_alpha and the L2 regularization parameter reg_lambda are used to control overfitting. It should be noted that the above model parameters are only examples and do not constitute a limitation of this application.

[0087] According to some embodiments of the present application, optionally, before training the lightweight gradient boosting machine learning model based on the training set in S202, the feature data of the sample positions in the training set and the test set may be preprocessed. For example, missing values ​​may be filled in for numerical feature data (such as filling the median), and missing values ​​may be filled in for categorical feature data (such as filling constants) and One-hot encoding may be performed. One-hot encoding is a technique for converting categorical variables into formatted data. For example, each category may be represented as a binary vector in which only one element is 1 and all other elements are 0. This approach allows the model to better handle non-numerical features.

[0088] S203: When the accuracy of the lightweight gradient boosting machine learning model in predicting the purchase of advertisements in the test set is greater than or equal to a preset threshold, a trained lightweight gradient boosting machine learning model is obtained.

[0089] During the testing phase, the test set can be used to evaluate the final performance of the model to ensure that the model is not overfitted on the training data. When the accuracy of the lightweight gradient boosting machine learning model predicting the purchase of advertisements in the test set is greater than or equal to the preset threshold, it means that the prediction accuracy of the lightweight gradient boosting machine learning model meets the requirements, thereby obtaining a trained lightweight gradient boosting machine learning model. The size of the preset threshold can be flexibly adjusted according to actual conditions, and this application does not limit this.

[0090] S204: Outputting a feature importance sorting graph based on the trained lightweight gradient boosting machine learning model, wherein multiple features corresponding to the multiple feature data in the feature importance sorting graph are sorted according to importance.

[0091] Based on the trained lightweight gradient boosting machine learning model, a feature importance ranking graph can be output. Figure 3 The feature importance ranking diagram is schematically shown. Figure 3 As shown in FIG. 1 , in the feature importance ranking diagram output by the trained lightweight gradient boosting machine learning model, multiple features can be ranked according to importance. The importance can be the importance of the prediction result of the job advertisement penetration rate or whether the job advertisement is purchased. For example, Figure 3 It can be seen that features such as whether the company has purchased advertisements in the last month (30 days), the company's advertising penetration rate, and HR's recent behavioral characteristics play a relatively important role in the training of the lightweight gradient boosting machine learning model. That is, these features have a greater impact on the prediction results of the job advertising penetration rate or whether the job has purchased advertisements.

[0092] S205: Select the top N features in the ranking as target features, and select feature data corresponding to the target features of at least one sample position of the purchased advertisement as second target feature data, where N is a positive integer.

[0093] The size of N can be flexibly adjusted according to actual conditions, and this application does not limit this. For example, in some examples, the top 1, top 3, top 5 or top 10 features in the ranking can be selected as target features. After determining the target feature, the second target feature data can be determined based on the feature data corresponding to the target feature of at least one sample position that purchased an advertisement. For example, the second target feature data can be that the company that posted the position has purchased advertisements in the past 30 days, the advertising penetration rate of the industry in which the company that posted the position is located in the past 3 days is greater than or equal to a preset value, and the average daily number of resume feedback from HR in the past 30 days is greater than or equal to a preset number, etc.

[0094] According to some embodiments of the present application, optionally, the trained lightweight gradient boosting machine learning model can also output a honeycomb diagram, through which the influence direction of each feature on the purchase advertising behavior can be determined. Figure 4 Schematic diagram showing the honeycomb diagram of the output of a lightweight gradient boosting machine learning model. Figure 4 As shown, for example, by Figure 4 The honeycomb chart shown shows that whether a company has purchased advertising in the past 30 days has a positive impact on the purchase of advertising, that is, companies that have purchased advertising before are more likely to purchase advertising again, while companies that have not purchased advertising are less likely to purchase advertising.

[0095] Accordingly, the target feature may include at least a first target feature that has a positive impact on the purchase of advertisements, such as whether the company has purchased advertisements in the past 30 days. The first target feature may be determined based on a honeycomb diagram output by a trained lightweight gradient boosting machine learning model.

