Training method of occupational orientation prediction model, occupational orientation prediction method and equipment

Abstract

The invention provides a training method of an occupational orientation prediction model, an occupational orientation prediction method and equipment, and can be applied to the technical field of artificial intelligence. The method comprises the steps that a heterogeneous graph sample set is obtained, the heterogeneous graph sample set comprises at least one heterogeneous graph sample, the heterogeneous graph sample comprises a plurality of nodes and connecting edges, and the nodes are constructed based on object samples, texts published by the object samples on a network or preset text categories; the preset text category is a category to which a preset word related to the occupational orientation in the target dictionary belongs, and the connecting edge is used for representing an association relationship among the plurality of nodes; and training the occupational orientation prediction model based on the heterogeneous graph sample set to obtain a trained occupational orientation prediction model.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and specifically to a training method, device, equipment, medium and program product for a career orientation prediction model, as well as a career orientation prediction method, device, equipment, medium and program product. Background Art

[0002] Career orientation theory aims to explain how individuals choose their careers and the sources of career satisfaction. The theory is based on the matching of personality and environment, and believes that career choice not only reflects personal interests and abilities, but is also influenced by social environment and cultural background. The current career orientation assessment mainly relies on test questionnaires filled out by the test subjects to assess their career orientation. This process usually involves analysis and comparison of questionnaire results, which is time-consuming and laborious, and is easily affected by the subjective factors of the test subjects. Incorrect assessment results will affect the selection of important positions, resulting in losses of manpower and financial resources.

[0003] In the process of realizing the concept of the present invention, the inventors found that there are at least the following problems in the related art: the related art mainly relies on test questionnaires filled out by the subjects to assess their career orientation, but the test questionnaires may not be filled out truthfully, resulting in low accuracy in the assessment of career orientation. Summary of the invention

[0004] In view of the above problems, the present disclosure provides a training method for a career orientation prediction model, a career orientation prediction method, an apparatus, a device, a medium and a program product.

[0005] According to one aspect of the present disclosure, a method for training a career orientation prediction model is provided, comprising: obtaining a heterogeneous graph sample set, wherein the heterogeneous graph sample set includes at least one heterogeneous graph sample, the heterogeneous graph sample includes a plurality of nodes and connecting edges, the nodes are constructed based on object samples, texts published by the object samples on the Internet, or preset text categories, the preset text categories are categories to which preset words related to career orientation in a target dictionary belong, and the connecting edges are used to characterize the association relationship between the plurality of nodes;

[0006] The career orientation prediction model is trained based on the heterogeneous graph sample set to obtain a trained career orientation prediction model.

[0007] According to another aspect of the present disclosure, a method for predicting career orientation is provided, comprising: obtaining a heterogeneous graph for a subject to be tested, wherein the heterogeneous graph comprises a plurality of nodes and connecting edges, the nodes are constructed based on the subject to be tested, texts published by the subject to be tested on the Internet, or preset text categories, the preset text categories are categories to which preset words related to career orientation in a target dictionary belong, and the edges are used to represent the association relationship between the plurality of nodes;

[0008] Vectorize the heterogeneous graph to obtain a vector representation of the heterogeneous graph;

[0009] The vector representation of the heterogeneous graph is input into the trained career orientation prediction model to obtain the career orientation prediction result for the object to be tested.

[0010] According to another aspect of the present disclosure, a training device for a career orientation prediction model is provided, comprising:

[0011] A first acquisition module is used to acquire a heterogeneous graph sample set, wherein the heterogeneous graph sample set includes at least one heterogeneous graph sample, the heterogeneous graph sample includes a plurality of nodes and connecting edges, the nodes are constructed based on the object sample, the text published by the object sample on the network, or a preset text category, the preset text category is a category to which a preset word related to career orientation in a target dictionary belongs, and the connecting edges are used to represent the association relationship between the plurality of nodes;

[0012] The training module is used to train the career orientation prediction model based on the heterogeneous graph sample set to obtain a trained career orientation prediction model.

[0013] According to another aspect of the present disclosure, a career orientation prediction device is provided, comprising:

[0014] A second acquisition module is used to acquire a heterogeneous graph for the object to be tested, wherein the heterogeneous graph includes a plurality of nodes and connecting edges, the nodes are constructed based on the object to be tested, the text published by the object to be tested on the Internet, or a preset text category, the preset text category is a category to which a preset word related to career orientation in a target dictionary belongs, and the edge is used to represent the association relationship between the plurality of nodes;

[0015] A vectorization module is used to vectorize the heterogeneous graph to obtain a vector representation of the heterogeneous graph;

[0016] The result determination module is used to input the vector representation of the heterogeneous graph into the trained career orientation prediction model to obtain the career orientation prediction result for the object to be tested.

[0017] Another aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0018] Another aspect of the present disclosure further provides a computer-readable storage medium having a computer program or instruction stored thereon, which implements the steps of the above method when the computer program or instruction is executed by a processor.

[0019] Another aspect of the present disclosure further provides a computer program product, including a computer program or instructions, which implement the steps of the above method when executed by a processor.

[0020] According to the training method of the career orientation prediction model disclosed in the present invention, the career orientation prediction model is trained by using a heterogeneous graph sample set to obtain a trained career orientation prediction model. Since the heterogeneous graph sample set includes at least one heterogeneous graph sample, and multiple nodes in the heterogeneous graph sample are constructed based on the object sample, the text published by the object sample on the Internet, and the preset text category, the trained career orientation prediction model can realize the mining of the text published by the object sample on the Internet and the dynamic aggregation of the text belonging to the same preset text category, so that the trained career orientation prediction model can output more accurate prediction results. Therefore, it at least partially solves the technical problem in the related art that the accuracy of career orientation assessment is low due to the possibility that the test questionnaire may not be filled out truthfully, and achieves the technical effect of being able to more accurately predict the career orientation type of the object to be tested through the trained career orientation prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above contents and other purposes, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0022] Figure 1 A diagram schematically illustrates an application scenario of a career orientation prediction model training method, a career orientation prediction method, an apparatus, a device, a medium, and a program product according to an embodiment of the present disclosure;

[0023] Figure 2 A flowchart schematically shows a method for training a career orientation prediction model according to an embodiment of the present disclosure;

[0024] Figure 3 A schematic diagram schematically shows a method of adding disturbance to an unlabeled sample to obtain an enhanced sample according to an embodiment of the present disclosure;

[0025] Figure 4 The structure diagram of the career orientation prediction model according to the embodiment of the present disclosure is schematically shown;

[0026] Figure 5 A schematic diagram of attention calculation of M target text category nodes having the highest correlation with the target object node in the attention layer according to an embodiment of the present disclosure is shown;

[0027] Figure 6 A model training structure diagram schematically illustrates a method for training a career orientation prediction model according to an embodiment of the present disclosure;

[0028] Figure 7 A flowchart of a method for training a career orientation prediction model according to another embodiment of the present disclosure is schematically shown;

[0029] Figure 8 A flowchart of a method for predicting career orientation according to an embodiment of the present disclosure is schematically shown;

[0030] Fig. 9 The structure block diagram of the training device of the career orientation prediction model according to the embodiment of the present disclosure is schematically shown;

[0031] Fig.10 A structural block diagram of a career orientation prediction device according to an embodiment of the present disclosure is schematically shown; and

[0032] Fig.11 A block diagram of an electronic device suitable for implementing a career orientation prediction model method and a career orientation prediction method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0033] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0034] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0035] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0036] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0037] It should be noted that the career orientation prediction model method and career orientation prediction method disclosed in the present invention can be used in the field of artificial intelligence technology, and can also be used in any field other than the field of artificial intelligence technology, such as: the field of big data technology, the field of computer technology, etc. The present disclosure does not limit the application fields of the career orientation prediction model method and career orientation prediction method.

