Employment recommendation method and device, electronic equipment and storage medium

Abstract

The present disclosure relates to an employment recommendation method and device, electronic equipment and storage medium, determining a user position set, and user resume information, position description text information and historical behavior information corresponding to the user. Through a large language model, a first matching score of each user position is determined based on the user position set, the user resume information and the position description text information. Then, through a deep knowledge graph convolution network and a graph enhancement model, a second matching score and a third matching score of each user position are determined based on the user position set and the historical behavior information. A comprehensive score is determined according to the first matching score, the second matching score and the third matching score of each user position, and at least one target user position is recommended for the user according to the comprehensive score. The present disclosure determines the position matching from three different dimensions through different combined features, and determines the final recommendation result together, thereby improving the accuracy of the employment recommendation result.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and more particularly to a job recommendation method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the increasing number of media college graduates this year, the employment pressure on college graduates is increasing, and the employment situation is severe. Furthermore, with the digitalization of traditional industries, online recruitment has gained popularity among job seekers due to its wide range and variety of job postings. However, existing recruitment systems suffer from problems such as crude job matching and recommendation methods, leading to decreased accuracy and failing to provide job seekers with satisfactory positions. Current technologies, when recommending jobs to job seekers, exhibit high homogeneity and reduced personalization, failing to provide the most suitable job information for each individual, resulting in poor job suitability and impacting employment outcomes. Summary of the Invention

[0003] In view of this, this disclosure proposes a job recommendation method, apparatus, electronic device, and storage medium designed to achieve personalized and accurate job recommendations for each job seeker.

[0004] According to a first aspect of this disclosure, a job recommendation method is provided, the method comprising:

[0005] Identify a set of user positions that includes multiple user positions;

[0006] Determine the user's corresponding resume information, job description text information, and historical behavior information;

[0007] Based on the user job set, user resume information, and job description text information, a first matching score is determined for each user job using a large language model.

[0008] Based on the user job set and the historical behavior information, a second matching score is determined for each user job using a deep knowledge graph convolutional network.

[0009] Based on the user job set and the historical behavior information, a graph augmentation model is used to determine the third matching score for each user job.

[0010] A comprehensive score is determined based on the first matching score, the second matching score, and the third matching score corresponding to each user position, and at least one target user position is recommended to the user based on the comprehensive score.

[0011] In one possible implementation, determining the first matching score for each user job based on the user job set, user resume information, and job description text information using a large language model includes:

[0012] The skill matching degree and industry matching degree are determined based on the user resume information and the job description text information;

[0013] The input information is determined based on the job description text, the user resume information, the skill matching degree, and the industry matching degree;

[0014] The input information and the set of user positions are input into the large language model, and the first matching score corresponding to each user position is output.

[0015] In one possible implementation, determining the skill matching degree and industry matching degree based on the user resume information and the job description text information includes:

[0016] Determine the job skill information and job professional information corresponding to the job description text information;

[0017] Determine the user skills and majors corresponding to the user resume information;

[0018] The skill matching degree is determined based on the job skill information and the user skills;

[0019] The industry matching degree is determined based on the job professional information and the user's professional background.

[0020] In one possible implementation, determining the input information based on the job description text information, the user resume information, the skill matching degree, and the industry matching degree includes:

[0021] The input information is obtained by concatenating the job description text, the user resume information, the skill matching degree, and the industry matching degree.

[0022] In one possible implementation, determining the second matching score for each user job based on the user job set and the historical behavior information using a deep knowledge graph convolutional network includes:

[0023] Construct a corresponding user knowledge graph based on the historical behavior information and the user job set;

[0024] The user knowledge graph is input into the deep knowledge graph convolutional network to obtain a second matching score for each user's job position.

[0025] In one possible implementation, the nodes of the user knowledge graph include users and user positions, the edges between nodes represent the interaction relationship between users and user positions, and the feature corresponding to each node is the attribute information of the user / user position.

[0026] In one possible implementation, the step of determining a third matching score for each user job based on the user job set and the historical behavior information using a graph augmentation model includes:

[0027] The graph augmentation model determines the user representation corresponding to the user based on the historical behavior information.

