Method and apparatus for recommending job information based on job search expectation transfer model

Abstract

The application discloses a generation method of a job-hunting expectation transfer model, a position information recommendation method and devices thereof. The generation method comprises the following steps: constructing a to-be-trained model; the to-be-trained model comprises a transformer structure and a multilayer perceptron; vector data of multiple conversion sequences output by the transformer structure is spliced and used as input of the multilayer perceptron; a training sample set is constructed; the training sample set comprises multiple groups; each group of the training sample set comprises position information and location information; the location information is a sequence position of multiple positions to which a job seeker has delivered a job application in a sequence according to a time interval between a delivery time and a current time; the to-be-trained model is trained by using the training sample set, so that a trained job-hunting expectation transfer model is obtained; and the job-hunting expectation transfer model is used for predicting delivery sequence data of a current job seeker and related scores of multiple to-be-predicted positions. The trained job-hunting expectation transfer model can predict an expectation change of the current job seeker, and is favorable for improving the accuracy of position information recommendation.

Description

Technical Field

[0001] This application relates to the technical fields of computers and the Internet, and in particular to a method for generating a job expectation transfer model, a method for recommending job information, an apparatus, an electronic device, and a storage medium. Background Technology

[0002] In the recruitment field, candidates' expectations can change to varying degrees throughout the job search cycle, including resume writing, job searching, application submission, and interviews. For example, initially, they might expect and apply for high salaries and preferred positions. However, as feedback and interview results become less than ideal, these initial expectations may gradually decrease, leading to lower salaries, changes in the city of work, or even different job titles. Current job matching and recommendation systems rely solely on user-provided expectations, failing to accurately capture and understand these shifts in expectations. This can result in low accuracy in job matching and recommendations. Summary of the Invention

[0003] The purpose of this application is to provide a method for generating a job expectation transfer model, a method for recommending job information, an apparatus, an electronic device, and a storage medium.

[0004] This application provides a method for generating a job-seeking expectation transfer model, including:

[0005] Construct a model to be trained; the model to be trained includes a transformer structure and a multilayer perceptron. The vector data of multiple transformation sequences output by the transformer structure are concatenated and used as the input of the multilayer perceptron.

[0006] Construct a training sample set; the training sample set includes multiple groups, each group of training sample sets includes job information and location information; the location information is the sequence position of multiple job applications ordered in order of application time from the current time.

[0007] The constructed model to be trained is trained using a training sample set to obtain a well-trained job expectation transfer model; the job expectation transfer model is used to predict the current job seeker's application sequence data and the relevant scores of multiple positions to be predicted.

[0008] Furthermore, the step of training the constructed model to be trained using a training sample set to obtain a trained job expectation transfer model also includes:

[0009] Extract the various influencing factors and location information of job information for each sample set;

[0010] Map each influencing factor in the job information to a first-class feature vector of the corresponding dimension;

[0011] The location information is mapped to a second type of feature vector with corresponding dimensions;

[0012] The first-class feature vectors and the second-class feature vectors are concatenated to obtain the concatenated first-class vector data.

[0013] The concatenated first-class vector data is input into the transformer structure, and the transformer structure is trained to output vector data of multiple transformation sequences.

[0014] The vector data of multiple transformation sequences are concatenated to obtain the concatenated second type of vector data;

[0015] The concatenated second-class vector data is input into a multilayer perceptron to train the multilayer perceptron and output the first correlation score between the target position and the multiple positions applied for.

[0016] Further, mapping the location information into a second type of feature vector of corresponding dimensions includes:

[0017] The vector PE at position t can be calculated using the following formula:

[0018]

[0019] Where: d is the vector dimension; k takes the value of (vector dimension d / 2-1);

[0020] When i = 2k, the vector is in an even position; when i = 2k+1, the vector is in an odd position.

[0021] Furthermore, the step of training the constructed model to be trained using the training sample set to obtain a trained job expectation transfer model further includes:

[0022] Input the relevant scores into the cross-entropy loss function to calculate the loss value;

[0023] Based on the calculated loss value, an optimization algorithm is used to calculate and update the gradient, and backpropagation is performed to update the parameters of the job expectation transfer model and the underlying first-class and second-class feature vector data; the loss function is shown in the following conditional expression:

[0024]

[0025] In the formula, f(x) i ) represents the first relevance score; when the training samples are positive samples, y i The value is 1; when the training samples are negative samples, y i 0; m is the total number of sample groups in the sample set.

[0026] Furthermore, job information includes one or more of the following influencing factors: job ID, function, industry, city of work, salary range, and the time difference between the nth job and the target job application.

[0027] This application provides a method for recommending job information, based on a job expectation transfer model generated by the method, including:

[0028] Obtain the second historical job posting data of the current job seeker within the second time period; the second historical job posting data includes job information for multiple job postings;

[0029] The second historical job posting data is sorted sequentially from the current time to the front to obtain the second posting sequence data; wherein, the second posting sequence data includes the position information of multiple job postings sorted in order of their closest time to the current time;

[0030] Obtain multiple job postings to be predicted;

[0031] Multiple job postings to be predicted and second application sequence data are input into a trained job expectation transfer model to obtain a prediction relevance score for each job posting to be predicted.

[0032] Recommended job information is obtained based on the prediction-related score of each job to be predicted.

[0033] Further, the step of obtaining recommended job information based on the prediction relevance score of each job to be predicted includes:

[0034] The prediction relevance scores of each job to be predicted are sorted in descending order, and the job to be predicted corresponding to the highest prediction relevance score is taken as the recommended job information.

[0035] Furthermore, the acquisition of multiple job titles to be predicted includes:

[0036] Obtain the jobs that job seekers have clicked and / or viewed, and select the most recent clicked and / or viewed jobs as the jobs to be predicted; or...

[0037] Obtain job postings from recruiters, and select the most recent posted positions as candidates for prediction; or...

[0038] We use a recall approach to identify positions with high application rates from the past job applications of current job seekers and select these positions for prediction.

[0039] This application provides an apparatus for generating a job-seeking expectation transfer model, comprising:

[0040] The first building module is used to build the model to be trained; the model to be trained includes a transformer structure and a multilayer perceptron, and the vector data of multiple transformation sequences output by the transformer structure are concatenated and used as the input of the multilayer perceptron.

[0041] The second construction module is used to construct a training sample set; the training sample set includes multiple sets, each set including job information and location information; the location information is the sequence position of multiple job applications ordered in order of application time from the current time.