[0096] In this way, based on the feature importance ranking diagram and honeycomb diagram output by the lightweight gradient boosting machine learning model, we can analyze the impact of each feature on the job advertisement penetration rate or the job advertisement purchase results, analyze the direction of influence of each feature on the advertisement purchase behavior, and screen out important features that affect the job advertisement penetration rate and have a positive impact on the purchase of advertisements, namely, target features. This can facilitate the subsequent verification of the prediction results of whether the target job will purchase advertisements, and provide data guidance for recommenders.

[0097] According to some embodiments of the present application, optionally, before inputting the multiple feature data of the target position into the trained deep neural network model, the deep neural network model can also be trained based on the feature data of multiple sample positions. The feature data of the sample positions used to train the deep neural network model have been described above and will not be repeated here.

[0098] According to some embodiments of the present application, optionally, similar to the multiple feature data of the sample position, the multiple feature data of the target position may include at least one of the following: recent behavioral feature data of the recruiter, feature data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the display-to-investment ratio of the target position, the average daily number of submissions for the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, the average daily number of exposures of the target position under the work location-function grid, the display-to-investment ratio of the work location-function grid, the advertising display-to-investment ratio of the work location-function grid, and the difference between the advertising display-to-investment ratio and the natural display-to-investment ratio of the work location-function grid.

[0099] The recent behavioral characteristic data of the recruiter (the recruiter who posted the target position) may include at least one of the following: the average number of HR logins per day in the last n days, the average number of HR chat replies per day in the last n days, the average number of HR active chats per day in the last n days, the average number of HR resume feedback per day in the last n days, the average number of HR-position chat replies per day in the last n days, the average number of HR-position active chats per day in the last n days, and the average number of HR-position resume feedback per day in the last n days. n is a positive integer, and the size of n can be flexibly adjusted according to actual conditions. For example, n can be equal to 1, 3, 7 or 30, etc., and this application does not limit this. Among them, for the target position, the average number of HR-position chat replies per day in the last n days can be the average number of chat replies per day by HR for the target position.

[0100] The characteristic data of the company where the target position is located may include at least one of the size of the company where the target position is located, the nature of the company where the target position is located, the industry of the company where the target position is located, and whether the company where the target position is located has purchased advertisements in the last month.

[0101] Accordingly, S102: inputting the plurality of feature data into the trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position may include the following steps:

[0102] The first part of the feature data among the multiple feature data is input into the trained lightweight gradient boosting machine learning model to obtain the first prediction result of the target position.

[0103] Among them, the first part of characteristic data includes at least one of the following: recent behavioral characteristic data of recruiters, characteristic data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, and the average daily exposure number of the target position under the work location-function grid.

[0104] In this way, by inputting the first part of the feature data among the multiple feature data into the trained lightweight gradient boosting machine learning model, the first prediction result of the target position output by the lightweight gradient boosting machine learning model is obtained.

[0105] According to some embodiments of the present application, optionally, S103: inputting multiple feature data into a trained deep neural network model to obtain a second prediction result for the target position may include the following steps three and four.

[0106] Step 3: When the target position is an existing position in the recruitment platform, multiple feature data are input into the trained deep neural network model to obtain the second prediction result of the target position.

[0107] Step 4: When the target position is a new position, the other feature data except the investment-to-expansion ratio of the target position among the multiple feature data are input into the trained deep neural network model to obtain the second prediction result of the target position.

[0108] For example, in some embodiments, when the target position is a new position, for example, the recent behavioral characteristic data of the recruiter, except for the display-to-investment ratio of the target position, the characteristic data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the average daily number of submissions for the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, the average daily number of exposures of the target position in the work location-function grid, the display-to-investment ratio of the work location-function grid, the advertising display-to-investment ratio of the work location-function grid, and the difference between the advertising display-to-investment ratio of the work location-function grid and the natural display-to-investment ratio can be input into the trained deep neural network model to obtain a second prediction result for the target position.

[0109] According to some embodiments of the present application, optionally, the prediction method of job purchase advertisements provided in the embodiments of the present application may further include the following steps: outputting an analysis report based on the target features. The analysis report may be used as a reference for recommenders when recommending purchase advertisements.