[0038] In the technical solution of the present disclosure, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0039] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating a person's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs, and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.

[0040] During the research, it was found that the career orientation theory was proposed by psychologist John Holland in the 1960s to explain how individuals choose their careers and the source of career satisfaction. The theory is based on the matching of personality and environment, and believes that career choice not only reflects personal interests and abilities, but is also influenced by social environment and cultural background. Career orientation is divided into six types: operational, exploratory, cooperative, managerial, secretarial, and meticulous, which examine personality and personality characteristics. These types are interrelated to form a model to help individuals identify the career path that suits them. Among them, the operational type mainly examines the degree of orientation towards work that requires operation such as machinery and tools; the exploratory type mainly examines the degree of orientation towards curiosity, observation and sensitivity towards new things; the cooperative type mainly examines the degree of orientation towards attitude, willingness to cooperate and sense of responsibility in cooperating with others to solve problems; the managerial type mainly examines the degree of orientation towards self-drive, planning management, problem solving, communication and coordination, and flexibility; the secretarial type mainly examines the degree of orientation towards open-mindedness, careful thinking and firm will; the meticulous type mainly examines the degree of orientation towards carefulness, rigor, attention to details and patience.

[0041] In the career orientation assessment methods in related technologies, the subjects are usually assessed through assessment questionnaires, but the assessment is too subjective and there is a possibility that the assessment results are not filled out truthfully. If machine learning, deep learning and other methods are used, the labeling of career orientation usually requires complex questionnaires or the participation of professionals, which makes the manpower and time cost of labeling data high, resulting in insufficient data with career orientation labels, making the career orientation assessment methods using machine learning and deep learning more susceptible to interference from noise data, and then overfitting or underfitting problems occur.

[0042] An embodiment of the present disclosure provides a method for training a career orientation prediction model, including: obtaining a heterogeneous graph sample set, wherein the heterogeneous graph sample set includes at least one heterogeneous graph sample, the heterogeneous graph sample includes multiple nodes and connecting edges, the nodes are constructed based on object samples, texts published by the object samples on the Internet, or preset text categories, the preset text categories are categories to which preset words related to career orientation in a target dictionary belong, and the connecting edges are used to characterize the association relationship between multiple nodes; training the career orientation prediction model based on the heterogeneous graph sample set to obtain a trained career orientation prediction model.

[0043] Figure 1 The application scenario diagram of the career orientation prediction model training method, career orientation prediction method, device, equipment, medium and program product according to the embodiments of the present disclosure is schematically shown.

[0044] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used to provide a medium for a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0045] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0046] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0047] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0048] It should be noted that the career orientation prediction model method and the career orientation prediction method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the career orientation prediction model device and the career orientation prediction device provided in the embodiment of the present disclosure can generally be set in the server 105. The career orientation prediction model method and the career orientation prediction method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the career orientation prediction model device and the career orientation prediction device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0049] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0050] The following will be based on Figure 1 The scene described by Figure 2~Figure 8 The career orientation prediction model method and career orientation prediction method of the disclosed embodiment are described in detail.

[0051] Figure 2 A flowchart schematically shows a method for training a career orientation prediction model according to an embodiment of the present disclosure.

[0052] like Figure 2 As shown, the method includes operations S210 to S220.

[0053] In operation S210, a heterogeneous graph sample set is obtained, wherein the heterogeneous graph sample set includes at least one heterogeneous graph sample, the heterogeneous graph sample includes multiple nodes and connecting edges, the nodes are constructed based on object samples, texts published by the object samples on the Internet, or preset text categories, the preset text categories are categories to which preset words related to career orientation in the target dictionary belong, and the connecting edges are used to characterize the association relationship between the multiple nodes.

[0054] In operation S220, the career orientation prediction model is trained based on the heterogeneous graph sample set to obtain a trained career orientation prediction model.

[0055] According to an embodiment of the present disclosure, the object sample may be a posting user on social media, and the text published by the object sample on the network may be a posting text on social media.

[0056] According to an embodiment of the present disclosure, the preset category can be a category to which words in the target dictionary that are determined to have a high correlation with career orientation based on the meaning, part of speech, etc. of the preset words belong, and the preset category can be as follows: positive adjectives, corresponding to preset words such as: worthy, willing, just right, etc.; negative adjectives, corresponding to preset words such as: at a loss, unhappy, none, etc.; anxiety, corresponding to preset words such as: worry, bad, upset, etc.; anger, corresponding to preset words such as: malice, insult, exasperated, etc.; interpersonal communication, corresponding to preset words such as: human touch, acquaintance, etc.

[0057] According to an embodiment of the present disclosure, the target dictionary may be an emotion dictionary consisting of words representing emotions and cognition mined from aspects such as sociology, health science, and psychology. The target dictionary may be pre-constructed by mining the above words, or may be an existing emotion dictionary. Specifically, it may be an emotion dictionary similar to the C-LIWC dictionary (Chinese Linguistic Inquiry and Word Count).

[0058] According to an embodiment of the present disclosure, a heterogeneous graph sample may be a topological graph having nodes of various categories, which is composed of various information such as object samples, texts published by the object samples on the Internet, and preset text categories.

[0059] According to the embodiments of the present disclosure, there is no limitation on the implementation method of the career orientation prediction model, which can be implemented through graph convolutional neural networks (GCNs for short).

[0060] According to the embodiments of the present disclosure, heterogeneous graph samples can effectively represent user information, text information published by users, and preliminary career orientation ability information. By training the career orientation prediction model with a heterogeneous graph sample set, the trained career orientation prediction model can dynamically aggregate text information published by users of the same type through nodes obtained from preset text categories, thereby predicting the corresponding career orientation results.

[0061] According to the training method of the career orientation prediction model disclosed in the present invention, the career orientation prediction model is trained by using a heterogeneous graph sample set to obtain a trained career orientation prediction model. Since the heterogeneous graph sample set includes at least one heterogeneous graph sample, and multiple nodes in the heterogeneous graph sample are constructed based on the object sample, the text published by the object sample on the Internet, and the preset text category, the trained career orientation prediction model can realize the mining of the text published by the object sample on the Internet and the dynamic aggregation of the text belonging to the same preset text category, so that the trained career orientation prediction model can output more accurate prediction results. Therefore, it at least partially solves the technical problem in the related art that the accuracy of career orientation assessment is low due to the possibility that the test questionnaire may not be filled out truthfully, and achieves the technical effect of being able to more accurately predict the career orientation type of the object to be tested through the trained career orientation prediction model.

[0062] According to an embodiment of the present disclosure, there are multiple preset text categories; the heterogeneous graph samples are generated in the following manner.