[0042] The training module is used to train the constructed model to be trained using the training sample set to obtain a trained job expectation transfer model; wherein,

[0043] The job expectation transfer model is used to predict the current job seeker's application sequence data and the relevant scores of multiple positions to be predicted.

[0044] Furthermore, the training module includes:

[0045] The extraction unit is used to extract the various influencing factors and location information of job information in each sample set;

[0046] The first mapping unit is used to map each influencing factor in the job information into a first-class feature vector of the corresponding dimension.

[0047] The second mapping unit is used to map the location information into a second type of feature vector with a corresponding dimension;

[0048] The first concatenation unit is used to concatenate the first type feature vectors and the second type feature vectors to obtain the concatenated first type vector data.

[0049] The first training unit is used to input the concatenated first-class vector data into the transformer structure, train the transformer structure, and output vector data of multiple transformation sequences.

[0050] The second concatenation unit is used to concatenate vector data from multiple transformation sequences to obtain concatenated second-class vector data.

[0051] The second training unit is used to input the concatenated second-class vector data into the multilayer perceptron to train the multilayer perceptron and output the first correlation score between the target position and the multiple positions applied for.

[0052] This application provides a job information recommendation device, which, based on a job expectation transfer model generated by the method described above, includes:

[0053] The second acquisition module is used to acquire the second historical job application data of the current job seeker within the second time period; wherein, the second historical job application data includes job information of multiple job applications submitted;

[0054] The second sequence module is used to sort the second historical job posting data in order from the latest to the earliest time to obtain the second delivery sequence data; wherein, the second delivery sequence data includes the position information of multiple job postings sorted in order of the closest time to the current time;

[0055] The third acquisition module is used to acquire multiple job positions to be predicted.

[0056] The input module is used to input multiple job postings to be predicted and second application sequence data into the trained job expectation transfer model to obtain the prediction relevance score for each job posting.

[0057] The recommendation module is used to obtain recommended job information based on the prediction relevance score of each job to be predicted.

[0058] This application provides an electronic device, which includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the steps of the method described above.

[0059] This application provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the steps of the method described above.

[0060] This application provides a computer program product, which includes computer program instructions that, when executed by a processor, implement the steps of the method described above.

[0061] The above-mentioned technical solution of this application has the following beneficial technical effects:

[0062] In this embodiment, there is a significant relationship between job seekers' expectation shifts and time intervals in the recruitment field. These shifts can be reflected in changes to any element of job information, such as job ID, job function, industry, work city, salary range, and benefits. By providing positive and negative samples to train the model, the model can learn the feature changes in the training samples, thus obtaining a well-trained job expectation shift model. This model can predict the current job seeker's expectation changes, which helps improve the accuracy of job information recommendations and increase the matching degree between job seekers and available positions. Therefore, time interval information is introduced into the training sample sequence information to allow the model to better learn and characterize the impact of time factors on expectation shifts. Introducing location information can characterize the order in which job seekers submit job applications, thereby obtaining high-quality training samples and optimizing the model training effect. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings of the embodiments of this application will be briefly described below.

[0064] Figure 1 This is a schematic diagram of the system architecture of an embodiment of this application.

[0065] Figure 2 This is a flowchart illustrating a method for constructing training samples according to an embodiment of this application.

[0066] Figure 3 This is a flowchart illustrating a method for generating a job expectation transfer model according to an embodiment of this application.

[0067] Figure 4 This is a flowchart of a job information recommendation method according to an embodiment of this application.

[0068] Figure 5 This is a schematic diagram illustrating the processing procedure of the sample sequence in an embodiment of this application.

[0069] Figure 6 This is a schematic diagram of the training process of the job expectation transfer model in an embodiment of this application.

[0070] Figure 7 This is a structural block diagram of a training sample construction device according to an embodiment of this application.

[0071] Figure 8 This is a structural block diagram of a job-seeking expectation transfer model generation device according to an embodiment of this application.

[0072] Figure 9 This is a structural block diagram of a job information recommendation device according to an embodiment of this application.

[0073] Figure 10This is a schematic diagram of an electronic device used to implement the training sample construction method, the job expectation transfer model generation method, and the job information recommendation method of the embodiments of this application. Detailed Implementation

[0074] The principles and spirit of this application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided to make the principles and spirit of this application clearer and more thorough, enabling those skilled in the art to better understand and implement the principles and spirit of this application. The exemplary embodiments provided herein are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments described herein without inventive effort are within the scope of protection of this application.

[0075] Embodiments of this application relate to terminal devices and / or servers. Those skilled in the art will understand that embodiments of this application can be implemented as a system, apparatus, device, method, computer-readable storage medium, or computer program product. Therefore, this disclosure can be specifically implemented in at least one of the following forms: entirely hardware, entirely software, or a combination of hardware and software. According to embodiments of this application, this application claims protection for a method and apparatus for constructing training samples, a method and apparatus for generating a job expectation transfer model, a method and apparatus for recommending job information, electronic devices, computer-readable storage media, and computer program products. Figure 1 A schematic diagram of a system architecture according to an embodiment of this application is shown. For example... Figure 1 As shown, the system includes a terminal device 102 and a server 104. The terminal device 102 can include at least one of the following: a smartphone, tablet, laptop, desktop computer, smart TV, various wearable devices, augmented reality (AR) devices, virtual reality (VR) devices, etc. A client can be installed on the terminal device 102. For example, the client can be a client specifically designed to perform a particular function (such as an application app), or a client with multiple embedded application applets (with different functions), or a client logged in via a browser. Users can operate on the terminal device 102. For example, a user can open the client installed on the terminal device 102 and input commands through the client, or a user can open the browser installed on the terminal device 102 and input commands through the browser. After receiving the user's input command, the terminal device 102 sends a request message containing the command to the server 104. After receiving the request message, the server 104 performs the corresponding processing and then returns the processing result information to the terminal device 102. The user command is completed through a series of data processing and information interaction.

[0076] In this document, terms such as first, second, and third are used only to distinguish one entity (or operation) from another, and are not intended to require or imply any order or relationship between these entities (or operations).

[0077] The following is a brief description of the concepts and technical terms that may be involved in the embodiments of this application.

[0078] Function: The job positions applied for by job seekers can be classified into three levels, such as R&D / Algorithm Engineer / Recommendation Algorithm Engineer. Among them, Recommendation Algorithm Engineer is a third-level category, which can be defined as "Function"; R&D is a first-level category, which can be defined as "Industry".