[0110] For example, if the target features are company size and company nature, data analysis of multiple companies with company sizes ranging from 50 to 150 people shows that the smaller the company, the lower the channel exposure ratio, and the higher the proportion of positions that have purchased advertisements to the company's positions, that is, the stronger the willingness to purchase advertisements. Data analysis of multiple units of different company natures shows that non-profit organizations, private companies and other units have a high proportion of advertising purchases, that is, the willingness to purchase advertisements is strong. Therefore, the analysis report can show the following content: for example, when the company size is in the first interval, the willingness to purchase advertisements is strong; when the company size is in the second interval, the willingness to purchase advertisements is weak; when the company is a non-profit organization or a private company, the willingness to purchase advertisements is strong.

[0111] Based on the same technical concept as the method for predicting job purchase advertisements provided in the above embodiment, the present application also provides a device for predicting job purchase advertisements.

[0112] Figure 5 A schematic diagram of a structure of a prediction device for job purchase advertisements provided in an embodiment of the present application. Figure 5 As shown, the prediction device 50 for job purchase advertisements provided in the embodiment of the present application may include the following modules:

[0113] An acquisition module 501 is used to acquire multiple characteristic data of a target position;

[0114] A first prediction module 502 is used to input multiple feature data into a trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position, where the first prediction result is a prediction of whether the target position will purchase an advertisement;

[0115] The second prediction module 503 is used to input the multiple feature data into the trained deep neural network model to obtain a second prediction result of the target position, where the second prediction result is the probability of purchasing an advertisement for the target position;

[0116] A preliminary determination module 504 is used to combine the first prediction result and the second prediction result to preliminarily determine whether the target position will purchase an advertisement;

[0117] Verification module 505 is used to screen out first target feature data corresponding to the target feature from multiple feature data if it is preliminarily determined that the target position will purchase advertisements, and verify whether the first target feature data is consistent with the second target feature data for purchasing advertisements; wherein the first target feature data corresponds one-to-one to the second target feature data, the target feature is an important feature affecting the penetration rate of position advertisements determined by a lightweight gradient boosting machine learning model based on feature data of multiple sample positions, and the second target feature data is feature data corresponding to the target feature of at least one sample position that purchases advertisements; if they are consistent, it is determined that the target position will purchase advertisements.

[0118] The prediction device for job purchase advertisements provided by the embodiment of the present application combines the first prediction result of the target job by the lightweight gradient boosting machine learning model and the second prediction result of the target job by the deep neural network model. The two verify each other, and can realize the accurate prediction of whether the target job will purchase advertisements. On this basis, the first target feature data of the target job is verified by the second target feature data of the important feature that affects the penetration rate of job advertisements, which is equivalent to further verifying the prediction result of whether the target job will purchase advertisements from the perspective of the penetration rate of job advertisements, and can further improve the accuracy of the prediction result of whether the target job will purchase advertisements. It can not only provide a reference for users who publish target jobs to purchase advertisements, which helps to improve the return rate of purchasing advertisements, but also provide a reference for recommenders, which helps to help recommenders accurately find users / jobs with the need to purchase advertisements, improve the accuracy and efficiency of recommending purchase advertisements, reduce the disturbance to other users, and improve the user experience. In addition, the first target feature data and the second target feature data of the target feature are conducive to providing data guidance for recommenders, helping recommenders to select suitable jobs for advertisement recommendation according to their needs.

[0119] According to some embodiments of the present application, optionally, the verification module 505 can also be used to screen out first target feature data corresponding to the target feature from multiple feature data if it is preliminarily determined that the target position will not purchase advertisements, and verify whether the first target feature data is consistent with third target feature data that has not purchased advertisements; wherein the first target feature data corresponds one-to-one to the third target feature data, and the third target feature data is feature data corresponding to the target feature of at least one sample position that has not purchased advertisements; if they are consistent, it is determined that the target position will not purchase advertisements; if they are inconsistent, a prompt message is sent to the target terminal device.