[0028] The graph augmentation model determines the graph representation of each user job based on the set of user jobs.

[0029] Based on the user representation and the graph representation, the user's interest in each user position is calculated to obtain a third matching score corresponding to each user position.

[0030] According to a second aspect of this disclosure, a job recommendation device is provided, the device comprising:

[0031] The user job determination module is used to determine a set of user jobs that includes multiple user jobs.

[0032] The user information determination module is used to determine the user's resume information, job description text information, and historical behavior information.

[0033] The first matching module is used to determine the first matching score corresponding to each user job based on the user job set, the user resume information and the job description text information using a large language model;

[0034] The second matching module is used to determine the second matching score corresponding to each user position based on the user position set and the historical behavior information through a deep knowledge graph convolutional network.

[0035] The third matching module is used to determine the third matching score corresponding to each user job based on the user job set and the historical behavior information using a graph augmentation model.

[0036] The job recommendation module is used to determine a comprehensive score based on the first matching score, the second matching score, and the third matching score corresponding to each user job, and to recommend at least one target user job to the user based on the comprehensive score.

[0037] In one possible implementation, the first matching module is further configured to:

[0038] The skill matching degree and industry matching degree are determined based on the user resume information and the job description text information;

[0039] The input information is determined based on the job description text, the user resume information, the skill matching degree, and the industry matching degree;

[0040] The input information and the set of user positions are input into the large language model, and the first matching score corresponding to each user position is output.

[0041] In one possible implementation, the first matching module is further configured to:

[0042] Determine the job skill information and job professional information corresponding to the job description text information;

[0043] Determine the user skills and majors corresponding to the user resume information;

[0044] The skill matching degree is determined based on the job skill information and the user skills;

[0045] The industry matching degree is determined based on the job professional information and the user's professional background.

[0046] In one possible implementation, the first matching module is further configured to:

[0047] The input information is obtained by concatenating the job description text, the user resume information, the skill matching degree, and the industry matching degree.

[0048] In one possible implementation, the second matching module is further configured to:

[0049] Construct a corresponding user knowledge graph based on the historical behavior information and the user job set;

[0050] The user knowledge graph is input into the deep knowledge graph convolutional network to obtain a second matching score for each user's job position.

[0051] In one possible implementation, the nodes of the user knowledge graph include users and user positions, the edges between nodes represent the interaction relationship between users and user positions, and the feature corresponding to each node is the attribute information of the user / user position.

[0052] In one possible implementation, the third matching module is further configured to:

[0053] The graph augmentation model determines the user representation corresponding to the user based on the historical behavior information.

[0054] The graph augmentation model determines the graph representation of each user job based on the set of user jobs.

[0055] Based on the user representation and the graph representation, the user's interest in each user position is calculated to obtain a third matching score corresponding to each user position.

[0056] According to a third aspect of this disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described method when executing instructions stored in the memory.

[0057] According to a fourth aspect of this disclosure, a non-volatile computer-readable storage medium is provided that stores computer program instructions thereon, wherein the computer program instructions, when executed by a processor, implement the above-described method.

[0058] According to a fifth aspect of this disclosure, a computer program product is provided, including computer-readable code or a non-volatile computer-readable storage medium carrying the computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0059] In this embodiment, a set of user job titles is determined, along with the user's corresponding resume information, job description text information, and historical behavior information. A large language model is used to determine a first matching score for each user job title based on the user job title set, resume information, and job description text information. Then, a deep knowledge graph convolutional network and a graph augmentation model are used respectively to determine a second and third matching score for each user job title based on the user job title set and historical behavior information. A comprehensive score is determined based on the first, second, and third matching scores for each user job title, and at least one target user job is recommended to the user based on the comprehensive score. This disclosure improves the accuracy of job recommendation results by performing job matching from three different dimensions using different combinations of features and jointly determining the final recommendation result.