[0079] Figure 2 The diagram illustrates a flowchart of a method for constructing training samples according to an embodiment of this application. The method includes the following steps:

[0080] S110: Obtain the first historical job posting data of a job seeker within the first time period; wherein, the first historical job posting data includes job information of multiple job postings.

[0081] Specifically, the first time period can be set as a predetermined cycle, such as a job search lifecycle, which is the entire process from when a job seeker starts looking for a job to when they go through an interview. When a job seeker browses the job description of a job posted by a recruiter on terminal device 102, the job description may include job ID, function, industry, work city, salary range, and benefits. The job seeker can click the "Submit Resume" button to send their resume to the recruiter for a job that matches their job preferences. Therefore, by obtaining the resumes sent by each job seeker after browsing different job descriptions on terminal device 102 on the recruitment platform, the job information of multiple jobs that the job seeker has applied for can be obtained. For example, if the first time period is set to two months, the corresponding historical job application data of different job seekers can be obtained.

[0082] S120: Sort the first historical job posting data in order from the current time to the front to obtain the first posting sequence data; wherein, the first posting sequence data includes the position information of multiple job postings sorted in order from the current time to the front of the posting time.

[0083] Specifically, the job information of each job seeker who has applied for multiple positions is sorted in order from the latest to the earliest time from the current time to obtain the first application sequence data. For example, if a job seeker applied for a total of 100 positions in two months, the last job applied for can be identified as sequence number "1". According to the order of application time from the latest to the earliest time from the current time, the earliest job applied for can be identified as sequence number "100", and so on for other positions. The location information is correspondingly the position identification sequence number of the multiple job applications.

[0084] S130: The first delivery sequence data is divided into multiple positive sample sets according to a preset segmentation strategy; wherein each positive sample set includes a target position and n submitted positions, the target position is the position in each positive sample set whose submission time is closest to the current time; n is greater than or equal to 2.

[0085] Specifically, each positive sample set can be set as a part of the first submission sequence data. That is, the first submission sequence data is a long sequence set, and each positive sample set is a short sequence set in the long sequence set. Each positive sample set may include a target position and n submitted positions. The target position is the position in each positive sample set whose submission time is closest to the current time. The n submitted positions are other positions whose submission time is slightly later than the submission time of the target position and are sorted in order from the latest to the earliest from the current time.

[0086] S140: Randomly select multiple positions from the delivery sequence data to obtain pre-selected positions; wherein, the pre-selected positions are non-target positions.

[0087] S150: Replace the target job in each positive sample set with the pre-selected job to obtain a corresponding set of multiple negative samples.

[0088] Specifically, randomly selected job titles replace the target job titles in each positive sample set, while the sequence of the other n submitted job titles remains unchanged, to obtain the corresponding negative sample set. The number of negative sample sets to positive sample sets can be generated in a 1:10 ratio. By providing positive and negative samples, the model to be trained, such as a neural network model, learns to distinguish which features are associated with the category of interest and which are not. This helps improve the model's discriminative ability and robustness.

[0089] S160: Obtain the training sample set based on the positive sample set and the negative sample set. Specifically, the positive sample set and the negative sample set can be shuffled to generate a random order to obtain the training sample set.

[0090] In this embodiment, there is a significant relationship between job seekers' expectation shifts and time intervals in the recruitment field. These shifts can be reflected in changes to any element of job information, such as job ID, job function, industry, work city, salary range, and benefits. By providing positive and negative samples to train the model, the model can learn the feature changes in the training samples, thus obtaining a well-trained job expectation shift model. This model can predict the current job seeker's expectation changes, which helps improve the accuracy of job information recommendations and increase the matching degree between job seekers and available positions. Therefore, time interval information is introduced into the training sample sequence information to allow the model to better learn and characterize the impact of time factors on expectation shifts. Introducing location information can characterize the order in which job seekers submit job applications, thereby obtaining high-quality training samples and optimizing the model training effect.

[0091] In some embodiments, step S130, which involves dividing the first delivery sequence data into multiple sets of positive samples according to a preset segmentation strategy, may include the following specific steps:

[0092] S131: Use a set sliding window to sequentially capture the first delivery sequence data to obtain multiple sets of sequence samples;

[0093] S132: Mark the job corresponding to the sequence in each set of sequence samples as a positive sample, and take the job corresponding to the first number in each set of sequence samples as the target job.

[0094] Specifically, the sliding window can include multiple application sequence numbers, meaning a certain number of samples can be set in the sliding window. Since the job information of each job seeker's multiple applications is sorted in order from the latest to the earliest time, the set sliding window is used to sequentially extract the first application sequence data to obtain multiple sets of sequence samples that can be labeled as positive samples. The target job is the last job applied for in each set of sequence samples. In this way, time interval information is introduced into the training samples so that the model can better learn and characterize the impact of time factors on the expected transfer.

[0095] In some embodiments, step S130, which involves dividing the first delivery sequence data into multiple sets of positive samples according to a preset segmentation strategy, may further include the following specific steps:

[0096] S133: If the number of samples in the current sequence does not meet the number of samples set in the sliding window, fill the end of the sequence of the current sequence with "0" so that the number of samples in the current sequence meets the number of samples set in the sliding window.

[0097] In the exemplary embodiment, job information includes one or more of the following influencing factors: job ID, job function, industry, work city, salary range, and the time difference between the nth job and the target job application. During the job search process, a job seeker's job ID, job function, industry, work city, salary range, etc., may change over time. Introducing these influencing factors into the training samples can characterize the trajectory changes of a job seeker's expected migration.

[0098] This application provides a method for generating a job-seeking expectation transfer model, based on a training sample set constructed using the method described above, such as... Figure 3 As shown, the specific steps may include the following:

[0099] S210: Construct the model to be trained; wherein, the model to be trained includes a transformer structure and a multilayer perceptron, and the vector data of multiple transformation sequences output by the transformer structure are concatenated and used as the input of the multilayer perceptron;

[0100] S230: The training sample set is used to train the constructed model to obtain the trained job expectation transfer model; wherein, the job expectation transfer model is used to predict the current job seeker's application sequence data and the relevant scores of multiple positions to be predicted.