[0120] According to some embodiments of the present application, optionally, the prediction device 50 for job purchase advertisements provided in the embodiments of the present application may further include a training module for obtaining a training set and a test set, each of which includes multiple sample positions, and each sample position includes multiple feature data; a lightweight gradient boosting machine learning model is trained based on the training set, and when the accuracy of the lightweight gradient boosting machine learning model predicting the purchase advertisements of the training set is greater than or equal to a preset threshold, or when the number of iterations reaches a preset threshold, the lightweight gradient boosting machine learning model is verified based on the test set; when the accuracy of the lightweight gradient boosting machine learning model predicting the purchase advertisements of the test set is greater than or equal to the preset threshold, a trained lightweight gradient boosting machine learning model is obtained; a feature importance ranking diagram is output based on the trained lightweight gradient boosting machine learning model, and multiple features corresponding to multiple feature data in the feature importance ranking diagram are ranked according to importance; the top N ranked features are selected as target features, and feature data corresponding to the target features of at least one sample position that purchases advertisements are selected as second target feature data, where N is a positive integer.

[0121] According to some embodiments of the present application, optionally, the target feature includes at least a first target feature having a positive impact on the purchase of advertisements, and the first target feature is determined based on a honeycomb diagram output by a trained lightweight gradient boosting machine learning model.

[0122] According to some embodiments of the present application, optionally, the characteristic data of the sample positions in the test set are characteristic data of the sample positions on weekdays, the training set includes positive sample positions that purchased advertisements and negative sample positions that did not purchase advertisements, the negative sample positions are randomly selected from multiple negative sample positions, and the difference between the number of positive sample positions and the number of negative sample positions in the training set is less than a preset difference.

[0123] According to some embodiments of the present application, optionally, the multiple characteristic data of the target position include at least one of the following: recent behavioral characteristic data of the recruiter, characteristic data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the exhibition-to-investment ratio of the target position, the average daily number of submissions for the target position, the functional advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, the average daily number of exposures of the target position under the work location-function grid, the exhibition-to-investment ratio of the work location-function grid, the advertising exhibition-to-investment ratio of the work location-function grid, and the difference between the advertising exhibition-to-investment ratio and the natural exhibition-to-investment ratio of the work location-function grid.

[0124] The first prediction module 502 can be specifically used to input the first part of the feature data among the multiple feature data into the trained lightweight gradient boosting machine learning model to obtain the first prediction result of the target position. The first part of the feature data includes at least one of the following: recent behavioral feature data of the recruiter, feature data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum education level of the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, and the average daily exposure number of the target position under the work location-function grid.

[0125] According to some embodiments of the present application, optionally, the second prediction module 503 can be specifically used to input multiple feature data into a trained deep neural network model to obtain a second prediction result for the target position when the target position is an existing position in the recruitment platform; when the target position is a new position, input other feature data among the multiple feature data except the investment-to-expense ratio of the target position into the trained deep neural network model to obtain a second prediction result for the target position.

[0126] Figure 5 Each module / unit in the device shown has the function of implementing each step in the method for predicting job purchase advertisements provided in the above method embodiment and can achieve its corresponding technical effect, which will not be described in detail here for the sake of brevity.

[0127] The electronic device in the embodiment of the present application may be a user terminal device, a server, other computing devices, or a cloud server. Figure 6 This is a hardware structure diagram of an electronic device of an embodiment of the present application. The electronic device may include a processor 601 and a memory 602 storing computer program instructions. When the processor 601 executes the computer program instructions, it implements the process or function of any of the above-mentioned embodiment methods.

[0128] Specifically, the processor 601 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application. The memory 602 may include a large-capacity memory for data or instructions. For example, the memory 602 may be at least one of the following: a hard disk drive (HDD), a read-only memory (ROM), a random access memory (RAM), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, a universal serial bus (USB) drive or other physical / tangible memory storage device. For another example, the memory 602 may include a removable or non-removable (or fixed) medium. For another example, the memory 602 may be inside or outside the integrated gateway disaster recovery device. The memory 602 may be a non-volatile solid-state memory. In other words, the memory 602 usually includes a tangible (non-transitory) computer-readable storage medium (such as a memory device) encoded with computer-executable instructions, and when the software is executed (such as executed by one or more processors), the operation described in the method of the embodiment of the present application can be performed. The processor 601 implements the process or function of any method in the above embodiments by reading and executing the computer program instructions stored in the memory 602.