[0060] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0061] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0062] Figure 1 A flowchart illustrating an employment recommendation method according to an embodiment of the present disclosure is shown;

[0063] Figure 2 A schematic diagram of an employment recommendation process according to an embodiment of the present disclosure is shown;

[0064] Figure 3 A schematic diagram of an employment recommendation device according to an embodiment of the present disclosure is shown;

[0065] Figure 4 A schematic diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0066] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0067] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0068] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0069] The job recommendation method of this disclosure can be executed by an electronic device such as a terminal device or a server. The terminal device can be any fixed or mobile terminal, such as a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, or wearable device. The server can be a single server or a server cluster consisting of multiple servers. Any electronic device can implement the job recommendation method of this disclosure by having its processor call computer-readable instructions stored in its memory.

[0070] Figure 1 A flowchart illustrating a job recommendation method according to an embodiment of this disclosure is shown. Figure 1 As shown, the employment recommendation method of this disclosure embodiment may include the following steps S10-S60.

[0071] Step S10: Determine the set of user positions that includes multiple user positions.

[0072] In one possible implementation, a user job set comprising multiple user positions can be determined via an electronic device. Each user job can have corresponding job information and job tags. The job information characterizes the job requirements, such as years of work experience, educational requirements, job certifications, and job characteristics. The job tags can include at least one category to which the job belongs, where multiple categories can have a nested relationship of major categories and subcategories. For example, a user job's category could include an industry category such as "Internet industry, catering industry, telecommunications industry, media industry, etc."; department categories within the industry category such as "technology department, media department, operations department, marketing department, administration department, and human resources department"; and specific job categories within the department category such as "producer, screenwriter, video editor, director, scriptwriter, editor, script planner, content planner." Alternatively, the job tag can simply include the specific job category to which the job belongs. Furthermore, the job tag at the lowest level of the nested relationship, i.e., the most finely divided, can also have corresponding job professional information and job skill information, used to define the professional background of the user job and the professional skills required for the user job, respectively.

[0073] Alternatively, the user job collection can be stored directly on the internal storage of the electronic device, or retrieved from other data storage devices by the electronic device.

[0074] Step S20: Determine the user's resume information, job description text information, and historical behavior information.

[0075] In one possible implementation, electronic devices can determine the user's resume information, job description text information, and historical behavior information corresponding to the user seeking job recommendations. The user resume information represents the job-seeking information of the corresponding job seeker, and may include information such as the user's major, years of work experience, skills, certifications, and work history. The job description text information represents the information of the job the user is looking for, and may include job tags representing the category of the job the user is looking for, or job skills and major information representing the job skills required for the job, respectively. The historical behavior information represents the user's interaction behavior with each job in the user's job set, and may include an interaction score for each job. This interaction score can be calculated based on the user's actions such as liking, disliking, and browsing time after viewing the corresponding job, according to a scoring rule.

[0076] Optionally, the acquisition of user resume information, job description text information, and historical behavior information is not subject to a specific order restriction with the acquisition of the user job set; they can be acquired sequentially or simultaneously. After obtaining the user's corresponding resume information, job description text information, and historical behavior information, data preprocessing can be performed through noise reduction, missing value handling, and other methods to optimize the data.

[0077] Step S30: Based on the user job set, user resume information and job description text information, determine the first matching score corresponding to each user job using a large language model.

[0078] In one possible implementation, after determining the user job set, user resume information, and job description text information, the electronic device can use a pre-trained large language model to determine the degree of matching between each user job in the user job set and the user, based on the user job set, user resume information, and job description text, and obtain the corresponding first matching score. Optionally, the electronic device can first determine the skill matching degree and industry matching degree based on the user resume information and job description text information, and determine the input information based on the job description text information, user resume information, skill matching degree, and industry matching degree. Then, the input information and the user job set are input into the large language model, and the first matching score corresponding to each user job is output.

[0079] Furthermore, skill matching degree characterizes the degree of match between a user's skills and the skills required for their desired job, while industry matching degree characterizes the degree of match between a user's major and the major required for their desired job. The calculation methods for skill matching degree and industry matching degree can include determining the job skill information and job major information corresponding to the job description text, and determining the user's skills and major corresponding to the user's resume information. Skill matching degree is determined based on the job skill information and user skills, and industry matching degree is determined based on the job major information and user major. Specifically, if the job description text includes job skill information and job major information, these can be directly determined. If the job description text includes job tags, the job skill information and job major information corresponding to the job tags are determined using an electronic device. The electronic device can directly extract user skills and major information from the user's resume information.