[0101] Specifically, in the field of deep learning, the Transformer architecture is a model architecture used to process sequential data. The Transformer architecture can include multi-head self-attention, positional encoding, feedforward neural networks, and residual connections and layer normalization. Multi-head self-attention can characterize and learn the relationships between job sequences, better capturing the multiple relationships and subtle differences between elements in each job sequence. Positional encoding provides location information within the sequence, and a fully connected feedforward neural network performs parallel computation on each position. Residual connections are used between Transformer layers, and layer normalization is applied before or after each sub-layer to enable training deeper network structures. The vector data of multiple transformed sequences output by the Transformer architecture are concatenated and used as input to a multilayer perceptron (MLP). The MLP is a feedforward neural network model that can... It includes an input layer, one or more hidden layers, and an output layer. After learning and training with a multilayer perceptron (MLP), it can output the correlation scores between the target job and other jobs in each training sample based on the activation function. The trained job expectation transfer model can predict the current job seeker's application sequence data and the correlation scores of multiple jobs to be predicted. The corresponding predicted correlation scores of each job to be predicted are sorted from largest to smallest, thereby obtaining recommended job information. That is, based on the correlation scores, it can be determined which job the job seeker is most likely to apply for after expectation transfer. This can accurately capture and understand the information of job seeker's expectation changes and transfers, which is of great help for optimizing job matching and job recommendation. In this embodiment, there is a significant relationship between job seekers' expectations shift and time intervals in the recruitment field. The shift in job seekers' expectations can be reflected in any element of job information such as job ID, function, industry, work city, salary range, and benefits. A well-trained job expectation shift model can learn the feature changes in the training samples. In this way, time interval information is introduced into the training sample sequence information so that the model can better learn and characterize the impact of time factors on expectation shift. Introducing location information can characterize the order in which job seekers submit job applications. The job expectation shift model can predict the current job seeker's expectation changes, which is beneficial to improving the accuracy of job information recommendations and increasing the matching degree between job seekers and the positions to be recruited.

[0102] In this embodiment, the training sample set constructed through steps S110-S160 of the above method may include multiple sets of positive samples and negative samples. Each set of positive samples includes a target position and n applied positions. The target position is the position in each set of positive samples whose application time is closest to the current time. The n applied positions are ordered in reverse order of their application time from the current time; n is greater than or equal to 2. Correspondingly, each set of negative samples includes a pre-selected position and n applied positions. The pre-selected position is used to replace the target position in each set of positive samples. The pre-selected position is a position randomly selected from the n applied positions in each set of positive samples. In other embodiments, the method for generating the job expectation transfer model may further include the following steps:

[0103] S220: Construct a training sample set; wherein the training sample set includes multiple groups, each group including job information and location information; the location information is the sequence position of multiple job applications ordered chronologically from the application time to the current time. Specifically, the training sample set can also be constructed based on job seekers' behavior to build positive and negative samples. For example, job seekers' behavior may include searching for keywords, browsing exposed jobs, clicking on jobs, saving jobs, applying for jobs, etc. The series of behaviors after a job seeker enters search keywords can be analyzed, with the job ultimately applied for by the job seeker as a positive sample, and the exposed but not applied for and / or not clicked jobs as negative samples. These positive and negative samples reflect the job seeker's selection tendency to a certain extent and can be used as training samples for the model to be trained. Furthermore, the training sample set can also be constructed based on job seekers' resume submission behavior. Multiple job postings can be sorted sequentially according to their submission time relative to the current time to obtain a corresponding submission sequence. A predetermined number of jobs can be sequentially extracted from this sequence to form multiple training sample sets for the model to be trained. Each training sample set can include job information and position information within the submission sequence. It is understood that the training sample set can also be constructed in other ways, which will not be listed here.

[0104] In some embodiments, step S230, which involves training the constructed model to be trained using the training sample set to obtain a trained job-seeking expectation transfer model, may include the following specific steps:

[0105] S231: Extract the various influencing factors and location information of job information from each sample set;

[0106] S232: Map each influencing factor in the job information to a first-class feature vector of the corresponding dimension;

[0107] S233: Map the location information into a second type of feature vector with corresponding dimensions;

[0108] S234: Concatenate the first-class feature vectors and the second-class feature vectors to obtain the concatenated first-class vector data;

[0109] S235: Input the concatenated first-class vector data into the transformer structure, train the transformer structure to output vector data of multiple transformation sequences;

[0110] S236: Concatenate the vector data of multiple transformation sequences to obtain concatenated second-class vector data;

[0111] S237: Input the concatenated second-class vector data into the multilayer perceptron to train the multilayer perceptron and output the first correlation score between the target position and the multiple positions applied for.

[0112] Specifically, a linear mapping is applied to the underlying embedding before inputting the Transformer structure, which maps each vector to the same vector space.

[0113] In some embodiments, step S233, mapping the location information into a second type of feature vector of corresponding dimensions, includes:

[0114] The vector PE at position t can be calculated using the following formula:

[0115]

[0116] Where: d is the vector dimension; k takes the value of (vector dimension d / 2-1);

[0117] When i = 2k, the vector is in an even position; when i = 2k+1, the vector is in an odd position.

[0118] Specifically, k is the vector dimension index. For example, if the vector dimension d = 512, it can be grouped into pairs, resulting in 256 groups. k = 0 to 255. 2k corresponds to 0, 2, 4...510, which are even positions; 2k+1 corresponds to 1, 3, 5...511, which are odd positions.

[0119] In some embodiments, step S230, which involves training the constructed model to be trained using the training sample set to obtain a trained job-seeking expectation transfer model, further includes the following specific steps:

[0120] S238: Input the relevant scores into the cross-entropy loss function and calculate the loss value;

[0121] S239: Based on the calculated loss value, an optimization algorithm is used to calculate and update the gradient, and backpropagation is performed to update the parameters of the job expectation transfer model and the underlying first-class and second-class feature vector data; wherein, the loss function is shown in the following conditional expression:

[0122]

[0123] In the formula, f(x) i ) represents the first relevance score; when the training samples are positive samples, y i The value is 1; when the training samples are negative samples, y i 0; m is the total number of sample groups in the sample set.

[0124] This application provides a method for recommending job information, based on a job expectation transfer model generated by the method, such as... Figure 4 As shown, the specific steps may include the following:

[0125] S310: Obtain the second historical job posting data of the current job seeker within the second time period; wherein, the second historical job posting data includes job information of multiple job postings;

[0126] S320: Sort the second historical job posting data in reverse order from the current time to obtain the second posting sequence data; wherein, the second posting sequence data includes the position information of multiple job postings sorted in order of their closest time to the current time;

[0127] S330: Obtain multiple job postings to be predicted;

[0128] S340: Input multiple job postings to be predicted and second application sequence data into the trained job expectation transfer model to obtain the prediction relevance score for each job posting to be predicted;

[0129] S350: Based on the prediction relevance score of each job to be predicted, recommended job information is obtained.