[0129] In one example, Figure 6 The electronic device shown may also include a communication interface 603 and a bus 610. Among them, the processor 601, the memory 602, and the communication interface 603 are connected through the bus 610 and complete the communication between each other. The communication interface 603 is mainly used to realize the communication between each module, device, unit and / or device in the embodiment of the present application. The bus 610 includes hardware, software or both, and can couple the components of the online data traffic billing device to each other. For example, the bus may include at least one of the following: an accelerated graphics port (AGP) or other graphics bus, an enhanced industrial standard architecture (EISA) bus, a front-end bus (FSB), a hypertransport (HT) interconnect, an industrial standard architecture (ISA) bus, an infinite bandwidth interconnect, a low pin count (LPC) bus, a memory bus, a micro channel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standard association local (VLB) bus or other suitable buses. The bus 610 may include one or more buses. Although the embodiments of the present application describe or illustrate a specific bus, the embodiments of the present application may consider any suitable bus or interconnection method.

[0130] In combination with the method in the above embodiments, an embodiment of the present application also provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the process or function of any method in the above embodiments is implemented.

[0131] In addition, an embodiment of the present application further provides a computer program product, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the process or function of any one of the methods in the above embodiments is implemented.

[0132] The above exemplarily describes the flowcharts and / or block diagrams of the methods, devices, systems and computer program products of the embodiments of the present application, and describes the relevant various aspects. It should be understood that each box or combination thereof in the flowchart and / or block diagram can be implemented by computer program instructions, or by dedicated hardware that performs a specified function or action, or by a combination of dedicated hardware and computer instructions. For example, these computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to form a machine that enables these instructions executed by such a processor to enable the implementation of the functions / actions specified in each box or combination thereof in the flowchart and / or block diagram. Such a processor can be a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit.

[0133] The functional blocks shown in the structured block diagram of the embodiment of the present application can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc.; when implemented in software, it is a program or code segment used to perform the required task. The program or code segment can be stored in a memory, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0134] It should be noted that the present application is not limited to the specific configurations and processes described above or shown in the figures. The above is only a specific implementation mode of the present application. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the described system, device, module or unit can refer to the corresponding process in the method embodiment without further description. It should be understood that the scope of protection of the present application is not limited to this. Any technician familiar with the technical field can think of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and these modifications or substitutions should be included in the scope of protection of the present application.

Claims

1. A method for predicting job purchase advertisements, characterized in that: include: Obtain multiple characteristic data of the target position; Input multiple feature data into the trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position, which is a prediction of whether the target position will purchase advertisements; Inputting the multiple feature data into the trained deep neural network model to obtain a second prediction result of the target position, where the second prediction result is a probability of purchasing an advertisement for the target position; Combining the first prediction result and the second prediction result, preliminarily determining whether the target position will purchase advertisements; If it is preliminarily determined that the target position will purchase advertisements, the first target feature data corresponding to the target feature is screened out from multiple feature data, and it is verified whether the first target feature data is consistent with the second target feature data for purchasing advertisements; wherein the first target feature data corresponds one-to-one to the second target feature data, the target feature is an important feature affecting the penetration rate of position advertisements determined by a lightweight gradient boosting machine learning model based on feature data training of multiple sample positions, and the second target feature data is feature data corresponding to the target feature of at least one sample position that purchases advertisements; if they are consistent, it is determined that the target position will purchase advertisements.

2. The method according to claim 1, characterized in that The method further comprises: If it is preliminarily determined that the target position will not purchase advertisements, the first target feature data corresponding to the target feature is screened out from multiple feature data, and it is verified whether the first target feature data is consistent with the third target feature data that did not purchase advertisements; wherein the first target feature data corresponds one-to-one to the third target feature data, and the third target feature data is feature data corresponding to the target feature of at least one sample position that did not purchase advertisements; if they are consistent, it is determined that the target position will not purchase advertisements; if they are inconsistent, a prompt message is sent to the target terminal device.