[0080] After extracting the user's corresponding job skills information, job specialization information, user skills, and user specialization from the electronic device, the above information can be converted into corresponding word vectors through an embedding layer. Then, the similarity between the job skills information and the user skills information is calculated to obtain the skill matching degree, and the similarity between the job specialization information and the user specialization is calculated to obtain the industry matching degree. For example, if the job skills information, job specialization information, user skills, and user specialization are Job Skills, Job Industry, Applicant Skills, and Applicant Industry respectively, the skill matching degree can be determined as Skills = Cosine_Similarity(Vector(Job Skills), Vector(Applicant Skills)), and the industry matching degree as Industry = Cosine_Similarity(Vector(Job Industry), Vector(ApplicantIndustry)). Here, the similarity function Cosine_Similarity can be a function that calculates the Euclidean distance between two vectors.

[0081] Furthermore, after determining the skill matching degree and industry matching degree, the electronic device can concatenate the job description text information, user resume information, skill matching degree, and industry matching degree to obtain the input information. For example, the input information can be determined as: X = Job Description + [SEP] + Resume + [SEP] + Skills + [SEP] + Industry, where JobDescription is the job description text information, Resume is the user resume information, Skills is the skill matching degree, Industry is the industry matching degree, and [SEP] is the separator.

[0082] In one possible implementation, after determining the input information, the electronic device inputs the input information and a set of user job titles into a large language model. The large language model then determines the degree of matching between the user and each user job title in the set based on the input information, obtaining a first matching score for each user job title. Specifically, the large language model in this embodiment includes a multi-classification output layer, which maps the model's output to multiple job categories.

[0083] Optionally, the training process of the large language model uses the cross-entropy loss function, which measures the difference between the model's output probability distribution and the probability distribution of the actual categories, as the objective function. That is, during the training of the large language model, minimizing the value of the cross-entropy function is the objective, and the formula for the cross-entropy loss function can be: ,in, It is the number of categories. It is the actual category probability distribution. This is the predicted probability distribution of a large language model. During model training, by providing training samples with guaranteed category labels, the large language model can learn to adjust its weights to minimize cross-entropy loss.

[0084] Ultimately, through a multi-classification layer design, the large language model can more effectively handle multi-category classification problems in any job search field. These job search fields can include media, internet, communications, and agriculture, among others.

[0085] By using a large language model to determine the degree of user matching with job title, rich semantic text with industry-specific characteristics can be captured through semantic text learning, which can improve the specificity of job recommendations when targeting specific industries.

[0086] Step S20: Based on the user job set and the historical behavior information, determine the second matching score corresponding to each user job using a deep knowledge graph convolutional network.

[0087] In one possible implementation, after acquiring a set of user job titles and user historical behavior information, the electronic device can construct a corresponding user knowledge graph based on the historical behavior information and the set of user job titles. This user knowledge graph is then input into a deep knowledge graph convolutional network to obtain a second matching score for each user job title. The nodes in the user knowledge graph include users and user job titles, with edges representing the interaction relationships between users and user job titles. The features corresponding to each node are the attribute information of the user / user job title. The type of this interaction relationship can be predetermined; for example, it can be determined that there is an interaction relationship between the user and the job title clicked, or between the user and the job title previously applied for, or between the user and the job title whose browsing time exceeds a preset duration.

[0088] Optionally, the features corresponding to each node in the user knowledge graph can be represented as a feature matrix, determined by two vectors obtained from the user's resume information and job description text information extracted by a pre-trained feature extraction model, or by two vectors obtained from the user's job information and job tags. In this embodiment, the selected features are represented as a feature matrix X, where each row corresponds to an output vector of the node input feature extraction model.