[0063] The text published on the Internet by the object sample is segmented and POS tagged to obtain multiple target words, wherein each of the multiple target words has a POS attribute; for each target word, based on the target word, the POS attribute of the target word and the preset words corresponding to each of the multiple preset text categories, a target text category associated with the target word is determined from the multiple preset text categories; based on the object sample, the multiple target words and the multiple preset text categories, multiple nodes are constructed; based on the relevance between each of the multiple target words, the target text category associated with the target word and the association between the object sample and each of the multiple target words, connecting edges between each of the multiple nodes are determined; based on the multiple nodes, the connecting edges between each of the multiple nodes and the weight values ​​corresponding to the connecting edges, a heterogeneous graph sample is obtained.

[0064] According to an embodiment of the present disclosure, the text published by the subject sample on the network may be obtained when the subject sample agrees or authorizes that the text published by the subject sample on the network may be obtained.

[0065] According to an embodiment of the present disclosure, the above-mentioned text may refer to sentences, words, etc. published by a user.

[0066] According to an embodiment of the present disclosure, when a text of an object sample published on the Internet is obtained, data access, cleaning and data preprocessing operations can be performed on the text, and data preprocessing includes removing irrelevant content such as numbers, stop words, punctuation marks, etc.

[0067] According to the embodiments of the present disclosure, there is no limitation on the specific implementation method of word segmentation and part-of-speech tagging of text, and the jieba word segmentation tool can be used.

[0068] According to an embodiment of the present disclosure, the association relationship between a target word and a preset text category, and between a target word and an object sample may be a subordinate relationship, and the association relationship between target words may be a correlation relationship, and the higher the correlation, the more related the two target words are considered to be.

[0069] According to an embodiment of the present disclosure, if a target word has an association relationship with a preset text category, a connecting edge is established between the target word and the nodes obtained from the preset text category; if the target word is a word in the text published by the object sample, a connecting edge is established between the target word and the nodes obtained from the object sample; if the correlation between the target word and the target word is greater than the preset correlation, a connecting edge is established between the nodes obtained from the above two target words, and the preset correlation is not limited and can be 0.

[0070] According to the embodiment of the present disclosure, by combining the target dictionary, a career orientation heterogeneous graph containing three nodes is constructed, including object nodes representing posting users, target word nodes represented by keywords in the text posted by users, and text category nodes related to career orientation features. For each object sample, the heterogeneous graph can be represented as ,in, is a collection of nodes, represents the edge between nodes. Here, Represents an object node, represents the target word node set, Represents a set of text category nodes, represents the p target word nodes determined by the p target words appearing in the texts published by the user on the Internet, is the q text category node determined by the q text categories selected from the target dictionary.

[0071] According to an embodiment of the present disclosure, each connecting edge may have a corresponding weight value, which is preset.

[0072] According to an embodiment of the present disclosure, based on a target word, a part-of-speech attribute of the target word, and a plurality of preset text categories respectively corresponding to the preset word, determining a target text category associated with the target word from a plurality of preset text categories may include the following operations.

[0073] Based on the part-of-speech attribute of the target word, a candidate text category is determined from multiple preset text categories; the target word is matched with the preset word corresponding to the candidate text category to obtain a matching result; when the matching result indicates that the target word matches the preset word, the candidate text category is determined to be a target text category that has an associated relationship with the target word.

[0074] According to an embodiment of the present disclosure, each preset text category may have its own part-of-speech attribute, which may be reflected in the identifier of the preset text category or by constructing a mapping relationship between the preset text category and the part-of-speech attribute.

[0075] According to an embodiment of the present disclosure, at least one candidate text category matching the part-of-speech attribute of the target word can be first screened from multiple preset text categories by using the part-of-speech attribute of the target word. Then, multiple preset words corresponding to the candidate text category are obtained from a storage unit, and the multiple preset words are matched with the target word. If it is determined that there is a preset word identical to the target word among the multiple preset words, the candidate text category is considered to be a target text category associated with the target word.

[0076] According to an embodiment of the present disclosure, by constructing part-of-speech attribute tags for both target words and preset text categories, the matching speed of target words and preset text categories can be improved, thereby improving the speed of determining whether multiple target words and multiple preset text categories have an association relationship.

[0077] According to an embodiment of the present disclosure, the relevance between each of the plurality of target words is determined in the following manner.

[0078] For each target word, determine at least one remaining word other than the target word; for each remaining word, determine the correlation between the target word and the remaining words based on a first probability that the target word and the remaining words appear at the same time, a second probability that the target word appears, and a third probability that the remaining words appear.

[0079] According to an embodiment of the present disclosure, the remaining words may be all target words among the multiple target words except the current target word.

[0080] According to an embodiment of the present disclosure, the probability of occurrence of each of the plurality of target words may be predetermined, and when calculating the relevance, the second probability of occurrence of the target word and the third probability of occurrence of each of the remaining words may be determined from the above probabilities.

[0081] According to an embodiment of the present disclosure, specifically, for any two target words and , the above correlation can be calculated by the following formula (1).

[0082]

[0083] in, represents the probability of two target words appearing at the same time, and Respectively represent the target word and target word Probability of occurrence.

[0084] According to an embodiment of the present disclosure, if PMI>0, it means that the two target words are related, and the larger the PMI value, the higher the correlation; if PMI=0, it means that the two target words are independent of each other and have no correlation; if PMI<0, it means that the two target words are unrelated and mutually exclusive.

[0085] According to the embodiments of the present disclosure, by calculating the relevance between each target word, the relevance of the target words can be quantified so that the subsequent career orientation prediction model can learn more effective information.

[0086] According to an embodiment of the present disclosure, when a heterogeneous graph sample set includes multiple heterogeneous graph samples, the multiple heterogeneous graph samples include an unlabeled sample and at least two labeled samples; the at least two labeled samples include a first labeled sample and a second labeled sample; training a career orientation prediction model based on the heterogeneous graph sample set to obtain a trained career orientation prediction model may include the following operations.

[0087] The career orientation prediction model is trained based on the first labeled sample to obtain a pre-trained career orientation prediction model; disturbance is added to the unlabeled sample to obtain an enhanced sample; the pre-trained career orientation prediction model is trained based on the enhanced sample, the unlabeled sample and the second labeled sample to obtain a trained career orientation prediction model.

[0088] According to the embodiments of the present disclosure, there is no limitation on the manner of adding disturbances to the unlabeled samples, and the disturbances may be added by performing random deletion or synonym replacement in the unlabeled samples.

[0089] According to an embodiment of the present disclosure, for each unlabeled sample , an enhanced sample can be generated by randomly perturbing it , thus constructing a sample pair .

[0090] According to an embodiment of the present disclosure, the number of the first labeled sample, the second labeled sample, and the number of the unlabeled sample may be multiple.

[0091] According to an embodiment of the present disclosure, semi-supervised training of a career orientation prediction model is implemented through first labeled data, enhanced samples, unlabeled samples, and second labeled samples. This can make use of limited labeled data and large-scale unlabeled data to improve the prediction accuracy of the model without increasing the cost of manual labeling.