3. The method according to claim 1, characterized in that Before inputting the plurality of feature data into the trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position, the method further includes: Obtaining a training set and a test set, wherein each of the training set and the test set includes a plurality of sample positions, and each sample position includes a plurality of feature data; The lightweight gradient boosting machine learning model is trained based on the training set, and when the accuracy of the lightweight gradient boosting machine learning model in predicting the purchase of advertisements in the training set is greater than or equal to a preset threshold, or when the number of iterations reaches a preset threshold, the lightweight gradient boosting machine learning model is verified based on the test set; When the accuracy of the lightweight gradient boosting machine learning model in predicting the purchase of advertisements in the test set is greater than or equal to a preset threshold, a trained lightweight gradient boosting machine learning model is obtained; A feature importance ranking graph is output based on the trained lightweight gradient boosting machine learning model, in which multiple features corresponding to multiple feature data are ranked according to importance; The first N features in the ranking are selected as target features, and feature data corresponding to the target features of at least one sample position of the purchased advertisement is selected as second target feature data, where N is a positive integer.

4. The method according to claim 3, characterized in that The target features include at least a first target feature that has a positive impact on the purchase of advertisements, and the first target feature is determined based on a honeycomb graph output by a trained lightweight gradient boosting machine learning model.

5. The method according to claim 3, characterized in that: The characteristic data of the sample positions in the test set are the characteristic data of the sample positions on weekdays. The training set includes positive sample positions that purchased advertisements and negative sample positions that did not purchase advertisements. The negative sample positions are randomly selected from multiple negative sample positions. The difference between the number of positive sample positions and the number of negative sample positions in the training set is less than a preset difference.

6. The method according to claim 1, characterized in that The multiple characteristic data of the target position include at least one of the following: recent behavioral characteristic data of the recruiter, characteristic data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the exposure ratio of the target position, the average daily number of submissions of the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, the average daily number of exposures of the target position in the work location-function grid, the exposure ratio of the work location-function grid, the advertising exposure ratio of the work location-function grid, and the difference between the advertising exposure ratio and the natural exposure ratio of the work location-function grid; Input multiple feature data into the trained lightweight gradient boosting machine learning model to obtain the first prediction result of the target position, including: Inputting a first part of the feature data from the plurality of feature data into the trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position; Among them, the first part of characteristic data includes at least one of the following: recent behavioral characteristic data of recruiters, characteristic data of the company where the target position is located, the function of the target position, the work location of the target position, the average salary level of the target position, the minimum educational level of the target position, the function advertising penetration rate, the work location advertising penetration rate, the company industry advertising penetration rate, and the average daily exposure number of the target position under the work location-function grid.

7. The method according to claim 6, characterized in that Input multiple feature data into the trained deep neural network model to obtain the second prediction result of the target position, including: When the target position is an existing position in the recruitment platform, multiple feature data are input into the trained deep neural network model to obtain a second prediction result of the target position; When the target position is a new position, the other feature data except the investment-to-expansion ratio of the target position among the multiple feature data are input into the trained deep neural network model to obtain a second prediction result of the target position.

8. A prediction device for job purchase advertisements, characterized in that: include: An acquisition module, used to acquire multiple characteristic data of a target position; A first prediction module is used to input multiple feature data into a trained lightweight gradient boosting machine learning model to obtain a first prediction result of the target position, where the first prediction result is to predict whether the target position will purchase an advertisement; A second prediction module is used to input multiple feature data into the trained deep neural network model to obtain a second prediction result of the target position, where the second prediction result is a probability of purchasing an advertisement for the target position; A preliminary determination module, used to combine the first prediction result and the second prediction result to preliminarily determine whether the target position will purchase an advertisement; A verification module is used to screen out first target feature data corresponding to the target feature from multiple feature data if it is preliminarily determined that the target position will purchase advertisements, and verify whether the first target feature data is consistent with the second target feature data for purchasing advertisements; wherein the first target feature data corresponds one-to-one to the second target feature data, the target feature is an important feature affecting the penetration rate of position advertisements determined by a lightweight gradient boosting machine learning model based on feature data training of multiple sample positions, and the second target feature data is feature data corresponding to the target feature of at least one sample position that purchases advertisements; if they are consistent, it is determined that the target position will purchase advertisements.

9. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; when the electronic device executes the computer program instructions, the method for predicting a job purchase advertisement according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method for predicting job purchase advertisements according to any one of claims 1 to 7 is implemented.

Images