[0089] Furthermore, the training of this feature extraction model can be achieved using the Lasso regression method, that is, feature selection is achieved by minimizing a loss function with an L1 regularization term. Its optimization objective can be expressed as: Where W is the parameter matrix composed of parameter vectors of the feature extraction model, and p is the number of parameter vectors included in the parameter matrix. Let x be the parameter vector of the j-th feature extraction model. i y is the feature vector of the i-th input sample during training. i λ is the target value of the i-th sample during training, n is the number of input samples, and λ is a regularization parameter used to control sparsity. During training, the sparsity of the feature extraction model parameters is controlled by adjusting the value of λ. A larger λ value will cause more parameters to tend towards 0, thus achieving stricter feature selection. After training, the features with non-zero parameters in the parameter vector w are the selected features.

[0090] Furthermore, to further improve the performance of the feature extraction model and combine the characteristics of feature selection and smoothness during training, the model loss can be performed using the SSLoss loss function. This loss function combines the sparsity and smoothness requirements of feature selection, and its form is as follows:

[0091]

[0092] in, It is the parameter vector of the feature extraction model. Let X be the parameter vector of the j-th feature extraction model, and let X be the feature matrix of the training set, where each row represents a feature vector; ij It is the j-th feature vector of the i-th input sample during training, x j It is a matrix composed of the j-th feature vectors of all input samples during the training process; y is the target value of the sample in the training set; n is the number of samples in the training set; p is the number of features; and The first term is the regularization parameter, which controls the strength of feature selection and smoothness, respectively. The loss function described above combines the requirements of sparsity and smoothness in feature selection. The second term is L1 regularization of the feature weights, causing the weights of many features to tend to zero, thereby achieving feature selection. The third term is a smoothing penalty for the feature weights, which enhances the smoothness of the model by limiting the variation of adjacent feature weights.

[0093] Furthermore, embodiments of this disclosure can represent the user knowledge graph as G=(V,E), where V is the set of nodes and E is the set of edges. The feature representation corresponding to each node serves as the attribute information of the node, describing the characteristics of the user and the job. The electronic device can input the user knowledge graph and the corresponding feature matrix into a deep knowledge graph convolutional model to determine the matching degree between the user and each user job, and obtain a second matching score corresponding to each user job.

[0094] For deep knowledge graph convolutional models, during application, the update rule for each graph convolutional layer l can be expressed as follows: ,in, It is a node In the Layer representation, It is an activation function. It is a node The neighborhood group, It is a weight matrix. This is the weight matrix used to update node attributes. Through multi-layer graph convolution operations, the neighbor information of a node is aggregated step by step, and the feature representation of the node is updated, outputting the corresponding prediction result. Here, u represents the neighbors of node v, and attr(v) represents the custom attributes of node v. During training, the loss function used by the deep knowledge graph convolutional model is... ,in, It is a binary label indicating whether user u has applied for job j. It is the predicted probability of the model.

[0095] Step S50: Based on the user job set and the historical behavior information, determine the third matching score corresponding to each user job using a graph augmentation model.

[0096] In one possible implementation, after acquiring a set of user job titles and user historical behavior information, the electronic device can use a graph augmentation model to determine the user representation based on the historical behavior information, and then use the same model to determine the graph representation of each user job title based on the set of user job titles. Finally, based on the user representation and the graph representation, the user's interest in each user job title is calculated, resulting in a third matching score for each user job title.

[0097] Optionally, the user representation can be a vector representing user behavior. The electronic device can determine the user representation by first constructing a corresponding directed graph structure based on the user's historical behavior information, and then inputting the constructed directed graph structure into a transformer encoding layer for encoding to obtain the corresponding user representation. Specifically, determining the directed graph structure involves transforming the user's historical behavior information into a directed graph structure, where the user behavior sequence serves as nodes, each node has a temporal relationship, and the edge weights represent information such as the intensity of the user's interaction with a particular job. Using a directed graph structure with temporal features and relationships allows the model to better understand the user's interaction patterns with jobs at different points in time, thereby more accurately capturing changes in user interest. Furthermore, the user behavior is input into the Transformer encoding layer in the form of a directed graph structure. The Transformer encoding layer utilizes a Long Short-Term Memory (LSTM) network mechanism to capture the relationships between different nodes within the user behavior sequence in the directed graph structure and generate the user representation.