[0092] According to an embodiment of the present disclosure, training a pre-trained career orientation prediction model based on an enhanced sample, an unlabeled sample, and a second labeled sample to obtain a trained career orientation prediction model may include the following operations.

[0093] The unlabeled samples, the enhanced samples and the second labeled samples are respectively vectorized to obtain a first vector representation of the unlabeled samples, a second vector representation of the enhanced samples and a third vector representation of the second labeled samples; the first vector representation is input into a pre-trained career orientation prediction model to obtain a first prediction result; the second vector representation is input into the pre-trained career orientation prediction model to obtain a second prediction result; the third vector representation is input into the pre-trained career orientation prediction model to obtain a third prediction result; based on the first prediction result and the second prediction result, a first loss value is obtained; based on the third prediction result and the label data corresponding to the second labeled sample, a second loss value is obtained; based on the first loss value, the preset weight corresponding to the first loss value and the second loss value, a target loss value is obtained; based on the target loss value, the parameters of the pre-trained career orientation prediction model are adjusted to obtain a trained career orientation prediction model.

[0094] According to an embodiment of the present disclosure, there is no limitation on the manner of vectorizing the unlabeled samples, enhanced samples, and second labeled samples, and they may be obtained through an embedding layer, which may be implemented by a vectorization model such as word2vec and BERT.

[0095] According to an embodiment of the present disclosure, when vectorizing any one of the unlabeled samples, the enhanced samples, and the second labeled samples, multiple target word nodes can be vectorized first to generate a target word vector sub-matrix. When vectorizing the object nodes and the text category dictionary, the dimensions of the target word vector matrix can be aligned by random assignment to obtain the object vector sub-matrix and the text category vector sub-matrix, and the final representation vector corresponding to the entire sample is obtained through the target word vector sub-matrix, the object vector sub-matrix, and the text category vector sub-matrix.

[0096] According to an embodiment of the present disclosure, the mean square error may be selected to calculate the first loss value and the second loss value. In the case where there are multiple unlabeled samples or multiple second labeled samples, multiple first loss values ​​determined based on multiple unlabeled samples or multiple second loss values ​​determined based on multiple second labeled samples may be averaged to obtain the overall consistency loss.

[0097] According to the embodiments of the present disclosure, the preset weight is not limited and can be obtained according to the requirements. In some embodiments, it can be 5. By limiting the preset weight to 5 and adjusting the parameters of the model according to the obtained target loss value, a better trained career orientation prediction model can be obtained.

[0098] According to an embodiment of the present disclosure, the ultimate optimization goal of the pre-trained career orientation prediction model is to minimize the target loss value.

[0099] According to an embodiment of the present disclosure, for each unlabeled sample pair , respectively calculate the predicted values ​​of career orientation under the original input and the perturbed input, that is, the first prediction result And the second prediction result The difference between the two prediction results is constrained by consistency regularization, that is, a first loss value is obtained based on the two prediction results, and the first loss value is used as part of the final target loss value to adjust the parameters of the pre-trained career orientation prediction model, thereby realizing the comparison of the prediction results of the model for the same sample before and after data enhancement, and based on this penalty model inconsistency, the model produces similar outputs for the same set of inputs, thereby improving the model's anti-interference ability to a certain extent.

[0100] According to an embodiment of the present disclosure, by training the model based on a second loss value obtained based on a second labeled sample and combining a preset weight with a target loss value obtained by the first loss value, the overfitting problem can be avoided to a certain extent.

[0101] According to the embodiments of the present disclosure, an overall loss function is formed by combining labeled sample data and large-scale unlabeled data, and the target loss value obtained by the overall loss function is used to update the model parameters. That is, under the consistency constraint, the combination of supervised training and unsupervised training is realized, thereby obtaining a more accurate trained career orientation prediction model, wherein the supervised training of the pre-trained career orientation prediction model based on the second labeled sample can help reduce the impact of input noise.

[0102] According to the embodiments of the present disclosure, through the above steps, the model can generate similar outputs for the same set of inputs. This unlabeled data regularization technology penalizes the inconsistency of the model by comparing the prediction results of the model before and after the disturbance, thereby reducing the sensitivity of the model to noise and random factors. It not only reduces the risk of overfitting, but also improves the accuracy of the career orientation prediction model.

[0103] According to an embodiment of the present disclosure, the multiple nodes include a target word node and a text category node; the target word node is a node determined by the target word, and the text category node is a node determined by a preset text category; adding perturbations to unlabeled samples to obtain enhanced samples may include the following operations.

[0104] From the target word nodes included in the unlabeled sample, determine the node to be replaced, wherein the node to be replaced is obtained from the target word to be replaced; based on the synonyms of the target word to be replaced determined from the target dictionary, replace the node to be replaced in the unlabeled sample with a synonym node obtained from the synonym to obtain a first sub-enhanced sample; delete the target deletion node in the first sub-enhanced sample or the unlabeled sample to obtain a second sub-enhanced sample, wherein the target deletion node is determined from the target word nodes having a connection edge with the text category node; based on the first sub-enhanced sample and / or the second sub-enhanced sample, obtain an enhanced sample.

[0105] According to an embodiment of the present disclosure, the node to be replaced may be replaced with a synonym node to obtain a first sub-enhanced sample, and based on the first sub-enhanced sample, a target deletion node may be determined from the first sub-enhanced sample and the target deletion node may be deleted to obtain a second sub-enhanced sample.

[0106] According to an embodiment of the present disclosure, a target deletion node may be determined in the unlabeled sample, and the target deletion node may be deleted in the unlabeled sample, thereby obtaining a second sub-enhanced sample.

[0107] According to an embodiment of the present disclosure, the target deletion node may be randomly determined from target word nodes having a connection edge with the text category node.

[0108] According to an embodiment of the present disclosure, the enhancement sample may include a first sub-enhancement sample and a second sub-enhancement sample, or may be the first sub-enhancement sample or the second sub-enhancement sample.

[0109] According to an embodiment of the present disclosure, random perturbation of unlabeled samples can be achieved by performing synonym replacement and / or random deletion on unlabeled samples, and the pre-trained career orientation prediction model can be trained by obtaining sample pairs, so as to improve the robustness of the trained career orientation prediction model obtained by subsequent training.

[0110] Figure 3 The diagram schematically shows a schematic diagram of adding disturbance to an unlabeled sample to obtain an enhanced sample according to an embodiment of the present disclosure.

[0111] like Figure 3As shown, adding disturbances to the unlabeled sample may include two methods: synonym replacement and random deletion. Specifically, during the synonym replacement process, the node to be replaced may be randomly determined in the unlabeled sample 310, such as Figure 3 The node representing attention to details or the node obtained by the target word: attention to details is replaced by a node representing a synonym of attention to details, such as: a node representing perfectionism, thereby obtaining enhanced sample 320.

[0112] According to an embodiment of the present disclosure, in the process of random deletion, one or more nodes representing patience, careful thinking, strong hands-on ability, etc. in the unlabeled sample 330 can be randomly deleted, such as deleting the node representing careful thinking, thereby obtaining an enhanced sample 340.

[0113] According to an embodiment of the present disclosure, the pre-trained career orientation prediction model includes: a convolutional layer, an attention layer and a fully connected layer; the multiple nodes also include: an object node; the object node is a node determined by an object sample; inputting the first vector representation into the pre-trained career orientation prediction model to obtain a first prediction result may include the following operations.