[0130] Specifically, the second time period can be the same as the first time period, both set to a predetermined cycle, such as a two-month period. By obtaining the resumes submitted by job seekers on the recruitment platform after browsing different job descriptions on terminal device 102, the job information of the multiple positions applied for by the job seeker can be obtained. The job information of the multiple positions applied for by the current job seeker can be sorted sequentially from the latest to the earliest time to obtain the second application sequence data; for example, if the current job seeker applied for a total of 50 positions within two months, the last applied position can be identified as sequence number "1", and the earliest applied position can be identified as sequence number "50" in the latest to the earliest time, and so on for the other positions; the location information corresponds to the position identification sequence number of the multiple applied positions. Multiple job postings to be predicted and second application sequence data are input into a trained job expectation transfer model to obtain a prediction relevance score for each job posting. The prediction relevance scores for each job posting are sorted from highest to lowest to obtain recommended job information. That is, based on the relevance score, the most likely job posting for the job seeker after expectation transfer can be determined. This accurately captures and understands the information regarding job seeker expectation changes and transfers, greatly aiding in optimizing job-person matching and job recommendations. In this embodiment, job seeker expectation transfer and time intervals are significantly related in the recruitment field. Job seeker expectation transfer can be reflected in changes to any element of job information such as job ID, function, industry, work city, salary range, and benefits. The trained job expectation transfer model can learn from the feature changes in the training samples. This introduces time interval information into the training sample sequence information so that the model can better learn and characterize the impact of time factors on expectation transfer. Introducing location information can characterize the order in which job seekers submit job applications. The job expectation transfer model can predict the current job seeker's expectation changes, which helps improve the accuracy of job information recommendations and increases the matching degree between job seekers and the positions to be recruited.

[0131] In some embodiments, step S350, obtaining recommended job information based on the prediction relevance score of each job to be predicted, includes the following specific steps:

[0132] The prediction relevance scores of each job to be predicted are sorted in descending order, and the job to be predicted corresponding to the highest prediction relevance score is taken as the recommended job information.

[0133] In some embodiments, step S310, obtaining multiple job titles to be predicted, includes the following specific steps:

[0134] Obtain the jobs that job seekers have clicked and / or viewed, and select the most recent clicked and / or viewed jobs as the jobs to be predicted; or...

[0135] Obtain job postings from recruiters, and select the most recent posted positions as candidates for prediction; or...

[0136] We use a recall approach to identify positions with high application rates from the past job applications of current job seekers and select these positions for prediction.

[0137] The implementation methods and advantages of the embodiments of this application have been described above through multiple examples. Specific examples are provided below for reference. Figure 5 and Figure 6 The specific processing steps of the embodiments of this application are described in detail.

[0138] I. The process of constructing training samples:

[0139] S11. Retrieve historical job posting data from the past two months, extract the posting sequence for each job seeker, and sort them by posting time, with the most recent posting at the top. That is, the last job posting in the sequence is numbered jd1.

[0140] time i ≥time j

[0141] Here, jd1 is the identifier number of the position for which the job seeker last submitted their resume; jd n The identifier number of the position for which the job seeker first submitted their resume;

[0142] S12. The logic for splitting the sample sequence can be found in [reference]. Figure 5 For example, if the sequence size in the sliding window is 51, containing 51 job postings, including one target job (the job the job seeker last submitted their resume for, located at the beginning of the sample sequence) and 50 other job postings, the sequence is padded with zeros to make it 51. For job seekers with more than 51 job postings, the sequence is split into multiple sample sequences using a sliding window of 51, resulting in n-50 sample sequences: 1-51, 2-52, ..., n-50-n. For instance, if a job seeker submitted resumes to 100 jobs within two months (n=100), 50 sample sequences can be obtained.

[0143] S13. Positive Sample Construction: Each sample sequence containing 51 positions generated in step S12 is labeled as a positive sample.

[0144] S14. Negative Sample Construction: For each sample sequence generated in step S12, the first job title is the target value; for example, randomly select 10 jobs from all jobs (excluding the target job title) and replace the first job title. This will generate 10 new sample sequences, which are the negative sample sequences.

[0145] S15. Shuffle the positive and negative sample sequences to generate the final sample set.

[0146] S16. Select 20% of the final sample set as test samples and 80% as training samples.

[0147] II. Training process of the job expectation transfer model:

[0148] S21. Feature extraction and construction of the underlying Embedding layer: job information and location information and their vector mapping.

[0149] a) Location information: The location information is set as a 128-dimensional vector. The formula for calculating the PE of the i-th vector at the t-th position is as follows:

[0150]

[0151] Where: d is the vector dimension; k takes the value of (vector dimension d / 2-1);

[0152] When i = 2k, the vector is in an even position; when i = 2k+1, the vector is in an odd position.

[0153] b)(Level 3 classification): Functional ID feature mapping is a 64-dimensional embedding;

[0154] c) Time span: The time difference between job n and the target job, in days or hours, is mapped to a 64-dimensional embedding. For example, a difference of 5 days can be mapped to a 64-dimensional embedding.

[0155] d) Position ID: The job ID feature is mapped to a 64-dimensional embedding;

[0156] e) Industry (Level 1 Function): Industry ID feature mapping is a 64-dimensional embedding;

[0157] f) City of work: City ID features can be mapped to a 64-dimensional Embedding;

[0158] g) Salary range: The salary ID feature is mapped to a 64-dimensional embedding. For example, a salary range of 6000-7999 can be labeled as 5 and mapped to a 64-dimensional embedding.

[0159] S22. Concat: Perform a concat operation on the location information and job information to concatenate the above 7 vectors into a large vector Embedding_concat_1. The concatenated vector has a dimension of 512.

[0160] S23. Input Embedding_concat_1 as input data into the Transformer structure.

[0161] S24. Construct the Transformer structure:

[0162] a) Multi-Head Self-Attention

[0163] i. The conversion formula is as follows:

[0164]

[0165] MultiHead(Q,K,V)=Concat(head1,..head n W o

[0166] head i =Attention(QW i Q KW i k VW i v )

[0167] Wherein: the number of heads is 8-dimensional, and each head vector is 64-dimensional;

[0168] ii. Assign the value of Embedding_concat_1 to QKV, that is, Q = Embedding_concat_1, K = Embedding_concat_1, V = Embedding_concat_1;

[0169] Perform a linear mapping on QKV respectively, mapping them to the same vector space, and then split them into 8 heads;

[0170]

[0171] Apply Attention(Q,K,V) to QKV, then concatenate the results of Attention(Q,K,V), and finally perform a fully connected (FC) linear mapping, denoted as Embedding_concat_2'.