[0098] Optionally, the graph representation can be a vector representing the corresponding user job. The method for determining the graph representation of each user job can be to use a graph attention network to capture the graph representation of the user job, enabling the model to more comprehensively understand the relationships between jobs in the directed graph structure. The graph attention network learns the representation of each job node, including the relationships between each job node and its neighboring nodes, as well as the temporal features between nodes. LayerNorm and residual connections are used to improve the performance and stability of the graph attention network, as shown in the formula: Head i This is the i-th head layer.

[0099] Furthermore, after obtaining the user representation and the graph representation of each user's job title, the job representation learned by the graph attention network can be combined with the user representation learned by the Transformer encoding layer to extract personalized interests from user behavior. The formula can be expressed as: ,in, The user representation, representing a user, is a user vector output by the LSTM coding layer. The graph representation of a job is a matrix of job node representations output by a graph attention network. Through a long short-term memory (LSTM) network mechanism, the graph augmentation model can more effectively capture long-term dependencies, increasing the reliability of the model's recommendation results. Then, by taking the dot product of the user representation and the job graph representation, the user's interest score for different jobs is calculated, serving as the corresponding third-match score. The formula can be expressed as: , For job description.

[0100] Step S60: Determine a comprehensive score based on the first matching score, second matching score, and third matching score corresponding to each user job, and recommend at least one target user job to the user based on the comprehensive score. In one possible implementation, after determining the first matching score, second matching score, and third matching score corresponding to each user job through three different methods, the electronic device can jointly determine the comprehensive score corresponding to each user job using the three matching scores, which is used to characterize the degree of matching between the user and the user job. Furthermore, at least one user job that meets the preset conditions in the comprehensive score can be selected as the target user job and recommended to the user.

[0101] Optionally, the process of determining the comprehensive score based on the first matching score, second matching score, and third matching score corresponding to the user's position can be implemented through a fusion layer. This fusion layer calculates the weighted sum of each matching score to obtain the corresponding score, which is Weighted_Output=Attention_WeightChatGPT. +Attention_WeightGCN +Attention_WeightTransformer The weight of each score can be an attention weight, representing the degree of attention paid to that score, and is obtained through learning. The weights of the first matching score, the second matching score, and the third matching score can be Attention_WeightChatGPT=σ( ), Attention_WeightGCN=σ( )、Attention_WeightTransformer=σ( ), where σ is the activation function and W is the learnable weight matrix. This indicates the score of the first match. This indicates the score of the second match. This represents the third matching score. Electronic devices can directly use the calculated weighted sum as the corresponding overall score, or they can input the weighted sum into an activation function to obtain a value within a preset range as the overall score.

[0102] Furthermore, the multiple models in this embodiment can be trained separately or together.

[0103] Figure 2 A schematic diagram of a job recommendation process according to an embodiment of the present disclosure is shown. Figure 2 As shown, when making job recommendations using embodiments of this disclosure, the electronic device can first determine a set of user positions, including multiple user jobs, as well as user resume information, job description text information, and historical behavior information corresponding to each user. Then, based on the user job set, user resume information, and job description text information, a large language model is used to determine a first matching score for each user job. Based on the user job set, user resume information, and job description text information, a deep knowledge graph convolutional network is used to determine a second matching score for each user job. Based on the user job set and historical behavior information, a graph augmentation model is used to determine a third matching score for each user job. Finally, a comprehensive score is determined based on the first, second, and third matching scores for each user job. At least one target user job matching the user is then selected based on the comprehensive score and recommended to the user.

[0104] Based on the above technical features, the embodiments of this disclosure can perform job matching from three different dimensions—large language model, review knowledge graph convolutional network, and graph augmentation model—through different combinations of features, and determine the final recommendation result based on the three matching results, thereby improving the accuracy of job recommendation results and user experience.

[0105] Figure 3 A schematic diagram of a job recommendation device according to an embodiment of the present disclosure is shown. Figure 3 As shown, the job recommendation device in this embodiment may include:

[0106] User job determination module 30 is used to determine a set of user jobs that includes multiple user jobs.