[0114] The first vector representation is input into the convolution layer to obtain a feature vector that has learned N-order neighborhood information, wherein N is a positive integer greater than or equal to 1, and the feature vector includes a vector representation of a target object node that has learned the target word node in the N-order neighborhood and a vector representation of a target text category node that has learned the target word node in the N-order neighborhood; the feature vector is input into the attention layer, and attention processing is performed on the M target text category nodes that are most associated with the target object node to obtain a processed feature vector, wherein M is a positive integer greater than or equal to 1; the processed feature vector is input into the fully connected layer, and the processed feature vector is mapped to obtain a first prediction result, wherein the prediction result includes the matching probability of the object sample corresponding to the unlabeled sample and each of the multiple occupational orientation types.

[0115] According to an embodiment of the present disclosure, the convolution layer may include multiple convolution sublayers, each convolution sublayer extracts first-order neighborhood information, and the convolution layer may be a graph convolution layer.

[0116] According to an embodiment of the present disclosure, the transmission of multi-level neighborhood information can be achieved by stacking multiple convolutional sublayers, and the mathematical formula for its propagation and conversion is shown in the following formula (2).

[0117]

[0118] Wherein, A represents a matrix after normalization and symmetry processing. In some embodiments, the normalization and symmetry operation of the matrix can use the Laplacian matrix to perform convolution operations to achieve feature extraction and reduce computational complexity. is the feature vector output by the convolutional sublayer of layer k, is the weight parameter of this layer, is the activation function, and the feature vector output by the convolutional sublayer of the k+1th layer is obtained after calculation In some embodiments, a ReLU function may be used as an activation function.

[0119] Figure 4 The structural diagram of the career orientation prediction model according to the embodiment of the present disclosure is schematically shown.

[0120] like Figure 4 As shown, taking the convolution layer including two convolution sub-layers as an example, the career orientation prediction model includes two convolution sub-layers, an attention layer and a fully connected layer.

[0121] According to an embodiment of the present disclosure, by inputting a heterogeneous graph sample into an embedding layer for vectorization, a vectorized representation of the heterogeneous graph sample is obtained, and by inputting the vectorized representation of the heterogeneous graph sample into a career orientation prediction model, the matching probability of the object sample and each of the multiple career orientations can be obtained. The object sample is an object sample corresponding to the object node included in the heterogeneous graph sample. The specific types of the multiple career orientation types are not limited, and different career orientation types can be divided in different scenarios. In some embodiments, it can include operational type, exploratory type, cooperative type, management type, secretarial type, and fine type, etc.

[0122] According to the embodiments of the present disclosure, N layers of convolutional sublayers can be used to avoid overfitting problems while ensuring that the model has sufficient feature learning capabilities, thereby effectively predicting the user's career orientation. The first convolutional sublayer integrates the first-order neighbor information of each node of the heterogeneous graph sample, and when calculating the second layer, the neighbor information of the neighbor node, that is, the second-order neighbor information, can also be included, so that the career orientation prediction model can learn more career orientation features. The calculation formula is as follows:

[0123]

[0124]

[0125] in, is the activation function, which can be the ReLU function, A is the Laplace matrix, is the input feature matrix, i.e. the vectorized heterogeneous graph sample, is the parameter of the first convolutional sublayer, and the intermediate features are obtained by calculation As the input of the second convolutional sublayer, is the parameter of the second convolutional sublayer, and finally the representation vector of the graph convolutional layer is obtained After two layers of graph convolutional layers, the vector representation of the target object node that has learned the target word node in the second-order neighborhood and the vector representation of the target text category node that has learned the target word node in the second-order neighborhood are obtained.

[0126] According to an embodiment of the present disclosure, the convolution layer may be followed by an attention layer, so that the target object node can obtain representation from the target text category node, effectively improving the prediction ability of the career orientation prediction model.

[0127] According to an embodiment of the present disclosure, in the attention layer, M target text category nodes with the highest correlation with the target object node can be selected. The selection method is not limited and can be determined by determining the preset text category in the text published by the object sample. Specifically, it can be determined by the connecting lines between multiple text category nodes and multiple target word nodes in the heterogeneous graph sample. The top M text category nodes with the highest number of connecting lines with the target word node can be determined as the M target text category nodes with the highest correlation with the target word node.

[0128] According to an embodiment of the present disclosure, the formula for performing attention processing on the M target text category nodes that are most associated with the target object node may be as shown in the following formulas (5) to (6).

[0129]

[0130]

[0131]

[0132] in, It is a learnable weight parameter matrix responsible for performing linear transformation operations on the node features of the target text category nodes and the node features of the target object nodes; It is the vector representation of the target object node obtained after the convolution layer; represents the i-th target text category node with the highest correlation with the target object node, which is the vector representation calculated by the convolution layer; is the activation function, which can be the LeakyReLU function. is the attention weight. The original attention score between the target text category node and the target object node is calculated by linearly transforming the node features of the target text category node and the node features of the target object node. Then, the dot product operation is performed on the concatenated features and the learnable weight matrix, and the softmax operation is applied to the original attention scores of all adjacent edges to obtain the attention weight. . is the weight parameter matrix. After all adjacent node features are weighted summed based on attention, they are calculated through the tanh activation function to obtain the representation of the corresponding dimension. .

[0133] Figure 5 The diagram schematically shows the attention calculation diagram of the M target text category nodes that are most highly associated with the target object node in the attention layer according to an embodiment of the present disclosure.

[0134] like Figure 5 As shown, the node represents In the above example, the object node and the text category node have learned the features of the target word node through the convolution layer, that is, the target object node and the target text category node are obtained. In order to further obtain more information from the text category node, the M target text category nodes with the highest correlation are selected for each object node, and the attention mechanism is used to enable the object node to effectively obtain relevant information. Figure 5 Six target text categories are schematically selected, and the attention weights are obtained through the attention mechanism. , i=1,2,...q.

[0135] And perform attention-based weighted summation of the vector representations of all target text category nodes as shown in Formula 7 to obtain the processed feature vector .

[0136] According to an embodiment of the present disclosure, in some embodiments, M may be equal to , and is equal to the number of occupational orientation categories.

[0137] According to the embodiments of the present disclosure, the fully connected layer is used as the final layer of the career orientation prediction model. The feature vector processed by the graph convolution layer and the attention layer is transformed in dimension to map the final output result. The calculation formula can be shown as the following formula (8).

[0138]

[0139] in, is the trainable weight matrix, is the feature vector output by the attention layer, is the offset, and after the Sigmoid function is calculated, it is the output result of the fully connected layer. , its dimension can be determined by the number of career orientation categories. For example, if there are six types of career orientation, the output dimension is 6.

[0140] According to the embodiments of the present disclosure, the model structures of the career orientation prediction model, the pre-trained career orientation prediction model, and the trained career orientation prediction model are the same, that is, they all include a convolutional layer, an attention layer, and a fully connected layer, and the model parameters are updated during the training process. In addition, the above three models also process the input data in the same way, and can all be performed as above. Figure 4 Processing shown.