[0172] Concat(head1, ...head) n W o

[0173] b) Add & Norm

[0174] Embedding_2=Embedding_concat_1+Embedding_concat_2'

[0175] After normalization, Embedding_3 = LN(Embedding_2).

[0176]

[0177] Normalization is performed at the layer level, where the mean μ and variance σ are the mean and variance of all feature vector dimensions at that layer level.

[0178] The purpose of Add is residual learning.

[0179] c) Feed Forward (feedforward network layer)

[0180] i. Two fully connected (FC) layers, with a hidden layer vector dimension of 512.

[0181] Embedding_4=FC2(FC1(Embedding_3))

[0182] FC(X)=W*X+B

[0183] In the formula, W is the weight, X is the input vector, and B is the bias;

[0184] d) Add & Norm

[0185] Same logic as the first Add&Norm above

[0186] Add:Embedding_5=Embedding_3+Embedding_4

[0187] Norm:Embedding_5=LN(Embedding_5)

[0188] S25. Concat: Concatenate the 51 embedding vectors output by the transformer layer together, denoted as Embedding_concat_6, and use it as input to the MLP.

[0189] S26. Construct an MLP (Multilayer Perceptron)

[0190] a) 4-layer fully connected (FC) layer, with 3 hidden layer vectors of dimensions 512, 512 and 256 respectively, and a 1-dimensional output layer. Finally, sigmoid is used as the activation function.

[0191] b)Embedding_7=FC4(FC3(FC2(FC1(Embedding_concat_6))))

[0192] FC(X)=W*X+B

[0193] predict=sigmoid(Embedding_7)

[0194] The conditional expression for the activation function is as follows:

[0195]

[0196] The relevant score can be obtained by calculating using the above formula. The value of the relevant score is between 0 and 1.

[0197] S27. Constructing the loss function

[0198] The loss function used is the cross-entropy loss function. The relevant score results calculated by the sigmoid activation function in step S26 are input into the cross-entropy loss function `loss` to calculate the loss value. The specific formula is as follows:

[0199]

[0200] Where f(x) i ) represents the relevant score calculated in step S26. When the training samples are positive samples, y i The value is 1; when the training samples are negative samples, y i 0; m is the total number of sample groups in the sample set.

[0201] S28. Model Parameter Update (Learning)

[0202] Based on the loss value calculated in step S27, the Adagrad optimization algorithm is used to calculate and update the gradient, and backpropagation is performed to update the model parameters and the underlying embedding vector. Steps S27-S28 are repeated until the predetermined number of training rounds is reached or other stopping conditions are met, at which point training stops, and the trained job expectation transfer model is obtained.

[0203] III. The reasoning process for recommending job information:

[0204] S31. For a job seeker, extract the multiple job applications submitted by the job seeker in the last 60 days, sort the application times of the job applications in order from the latest to the earliest to obtain the application job sequence (the order is the same as the sorting in step S11), and extract the top 50 job applications (if there are fewer than 50 job applications, fill the rest with 0) as the prediction sequence.

[0205] S32. Obtain the job positions to be predicted. For example, there are 3 job positions to be predicted: jd1, jb2, and jb3. Input the 3 job positions to be predicted and this prediction sequence into the trained job expectation transfer model to obtain 3 corresponding relevance scores. Compare the relevance scores and take the job position corresponding to the maximum relevance score as the recommended job information for the job seeker, that is, the job position that the job seeker is most likely to apply for after the expectation transfer.

[0206] The following example illustrates the shift in salary expectations during the job search lifecycle:

[0207] There is a group of job seekers who initially had high expectations, but as their resumes received less-than-ideal feedback and interview results, their salary expectations gradually decreased. Below is a list of their actual job applications, arranged from earliest to latest:

[0208] The application sequence is as follows: 6000-7999, 6000-7999, 8000-9999, 6000-7999, 4500-5999, 4500-5999, 6000-7999, 4500-5999, 4500-5999, 3000-4499, 4500-5999, 4500-5999 (the complete application sequence is longer than this; only a portion is shown here). From the data above, it's clear that initially, applications were concentrated in the 6000-7999 yuan range or even higher, while later applications concentrated in the 4500-5999 yuan range or even lower. Therefore, we can consider the 6000-7999 yuan range as application position 1 and the 4500-5999 yuan range as application position 2.

[0209] Sort the data sequentially from the latest to the earliest time, and the delivery sequence data is as follows:

[0210] 4500-5999, 4500-5999, 3000-4499, 4500-5999, 4500-5999, 6000-7999, 4500-5999, 4500-5999, 6000-7999, 8000-9999, 6000-7999, 6000-7999;

[0211] The predicted job positions 1 and 2 were input into the trained job expectation transfer model along with the aforementioned application sequence data. The output correlation scores were approximately 0.65 and 0.89, respectively. This indicates that positions with salaries between 4500 and 5999 yuan are more suitable for job seekers, thus achieving a prediction of the job seeker's expected salary transfer. In this embodiment, salary is used as an example. Job seekers' expected functions, industries, and work cities all have different transfer rates. This algorithm can fuse these influencing factors into a whole to reflect the job seeker's expected salary transfer.

[0212] This application proposes an algorithm based on the Transformer mechanism to characterize job seekers' expectations shift. When applied to an advertising ranking model, this algorithm can improve the display-to-display ratio (resume submissions / exposures) by 5.7%. Applying this algorithm to click-through rate (CTR) and conversion rate (CTR) models allows for CTR and CTR prediction.

[0213] Corresponding to the method embodiments of this application, this application also provides an apparatus for constructing training samples, such as... Figure 7 As shown, it includes:

[0214] The first acquisition module 710 is used to acquire the first historical job application data of job seekers within a first time period; wherein, the first historical job application data includes job information of multiple job applications submitted;

[0215] The first sequence module 720 is used to sort the first historical job posting data in order from the current time to the front, so as to obtain the first delivery sequence data; wherein, the first delivery sequence data includes the position information of multiple job postings that have been submitted in order from the current time to the front.

[0216] The segmentation module 730 is used to segment the first delivery sequence data into multiple positive sample sets according to a preset segmentation strategy; wherein each positive sample set includes a target position and n submitted positions, the target position is the position in each positive sample set whose submission time is closest to the current time; n is greater than or equal to 2;

[0217] Selection module 740 is used to randomly select multiple positions from the delivery sequence data to obtain pre-selected positions; wherein, the pre-selected positions are non-target positions;

[0218] Replacement module 750 is used to replace the target job in each set of positive samples with the pre-selected job to obtain a corresponding set of multiple sets of negative samples;

[0219] Training sample module 760 is used to obtain a training sample set based on the positive sample set and the negative sample set.