[0107] User information determination module 31 is used to determine the user's resume information, job description text information and historical behavior information corresponding to the user;

[0108] The first matching module 32 is used to determine the first matching score corresponding to each user job based on the user job set, the user resume information and the job description text information using a large language model;

[0109] The second matching module 33 is used to determine the second matching score corresponding to each user position based on the user position set and the historical behavior information through a deep knowledge graph convolutional network.

[0110] The third matching module 34 is used to determine the third matching score corresponding to each user job based on the user job set and the historical behavior information using a graph augmentation model.

[0111] The job recommendation module 35 is used to determine a comprehensive score based on the first matching score, the second matching score, and the third matching score corresponding to each user job, and to recommend at least one target user job to the user based on the comprehensive score.

[0112] In one possible implementation, the first matching module 32 is further configured to:

[0113] The skill matching degree and industry matching degree are determined based on the user resume information and the job description text information;

[0114] The input information is determined based on the job description text, the user resume information, the skill matching degree, and the industry matching degree;

[0115] The input information and the set of user positions are input into the large language model, and the first matching score corresponding to each user position is output.

[0116] In one possible implementation, the first matching module 32 is further configured to:

[0117] Determine the job skill information and job professional information corresponding to the job description text information;

[0118] Determine the user skills and majors corresponding to the user resume information;

[0119] The skill matching degree is determined based on the job skill information and the user skills;

[0120] The industry matching degree is determined based on the job professional information and the user's professional background.

[0121] In one possible implementation, the first matching module 32 is further configured to:

[0122] The input information is obtained by concatenating the job description text, the user resume information, the skill matching degree, and the industry matching degree.

[0123] In one possible implementation, the second matching module 33 is further configured to:

[0124] Construct a corresponding user knowledge graph based on the historical behavior information and the user job set;

[0125] The user knowledge graph is input into the deep knowledge graph convolutional network to obtain a second matching score for each user's job position.

[0126] In one possible implementation, the nodes of the user knowledge graph include users and user positions, the edges between nodes represent the interaction relationship between users and user positions, and the feature corresponding to each node is the attribute information of the user / user position.

[0127] In one possible implementation, the third matching module 34 is further configured to:

[0128] The graph augmentation model determines the user representation corresponding to the user based on the historical behavior information.

[0129] The graph augmentation model determines the graph representation of each user job based on the set of user jobs.

[0130] Based on the user representation and the graph representation, the user's interest in each user position is calculated to obtain a third matching score corresponding to each user position.

[0131] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0132] This disclosure also proposes a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the above-described method. The computer-readable storage medium can be volatile or non-volatile.

[0133] This disclosure also proposes an electronic device, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0134] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0135] Figure 4 A schematic diagram of an electronic device 1900 according to an embodiment of the present disclosure is shown. For example, the electronic device 1900 may be provided as a server or a terminal device. (Refer to...) Figure 4 The electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0136] Electronic device 1900 may also include a power supply component 1926 configured to perform power management of electronic device 1900, a wired or wireless network interface 1950 configured to connect electronic device 1900 to a network, and an input / output interface 1958 (I / O interface). Electronic device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM Mac OS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0137] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of an electronic device 1900 to perform the above-described method.