[0141] Figure 6 The model training structure diagram of the career orientation prediction model training method according to the embodiment of the present disclosure is schematically shown.

[0142] like Figure 6 As shown, the career orientation prediction model can be trained using the first labeled sample to obtain a pre-trained career orientation prediction model. During the training process, the first labeled sample can be vectorized to obtain a vectorized representation of the first labeled sample, and then the parameters of the career orientation prediction model can be adjusted to obtain a pre-trained career orientation prediction model. The pre-trained career orientation prediction model can output the matching probability of the sample object of the first labeled sample with multiple career orientation types.

[0143] According to an embodiment of the present disclosure, by sharing parameters, unlabeled samples and enhanced samples obtained by adding perturbations to the unlabeled samples are trained on the basis of a pre-trained career orientation prediction model, and a trained career orientation prediction model is obtained through consistency regularization constraints.

[0144] According to the embodiments of the present disclosure, in some embodiments, the pre-trained career orientation prediction model can also be trained based on a combination of enhanced samples, unlabeled samples and second labeled samples to obtain a trained career orientation prediction model.

[0145] Figure 7 A data flow diagram of a method for training a career orientation prediction model according to yet another embodiment of the present disclosure is schematically shown.

[0146] like Figure 7 As shown, data preprocessing is performed on the texts published on the Internet by multiple object samples to obtain multiple target texts 701. Based on the multiple target texts 701, multiple object samples 702 and multiple preset text categories 703, heterogeneous graph samples corresponding to each of the multiple object samples are obtained, that is, a heterogeneous graph sample set 704 is obtained. Perturbations are added to the unlabeled samples 705 included in the heterogeneous graph sample set 704 to obtain enhanced samples 708. Based on the first labeled samples 706 included in the heterogeneous graph sample set 704, a career orientation prediction model 709 is trained to obtain a pre-trained career orientation prediction model 710.

[0147] According to an embodiment of the present disclosure, the unlabeled sample 705, the enhanced sample 708, and the second labeled sample 707 included in the heterogeneous graph sample set 704 are respectively vectorized to obtain a first vector representation of the unlabeled sample 705, a second vector representation of the enhanced sample 708, and a third vector representation of the second labeled sample 707. The first vector representation of the unlabeled sample 705 is input into the pre-trained career orientation prediction model 710 to obtain a first prediction result. The second vector representation of the enhanced sample 708 is input into the pre-trained career orientation prediction model 710 to obtain a second prediction result. The third vector representation of the second labeled sample 707 is input into the pre-trained career orientation prediction model 710 to obtain a third prediction result. Based on the first prediction result and the second prediction result, a first loss value 711 is obtained. Based on the third prediction result and the label data corresponding to the second labeled sample 707, a second loss value 712 is obtained. The target loss value is obtained based on the first loss value 711, the preset weight corresponding to the first loss value 711, and the second loss value 712.

[0148] According to an embodiment of the present disclosure, parameters of the pre-trained career orientation prediction model 710 are adjusted based on the target loss value to obtain a trained career orientation prediction model 713.

[0149] Figure 8 The flowchart of the career orientation prediction method according to the embodiment of the present disclosure is schematically shown.

[0150] like Figure 8 As shown, the method includes operations S810 to S830.

[0151] In operation S810, a heterogeneous graph for the object to be tested is obtained, wherein the heterogeneous graph includes multiple nodes and connecting edges, the nodes are constructed based on the object to be tested, texts published by the object to be tested on the Internet, or preset text categories, the preset text categories are categories to which preset words related to career orientation in the target dictionary belong, and the edges are used to characterize the association relationships between the multiple nodes.

[0152] In operation S820, the heterogeneous graph is vectorized to obtain a vector representation of the heterogeneous graph.

[0153] In operation S830, the vector representation of the heterogeneous graph is input into a trained career orientation prediction model to obtain a career orientation prediction result for the subject to be tested.

[0154] According to an embodiment of the present disclosure, the process of determining a heterogeneous graph may be the same as the process of determining a heterogeneous graph sample.

[0155] According to an embodiment of the present disclosure, a heterogeneous graph may be input into an embedding layer for vectorization.

[0156] According to the embodiment of the present disclosure, the processing process of the trained career orientation prediction model for the heterogeneous graph is the same as the processing process of the trained career orientation prediction model for the heterogeneous graph sample, that is, Figure 5 The processing shown is the same.

[0157] According to an embodiment of the present disclosure, the career orientation prediction result for the subject to be tested includes the matching probability between the subject to be tested and each of the multiple career orientation types.

[0158] Based on the above-mentioned career orientation prediction model training method, the present disclosure also provides a career orientation prediction model training device. Fig. 9 The device is described in detail.

[0159] Fig. 9 The structural block diagram of the training device of the career orientation prediction model according to the embodiment of the present disclosure is schematically shown.

[0160] like Fig. 9 As shown, the training device 900 of the career orientation prediction model includes a first acquisition module 910 and a training module 920 .

[0161] The first acquisition module 910 is used to acquire a heterogeneous graph sample set, wherein the heterogeneous graph sample set includes at least one heterogeneous graph sample, and the heterogeneous graph sample includes multiple nodes and connecting edges. The nodes are constructed based on object samples, texts published by the object samples on the Internet, or preset text categories. The preset text categories are categories to which preset words related to career orientation in the target dictionary belong, and the connecting edges are used to characterize the association relationship between multiple nodes.

[0162] The training module 920 is used to train the career orientation prediction model based on the heterogeneous graph sample set to obtain a trained career orientation prediction model.

[0163] According to an embodiment of the present disclosure, there are multiple preset text categories. The training device 900 for the career orientation prediction model further includes: a processing module, a relationship determination module, a node determination module, an edge determination module, and a sample determination module.

[0164] The processing module is used to perform word segmentation and part-of-speech tagging on the text of the object sample published on the Internet to obtain multiple target words, wherein each of the multiple target words has a part-of-speech attribute.

[0165] The relationship determination module is used to determine, for each target word, a target text category associated with the target word from multiple preset text categories based on the target word, the part-of-speech attribute of the target word, and multiple preset text categories corresponding to each preset word.

[0166] The node determination module is used to construct multiple nodes based on the object sample, multiple target words and multiple preset text categories.

[0167] The edge determination module is used to determine the connection edges between multiple nodes based on the relevance between multiple target words, the target text category associated with the target words, and the association between the object sample and the multiple target words.

[0168] The sample determination module is used to obtain heterogeneous graph samples based on multiple nodes, connecting edges between the multiple nodes, and weight values ​​corresponding to the connecting edges.

[0169] According to an embodiment of the present disclosure, the relationship determination module includes: a first determination submodule, a matching submodule and a second determination submodule.

[0170] The first determination submodule is used to determine a candidate text category from a plurality of preset text categories based on the part-of-speech attribute of the target word.

[0171] The matching submodule is used to match the target word with the preset word corresponding to the candidate text category to obtain a matching result.

[0172] The second determination submodule is configured to determine the candidate text category as a target text category associated with the target word when the matching result indicates that the target word matches the preset word.

[0173] According to an embodiment of the present disclosure, the training device 900 for the career orientation prediction model further includes: a word determination module and a relevance determination module.