[0220] Corresponding to the method embodiments of this application, this application also provides a device for generating a job expectation transfer model, based on the training sample set constructed by the method, such as... Figure 8 As shown, specifically, it may include:

[0221] The first building module 810 is used to build the model to be trained; wherein the model to be trained includes a transformer structure and a multilayer perceptron, and the vector data of multiple transformation sequences output by the transformer structure are concatenated and used as the input of the multilayer perceptron;

[0222] The training module 830 is used to train the constructed model to be trained using the training sample set to obtain a trained job expectation transfer model; wherein the job expectation transfer model is used to predict the application sequence data of the current job seeker and the relevant scores of multiple positions to be predicted.

[0223] In other embodiments, the apparatus for generating the job expectation transfer model may further include: a second construction module 820, which is used to construct a training sample set; wherein the training sample set includes multiple sets, and each set of training sample sets includes job information and location information; the location information is the sequence position of multiple job applications ordered in order of their application time relative to the current time.

[0224] In some embodiments, the training module 830 includes:

[0225] The extraction unit is used to extract the various influencing factors and location information of job information in each sample set;

[0226] The first mapping unit is used to map each influencing factor in the job information into a first-class feature vector of the corresponding dimension.

[0227] The second mapping unit is used to map the location information into a second type of feature vector with a corresponding dimension;

[0228] The first concatenation unit is used to concatenate the first type feature vectors and the second type feature vectors to obtain the concatenated first type vector data.

[0229] The first training unit is used to input the concatenated first-class vector data into the transformer structure, train the transformer structure, and output vector data of multiple transformation sequences.

[0230] The second concatenation unit is used to concatenate vector data from multiple transformation sequences to obtain concatenated second-class vector data.

[0231] The second training unit is used to input the concatenated second-class vector data into the multilayer perceptron to train the multilayer perceptron and output the first correlation score between the target position and the multiple positions applied for.

[0232] Corresponding to the method embodiments of this application, this application also provides a job information recommendation device, based on the job expectation transfer model generated by the method, such as... Figure 9 As shown, it includes:

[0233] The second acquisition module 910 is used to acquire the second historical job application data of the current job seeker within the second time period; wherein, the second historical job application data includes job information of multiple job applications submitted;

[0234] The second sequence module 920 is used to sort the second historical job posting data in order from the latest to the earliest time to obtain the second delivery sequence data; wherein, the second delivery sequence data includes the position information of multiple job postings sorted in order of the closest time to the current time;

[0235] The third acquisition module 930 is used to acquire multiple job positions to be predicted;

[0236] Input module 940 is used to input multiple job positions to be predicted and second application sequence data into a trained job expectation transfer model to obtain a prediction relevance score for each job position to be predicted.

[0237] The recommendation module 950 is used to obtain recommended job information based on the prediction relevance score of each job to be predicted.

[0238] The electronic device in this application embodiment may be a user terminal device, a server, other computing devices, or a cloud server. Figure 10 The diagram illustrates the hardware structure of an electronic device according to an embodiment of this 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 methods described above.

[0239] Specifically, processor 601 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. Memory 602 may include mass storage for data or instructions. For example, memory 602 may be at least one of the following: hard disk drive (HDD), read-only memory (ROM), random access memory (RAM), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, universal serial bus (USB) drive, or other physical / tangible memory storage device. Alternatively, memory 602 may include removable or non-removable (or fixed) media. Furthermore, memory 602 may be internal or external to the integrated gateway disaster recovery device. Memory 602 may be non-volatile solid-state memory. In other words, typically memory 602 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 (e.g., by one or more processors), it can perform the operations described in the methods of the embodiments of this application. The processor 601 implements the process or function of any of the methods described in the above embodiments by reading and executing computer program instructions stored in the memory 602.

[0240] In one example Figure 10 The illustrated electronic device may also include a communication interface 603 and a bus 610. The processor 601, memory 602, and communication interface 603 are connected via bus 610 and communicate with each other. Communication interface 603 is primarily used to enable communication between modules, devices, units, and / or equipment in the embodiments of this application. Bus 610 may include hardware, software, or both, and can couple components of the online data traffic billing device together. For example, the bus may include at least one of the following: Accelerated Graphics Port (AGP) or other graphics bus, Enhanced Industry Standard Architecture (EISA) bus, Front Side Bus (FSB), HyperTransport (HT) Interconnect, Industry Standard Architecture (ISA) bus, Infinite Bandwidth Interconnect, Low Pin Count (LPC) bus, memory bus, Microchannel Architecture (MCA) bus, Peripheral Component Interconnect (PCI) bus, PCI-Express (PCI-X) bus, Serial Advanced Technology Attachment (SATA) bus, Video Electronics Standards Association Local (VLB) bus, or other suitable buses. Bus 610 may include one or more buses. Although specific buses are described or illustrated in the embodiments of this application, any suitable bus or interconnection method may be considered in the embodiments of this application.

[0241] In conjunction with the methods in the above embodiments, this application also provides a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the process or function of any of the methods in the above embodiments.

[0242] In addition, this application also provides a computer program product that stores computer program instructions, which, when executed by a processor, implement the process or function of any of the methods described above.

[0243] The flowcharts and / or block diagrams of methods, apparatuses, systems, and computer program products according to embodiments of this application have been exemplarily described above, and related aspects have been described. It should be understood that each block or combination thereof in the flowcharts and / or block diagrams may be implemented by computer program instructions, by dedicated hardware performing a specified function or action, or by a combination of dedicated hardware and computer instructions. For example, these computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to form a machine that enables the implementation of the function / action specified in each block or combination thereof in the flowcharts and / or block diagrams, executable via such processor. Such a processor may be a general-purpose processor, a dedicated processor, a special-purpose application processor, or a field-programmable logic circuit.

[0244] The functional blocks shown in the structural block diagrams of this application can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc.; when implemented in software, they are programs or code segments used to perform the required tasks. Programs or code segments can be stored in memory or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. Code segments can be downloaded via computer networks such as the Internet or intranets.

[0245] It should be noted that this application is not limited to the specific configurations and processes described above or shown in the figures. The above descriptions are merely specific embodiments of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the described systems, devices, modules, or units can be referred to the corresponding processes in the method embodiments, and need not be repeated here. It should be understood that the scope of protection of this application is not limited thereto. Any person skilled in the art can conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application.