[0138] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0139] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0140] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0141] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0142] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0143] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0144] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0145] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0146] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A job recommendation method, characterized in that, The method includes: A set of user positions is determined, which includes multiple user positions. Each user position has corresponding position information and position tags. The position tags include at least one category to which the position belongs, wherein the multiple categories have a nested relationship of major categories and subcategories. The system identifies the user resume information, job description text information, and historical behavior information corresponding to the user who needs job recommendations. The job description text information is used to represent the job information that the user wants to find. Based on the user job set, user resume information, and job description text information, a first matching score is determined for each user job using a large language model. Based on the user job set and the historical behavior information, a second matching score is determined for each user job using a deep knowledge graph convolutional network. Based on the user job set and the historical behavior information, a graph augmentation model is used to determine the third matching score for each user job. A comprehensive score is determined based on the first matching score, the second matching score, and the third matching score corresponding to each user's job position, and at least one target user job position is recommended to the user based on the comprehensive score. The step of determining a first matching score for each user job based on the user job set, user resume information, and job description text information using a large language model includes: determining skill matching degree and industry matching degree based on the user resume information and job description text information, whereby the skill matching degree represents the degree of matching between the user's skills and the skills required for the desired job, and the industry matching degree represents the degree of matching between the user's major and the major required for the desired job; determining input information based on the job description text information, the user resume information, the skill matching degree, and the industry matching degree; and inputting the input information and the user job set into the large language model to output the first matching score for each user job. The step of determining the second matching score corresponding to each user position based on the user position set and the historical behavior information using a deep knowledge graph convolutional network includes: constructing a corresponding user knowledge graph based on the historical behavior information and the user position set; inputting the user knowledge graph into the deep knowledge graph convolutional network to obtain the second matching score corresponding to each user position; wherein, the nodes of the user knowledge graph include users and user positions, the edges between nodes represent the interaction relationship between users and user positions, and the features corresponding to each node are the attribute information of the user / user position; The step of determining the third matching score for each user job based on the user job set and the historical behavior information using a graph augmentation model includes: determining the user representation corresponding to the user based on the historical behavior information using the graph augmentation model; determining the graph representation of each user job based on the user job set using the graph augmentation model; calculating the user's interest in each user job based on the user representation and the graph representation to obtain the third matching score for each user job.

2. The method according to claim 1, characterized in that, The step of determining skill matching degree and industry matching degree based on the user resume information and the job description text information includes: Determine the job skill information and job professional information corresponding to the job description text information; Determine the user skills and majors corresponding to the user resume information; The skill matching degree is determined based on the job skill information and the user skills; The industry matching degree is determined based on the job professional information and the user's professional background.

3. The method according to claim 1 or 2, characterized in that, The step of determining input information based on the job description text information, the user resume information, the skill matching degree, and the industry matching degree includes: The input information is obtained by concatenating the job description text, the user resume information, the skill matching degree, and the industry matching degree.

4. A job recommendation device, characterized in that, The device includes: The user job determination module is used to determine a set of user jobs that includes multiple user jobs. Each user job has corresponding job information and job tags. The job tags include at least one category to which the job belongs, and the multiple categories have a nested relationship between major categories and minor categories. The user information determination module is used to determine the user resume information, job description text information and historical behavior information of the user who needs to be recommended a job. The job description text information is used to represent the information of the job that the user wants to find. The first matching module is used to determine the first matching score corresponding to each user job based on the user job set, the user resume information and the job description text information using a large language model; The second matching module is used to determine the second matching score corresponding to each user position based on the user position set and the historical behavior information through a deep knowledge graph convolutional network. The third matching module is used to determine the third matching score corresponding to each user job based on the user job set and the historical behavior information using a graph augmentation model. The job recommendation module is used to determine a comprehensive score based on the first matching score, the second matching score, and the third matching score corresponding to each user job, and to recommend at least one target user job to the user based on the comprehensive score. The first matching module is configured to: determine skill matching degree and industry matching degree based on the user resume information and the job description text information, wherein the skill matching degree is used to characterize the degree of matching between the user's skills and the skills of the user's desired job, and the industry matching degree is used to characterize the degree of matching between the user's major and the major required for the user's desired job; determine input information based on the job description text information, the user resume information, the skill matching degree, and the industry matching degree; input the input information and the user job set into the large language model, and output a first matching score corresponding to each user job; The second matching module is used to: construct a corresponding user knowledge graph based on the historical behavior information and the user job set; input the user knowledge graph into the deep knowledge graph convolutional network to obtain a second matching score for each user job; wherein, the nodes of the user knowledge graph include users and user jobs, the edges between nodes represent the interaction relationship between users and user jobs, and the features corresponding to each node are the attribute information of users / user jobs; The third matching module is used to: determine the user representation corresponding to the user based on the historical behavior information through the graph augmentation model; determine the graph representation of each user job based on the user job set through the graph augmentation model; calculate the user's interest in each user job based on the user representation and the graph representation, and obtain the third matching score corresponding to each user job.

5. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to implement the method of any one of claims 1 to 3 when executing instructions stored in the memory.

6. A non-volatile computer-readable storage medium storing computer program instructions thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 3.

Images