[0174] The word determination module is used to determine at least one remaining word other than the target word for each target word.

[0175] The relevance determination module is used to determine the relevance between the target word and the remaining words for each remaining word based on a first probability that the target word and the remaining words appear at the same time, a second probability that the target word appears, and a third probability that the remaining words appear.

[0176] According to an embodiment of the present disclosure, when the heterogeneous graph sample set includes multiple heterogeneous graph samples, the multiple heterogeneous graph samples include unlabeled samples and at least two labeled samples. The at least two labeled samples include a first labeled sample and a second labeled sample. The training module 920 includes: a first training submodule, a disturbance adding submodule, and a second training submodule.

[0177] The first training submodule is used to train the career orientation prediction model based on the first labeled sample to obtain a pre-trained career orientation prediction model.

[0178] The perturbation adding submodule is used to add perturbations to unlabeled samples to obtain enhanced samples.

[0179] The second training submodule is used to train the pre-trained career orientation prediction model based on the enhanced samples, the unlabeled samples and the second labeled samples to obtain a trained career orientation prediction model.

[0180] According to an embodiment of the present disclosure, the plurality of nodes include a target word node and a text category node. The target word node is a node determined by the target word, and the text category node is a node determined by a preset text category. The disturbance adding submodule includes: a replacement node determination unit, a replacement unit, a deletion unit, and a sample determination unit.

[0181] The replacement node determination unit is used to determine a node to be replaced from the target word nodes included in the unlabeled sample, wherein the node to be replaced is obtained from the target word to be replaced.

[0182] The replacement unit is used to replace the node to be replaced in the unlabeled sample with a synonym node obtained from the synonym based on the synonym of the target word to be replaced determined from the target dictionary, so as to obtain a first sub-enhanced sample.

[0183] The deleting unit is used to delete the target deletion node in the first sub-enhanced sample or the unlabeled sample to obtain the second sub-enhanced sample, wherein the target deletion node is determined from the target word nodes that have a connection edge with the text category node.

[0184] The sample determination unit is configured to obtain an enhanced sample based on the first sub-enhanced sample and / or the second sub-enhanced sample.

[0185] According to an embodiment of the present disclosure, the second training submodule includes: a representation determination unit, a first prediction unit, a second prediction unit, a third prediction unit, a first loss determination unit, a second loss determination unit, and a target loss determination unit.

[0186] The representation determination unit is used to vectorize the unlabeled sample, the enhanced sample and the second labeled sample respectively to obtain a first vector representation of the unlabeled sample, a second vector representation of the enhanced sample and a third vector representation of the second labeled sample.

[0187] The first prediction unit is used to input the first vector representation into a pre-trained career orientation prediction model to obtain a first prediction result.

[0188] The second prediction unit is used to input the second vector representation into the pre-trained career orientation prediction model to obtain a second prediction result.

[0189] The third prediction unit is used to input the third vector representation into the pre-trained career orientation prediction model to obtain a third prediction result.

[0190] The first loss determining unit is used to obtain a first loss value based on the first prediction result and the second prediction result.

[0191] The second loss determination unit is used to obtain a second loss value based on the third prediction result and the label data corresponding to the second labeled sample.

[0192] The target loss determination unit is used to obtain a target loss value based on the first loss value, a preset weight corresponding to the first loss value, and a second loss value.

[0193] The training unit is used to adjust the parameters of the pre-trained career orientation prediction model based on the target loss value to obtain a trained career orientation prediction model.

[0194] According to an embodiment of the present disclosure, the pre-trained career orientation prediction model includes: a convolution layer, an attention layer and a fully connected layer. The multiple nodes also include: an object node. The object node is a node determined by an object sample. The first prediction unit includes: a convolution subunit, an attention subunit and a mapping subunit.

[0195] A convolution subunit is used to input the first vector representation into the convolution layer to obtain a feature vector that has learned N-order neighborhood information, wherein N is a positive integer greater than or equal to 1, and the feature vector includes a vector representation of a target object node that has learned the target word node in the N-order neighborhood and a vector representation of a target text category node that has learned the target word node in the N-order neighborhood.

[0196] The attention subunit is used to input the feature vector into the attention layer, perform attention processing on the M target text category nodes that are most associated with the target object node, and obtain the processed feature vector, where M is a positive integer greater than or equal to 1.

[0197] A mapping subunit is used to input the processed feature vector into the fully connected layer, map the processed feature vector, and obtain a first prediction result, wherein the prediction result includes the matching probability of the object sample corresponding to the unlabeled sample and each of the multiple occupational orientation types.

[0198] Fig.10 The structural block diagram of the career orientation prediction device according to an embodiment of the present disclosure is schematically shown.

[0199] like Fig.10 As shown, the career orientation prediction device 1000 includes a second acquisition module 1010 , a vectorization module 1020 and a result determination module 1030 .

[0200] The second acquisition module 1010 is used to obtain a heterogeneous graph for the object to be tested, wherein the heterogeneous graph includes multiple nodes and connecting edges, the nodes are constructed based on the object to be tested, the text published by the object to be tested on the Internet, or a preset text category, the preset text category is the category to which the preset words related to career orientation in the target dictionary belong, and the edges are used to characterize the association relationship between the multiple nodes.

[0201] The vectorization module 1020 is used to vectorize the heterogeneous graph to obtain a vector representation of the heterogeneous graph.

[0202] The result determination module 1030 is used to input the vector representation of the heterogeneous graph into the trained career orientation prediction model to obtain the career orientation prediction result for the subject to be tested.

[0203] According to an embodiment of the present disclosure, any multiple modules of the first acquisition module 910, the training module 920, the second acquisition module 1010, the vectorization module 1020 and the result determination module 1030 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the first acquisition module 910, the training module 920, the second acquisition module 1010, the vectorization module 1020 and the result determination module 1030 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in a suitable combination of any of them. Alternatively, at least one of the first acquisition module 910, the training module 920, the second acquisition module 1010, the vectorization module 1020 and the result determination module 1030 may be at least partially implemented as a computer program module, which may perform a corresponding function when executed.

[0204] Fig.11 A block diagram of an electronic device suitable for implementing a career orientation prediction model method and a career orientation prediction method according to an embodiment of the present disclosure is schematically shown.

[0205] like Fig.11As shown, the electronic device 1100 according to an embodiment of the present disclosure includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage part 1108 to a random access memory (RAM) 1103. The processor 1101 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include an onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0206] In RAM 1103, various programs and data required for the operation of electronic device 1100 are stored. Processor 1101, ROM 1102 and RAM 1103 are connected to each other through bus 1104. Processor 1101 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 1102 and / or RAM 1103. It should be noted that the program can also be stored in one or more memories other than ROM 1102 and RAM 1103. Processor 1101 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0207] According to an embodiment of the present disclosure, the electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to the bus 1104. The electronic device 1100 may further include one or more of the following components connected to the input / output (I / O) interface 1105: an input portion 1106 including a keyboard, a mouse, etc.; an output portion 1107 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 1108 including a hard disk, etc.; and a communication portion 1109 including a network interface card such as a LAN card, a modem, etc. The communication portion 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the input / output (I / O) interface 1105 as needed. A removable medium 1111, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1110 as needed, so that a computer program read therefrom is installed into the storage portion 1108 as needed.