Claims

1. A method for generating a job-seeking expectation transfer model, characterized in that, include: Construct a model to be trained; the model to be trained includes a transformer structure and a multilayer perceptron. The vector data of multiple transformation sequences output by the transformer structure are concatenated and used as the input of the multilayer perceptron. Construct a training sample set; the training sample set includes multiple groups, each group of training sample sets includes job information and location information; the location information is the sequence position of multiple job applications ordered in order of application time from the current time. The constructed model to be trained is trained using a training sample set to obtain a well-trained job expectation transfer model; among which, The job expectation transfer model is used to predict the current job seeker's application sequence data and the relevant scores of multiple positions to be predicted; The step of training the constructed model to be trained using the training sample set to obtain a trained job expectation transfer model includes: Extract the various influencing factors and location information of job information from each training sample set; Map each influencing factor in the job information to a first-class feature vector of the corresponding dimension; The location information is mapped to a second type of feature vector with corresponding dimensions; The first-class feature vectors and the second-class feature vectors are concatenated to obtain the concatenated first-class vector data. The concatenated first-class vector data is input into the transformer structure, and the transformer structure is trained to output vector data of multiple transformation sequences. The vector data of multiple transformation sequences are concatenated to obtain the concatenated second type of vector data; The concatenated second type of vector data is input into a multilayer perceptron to train the multilayer perceptron and output the first correlation score between the target position and multiple positions applied for. The step of mapping the location information into a second type of feature vector of corresponding dimensions includes: The vector PE at position t can be calculated using the following formula: Where: d is the vector dimension; k takes the value of vector dimension d / 2-1; when i = 2k, the vector is in an even position; when i = 2k+1, the vector is in an odd position.

2. The method according to claim 1, characterized in that, The step of training the constructed model to be trained using the training sample set to obtain a trained job expectation transfer model further includes: Input the relevant scores into the cross-entropy loss function to calculate the loss value; Based on the calculated loss value, an optimization algorithm is used to calculate and update the gradient, and backpropagation is performed to update the parameters of the job expectation transfer model and the underlying first-class and second-class feature vector data; the loss function is shown in the following conditional expression: In the formula, f(x) i ) represents the first relevance score; when the training samples are positive samples, y i The value is 1; when the training samples are negative samples, y i 0; m is the total number of sample groups in the sample set.

3. The method according to claim 1, characterized in that, Job information includes one or more of the following influencing factors: job ID, job function, industry, city of work, salary range, and the time difference between the nth job and the target job application.

4. A method for recommending job information, characterized in that, The job expectation transfer model generated based on the method described in any one of claims 1-3 includes: Obtain the second historical job posting data of the current job seeker within the second time period; the second historical job posting data includes job information for multiple job postings; The second historical job posting data is sorted sequentially from the current time to the front to obtain the second posting sequence data; wherein, the second posting sequence data includes the position information of multiple job postings sorted in order of their closest time to the current time; Obtain multiple job postings to be predicted; Multiple job postings to be predicted and second application sequence data are input into a trained job expectation transfer model to obtain a prediction relevance score for each job posting to be predicted. Recommended job information is obtained based on the prediction-related score of each job to be predicted.

5. The method according to claim 4, characterized in that, The process of obtaining recommended job information based on the prediction relevance score of each job to be predicted includes: The prediction relevance scores of each job to be predicted are sorted in descending order, and the job to be predicted corresponding to the highest prediction relevance score is taken as the recommended job information.

6. The method according to claim 4, characterized in that, The acquisition of multiple job titles to be predicted includes: Obtain the jobs that job seekers have clicked and / or viewed, and select the most recent clicked and / or viewed jobs as the jobs to be predicted; or... Obtain job postings from recruiters, and select the most recent posted positions as the jobs to be predicted; or... We use a recall approach to identify positions with high application rates from the past job applications of current job seekers and select these positions for prediction.

7. A device for generating a job-seeking expectation transfer model, characterized in that, include: The first building module is used to build the model to be trained; the model to be trained includes a transformer structure and a multilayer perceptron, and the vector data of multiple transformation sequences output by the transformer structure are concatenated and used as the input of the multilayer perceptron. The second construction module is used to construct a training sample set; the training sample set includes multiple sets, each set including job information and location information; the location information is the sequence position of multiple job applications ordered in order of application time from the current time. A training module is used to train the constructed model to be trained using the training sample set to obtain a trained job expectation transfer model; wherein, the job expectation transfer model is used to predict the current job seeker's application sequence data and the relevant scores of multiple positions to be predicted; the training module includes: The extraction unit is used to extract the various influencing factors and location information of job information in each sample set; The first mapping unit is used to map each influencing factor in the job information into a first-class feature vector of the corresponding dimension. The second mapping unit is used to map the location information into a second type of feature vector with a corresponding dimension; The first concatenation unit is used to concatenate the first type feature vectors and the second type feature vectors to obtain the concatenated first type vector data. The first training unit is used to input the concatenated first-class vector data into the transformer structure, train the transformer structure, and output vector data of multiple transformation sequences. The second concatenation unit is used to concatenate vector data from multiple transformation sequences to obtain concatenated second-class vector data. The second training unit is used to input the concatenated second-class vector data into the multilayer perceptron to train the multilayer perceptron and output the first correlation score between the target job and the multiple jobs applied for; wherein... The second mapping unit includes: the t-th position, and the i-th vector PE can be calculated using the following formula: Where: d is the vector dimension; k takes the value of vector dimension d / 2-1; when i = 2k, the vector is in an even position; when i = 2k+1, the vector is in an odd position.

8. A job information recommendation device, characterized in that, The job expectation transfer model generated based on the method described in any one of claims 1-3 includes: The second acquisition module is used to acquire the second historical job application data of the current job seeker within the second time period; wherein, the second historical job application data includes job information of multiple job applications submitted; The second sequence module is used to sort the second historical job posting data in order from the latest to the earliest time to obtain the second delivery sequence data; wherein, the second delivery sequence data includes the position information of multiple job postings sorted in order of the closest time to the current time; The third acquisition module is used to acquire multiple job positions to be predicted. The input module is used to input multiple job postings to be predicted and second application sequence data into the trained job expectation transfer model to obtain the prediction relevance score for each job posting. The recommendation module is used to obtain recommended job information based on the prediction relevance score of each job to be predicted.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory storing computer program instructions; when the electronic device executes the computer program instructions, it implements the method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the method as described in any one of claims 1-6.

Images