[0208] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0209] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus, or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 1102 and / or RAM 1103 described above and / or one or more memories other than ROM 1102 and RAM 1103.

[0210] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program includes a program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the career orientation prediction model method and career orientation prediction method provided by the embodiment of the present disclosure.

[0211] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 1101. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0212] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 1109, and / or installed from a removable medium 1111. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0213] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1109, and / or installed from the removable medium 1111. When the computer program is executed by the processor 1101, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0214] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0215] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0216] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0217] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A training method for a career orientation prediction model, characterized in that: The method comprises: Acquire a heterogeneous graph sample set, wherein the heterogeneous graph sample set includes at least one heterogeneous graph sample, the heterogeneous graph sample includes a plurality of nodes and connecting edges, the nodes are constructed based on object samples, texts published by the object samples on the Internet, or preset text categories, the preset text categories are categories to which preset words related to career orientation in a target dictionary belong, and the connecting edges are used to characterize associations between the plurality of nodes; The career orientation prediction model is trained based on the heterogeneous graph sample set to obtain a trained career orientation prediction model.

2. The method according to claim 1, characterized in that There are multiple preset text categories; the heterogeneous graph samples are generated in the following way: Segmenting and part-of-speech tagging the text of the object sample published on the Internet to obtain a plurality of target words, wherein each of the plurality of target words has a part-of-speech attribute; For each of the target words, based on the target word, the part-of-speech attribute of the target word, and the preset words corresponding to the plurality of preset text categories, a target text category associated with the target word is determined from the plurality of preset text categories; Constructing a plurality of nodes based on the object sample, the plurality of target words and the plurality of preset text categories; Determining connection edges between each of the plurality of nodes based on the relevance between each of the plurality of target words, the target text category associated with the target words, and the association between the object sample and each of the plurality of target words; The heterogeneous graph sample is obtained based on a plurality of nodes, connection edges between the plurality of nodes, and weight values ​​corresponding to the connection edges.

3. The method according to claim 1, characterized in that The step of determining a target text category associated with the target word from the plurality of preset text categories based on the target word, the part-of-speech attribute of the target word, and the preset words corresponding to the plurality of preset text categories, comprises: Based on the part-of-speech attribute of the target word, determining a candidate text category from a plurality of preset text categories; Matching the target word with a preset word corresponding to the candidate text category to obtain a matching result; In a case where the matching result indicates that the target word matches the preset word, the candidate text category is determined to be a target text category associated with the target word.

4. The method according to claim 1, characterized in that: The relevance between the plurality of target words is determined in the following manner: For each of the target words, determining at least one remaining word other than the target word; For each of the remaining words, the relevance between the target word and the remaining words is determined based on a first probability that the target word and the remaining words appear at the same time, a second probability that the target word appears, and a third probability that the remaining words appear.

5. The method according to claim 1, characterized in that In the case where the heterogeneous graph sample set includes a plurality of the heterogeneous graph samples, the plurality of the heterogeneous graph samples include an unlabeled sample and at least two labeled samples; the at least two labeled samples include a first labeled sample and a second labeled sample; The step of training the career orientation prediction model based on the heterogeneous graph sample set to obtain a trained career orientation prediction model includes: Training the career orientation prediction model based on the first labeled sample to obtain a pre-trained career orientation prediction model; Add disturbance to the unlabeled sample to obtain an enhanced sample; The pre-trained career orientation prediction model is trained based on the enhanced samples, the unlabeled samples and the second labeled samples to obtain a trained career orientation prediction model.

6. The method according to claim 5, characterized in that The plurality of nodes include a target word node and a text category node; the target word node is a node determined by the target word, and the text category node is a node determined by the preset text category; Adding disturbance to the unlabeled sample to obtain an enhanced sample includes: Determine a node to be replaced from the target word nodes included in the unlabeled sample, wherein the node to be replaced is obtained from the target word to be replaced; Based on the synonyms of the target word to be replaced determined from the target dictionary, the node to be replaced in the unlabeled sample is replaced with a synonym node obtained from the synonym to obtain a first sub-enhanced sample; Deleting a target deletion node in the first sub-enhanced sample or the unlabeled sample to obtain a second sub-enhanced sample, wherein the target deletion node is determined from target word nodes having a connection edge with the text category node; The enhanced sample is obtained based on the first sub-enhancement sample and / or the second sub-enhancement sample.

7. The method according to claim 5, characterized in that The pre-trained career orientation prediction model is trained based on the enhanced sample, the unlabeled sample and the second labeled sample to obtain a trained career orientation prediction model, including: Vectorizing the unlabeled sample, the enhanced sample, and the second labeled sample respectively to obtain a first vector representation of the unlabeled sample, a second vector representation of the enhanced sample, and a third vector representation of the second labeled sample; Inputting the first vector representation into the pre-trained career orientation prediction model to obtain a first prediction result; Inputting the second vector representation into the pre-trained career orientation prediction model to obtain a second prediction result; Inputting the third vector representation into the pre-trained career orientation prediction model to obtain a third prediction result; Obtaining a first loss value based on the first prediction result and the second prediction result; Obtaining a second loss value based on the third prediction result and the label data corresponding to the second labeled sample; Obtaining a target loss value based on the first loss value, a preset weight corresponding to the first loss value, and the second loss value; The parameters of the pre-trained career orientation prediction model are adjusted based on the target loss value to obtain the trained career orientation prediction model.

8. The method according to claim 7, characterized in that The pre-trained career orientation prediction model includes: a convolution layer, an attention layer and a fully connected layer; the plurality of nodes also include: an object node; the object node is a node determined by the object sample; The step of inputting the first vector representation into the pre-trained career orientation prediction model to obtain a first prediction result includes: Inputting the first vector representation into the convolutional layer to obtain a feature vector that has learned N-order neighborhood information, wherein N is a positive integer greater than or equal to 1, and the feature vector includes a vector representation of a target object node that has learned a target word node in the N-order neighborhood and a vector representation of a target text category node that has learned a target word node in the N-order neighborhood; Input the feature vector into the attention layer, perform attention processing on the M target text category nodes that are most associated with the target object node, and obtain a processed feature vector, where M is a positive integer greater than or equal to 1; The processed feature vector is input into the fully connected layer, and the processed feature vector is mapped to obtain a first prediction result, wherein the prediction result includes the matching probability of the object sample corresponding to the unlabeled sample and each of the multiple occupational orientation types.

9. A method for predicting career orientation, characterized in that: The method comprises: Acquire a heterogeneous graph for the object to be tested, wherein the heterogeneous graph includes a plurality of nodes and connecting edges, the nodes are constructed based on the object to be tested, texts published by the object to be tested on the Internet, or preset text categories, the preset text categories are categories to which preset words related to career orientation in a target dictionary belong, and the edges are used to characterize the association relationship between the plurality of nodes; Vectorizing the heterogeneous graph to obtain a vector representation of the heterogeneous graph; The vector representation of the heterogeneous graph is input into a trained career orientation prediction model to obtain a career orientation prediction result for the subject to be tested, wherein the trained career orientation prediction model is trained by the career orientation prediction model training method described in any one of claims 1 to 8.

10. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 9.

Images