Method, apparatus, device, and medium for obtaining recruitment information based on a time series model

Abstract

The present invention relates to the field of artificial intelligence, and discloses a method for obtaining recruitment information based on a time series model, including: adding and classifying the time series relationship of the historical user feature set and the historical expected feature set to statistically calculate the user feature probability of the user feature type combination and the expected feature probability of the expected feature type combination; identifying the association relationship between the historical user feature set and the historical expected feature set, and expanding the initial training data set according to the association relationship, the user feature probability and the expected feature probability, and using the obtained expanded data and historical data as the training data set to train the recruitment model; outputting the target expected features matching the candidate users through the trained recruitment model. The present invention also relates to a blockchain technology, and the user feature set and the expected feature set can be stored in the blockchain node. The present invention also proposes a device, equipment and medium for obtaining recruitment information based on a time series model. The present invention can improve the recruitment efficiency and quality.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and particularly to a method, device, equipment and storage medium for obtaining recruitment information based on a time series model. Background Art

[0002] In the process of enterprise recruitment, while considering the personal characteristics of candidates, enterprises also have multiple expected indicators for candidates. Generally, it is necessary to realize intelligent recruitment based on the information of historical candidates and expected indicators. Traditional recruitment methods usually train multiple fused regression models based on the personal characteristics of candidates and the characteristics of expected indicators, and output the corresponding target expected indicators of candidates through the trained fused models to realize recruitment decisions.

[0003] However, since traditional single regression models can only calculate a single target expected indicator of a candidate, and multiple models need to be fused to obtain all the expected indicators corresponding to the candidate, the efficiency of obtaining candidate information is low, thereby reducing the efficiency of recruitment; furthermore, traditional models do not consider the influence degree of each expected indicator among multiple expected indicators on recruitment decisions, resulting in low quality of obtaining candidate information, and further reducing the quality of recruitment. Summary of the Invention

[0004] The present invention provides a method, device, equipment and storage medium for obtaining recruitment information based on a time series model, and its main purpose is to improve the efficiency and quality of recruitment.

[0005] To achieve the above object, the present invention provides a method for obtaining recruitment information based on a time series model, including:

[0006] Obtain a historical user feature set and a historical expected feature set, respectively add time series relationships to the historical user feature set and the historical expected feature set to obtain a historical user time series feature set and a historical expected time series feature set, and respectively classify the historical user time series feature set and the historical expected time series feature set to obtain user feature types and expected feature types;

[0007] Respectively count the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types;

[0008] Use the historical user feature set and the historical expected feature set as an initial training data set, identify the association relationship between the historical user feature set and the historical expected feature set, and expand the initial training data set according to the association relationship, the user feature probabilities and the expected feature probabilities to obtain a new training data set, and merge the initial training data set and the new training data set as a target training data set;

[0009] Obtain the true expected feature set corresponding to the user feature set from the target training data set, and train the preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model;

[0010] Obtain the candidate user feature set of the user to be recruited, and input the candidate user feature set into the trained recruitment model to obtain the target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

[0011] Optionally, the step of using the fusion layer in the twin tower model to interactively splice the user interaction matrix set and the institution interaction matrix set to obtain a target user interaction feature set includes:

[0012] Use the dilated causal convolutional layer in the fusion layer to splice the user interaction matrix set and the institution interaction matrix set to obtain an initial target user interaction feature set;

[0013] Use the residual connection layer in the fusion layer to perform feature dimensionality reduction on the initial user-institution interaction feature set to obtain the target user interaction feature set.

[0014] Optionally, the step of expanding the initial training data set according to the association relationship, the user feature probability, and the expected feature probability to obtain a new training data set includes:

[0015] Generate an initial new training data set according to the user feature probability and the expected feature probability;

[0016] Traverse the initial new training data set according to the association relationship to check if there are user features identical to those in the historical user feature set;

[0017] When there are no user features in the initial new training data set that are identical to those in the historical user feature set, calculate the similarity between the user feature and the historical user features in the historical user feature set, and find the expected feature corresponding to the historical user feature that matches the user feature from the historical user feature set;

[0018] Add the user feature and the expected feature corresponding to the user feature to the initial new training data set to obtain the new training data set.

[0019] Optionally, the step of separately counting the user feature probabilities of combinations between user feature types and the expected feature probabilities of combinations between expected feature types includes:

[0020] Use a preset classifier to classify users and expectations for the user feature types and the expected feature types respectively, and obtain the prior probabilities between the user feature types and the prior probabilities between the expected feature types;

[0021] Determine the user feature probabilities of the categories to which the user feature types belong according to the prior probabilities between the user feature types, and then determine the expected feature probabilities of the categories to which the expected feature types belong according to the prior probabilities between the expected feature types.

[0022] Optionally, the step of classifying the historical user time series feature set and the historical expected time series feature set respectively to obtain user feature types and expected feature types includes:

[0023] Obtain the user dimension user class labels of the historical user time series feature set, and obtain the expected class labels of the historical expected time series feature set;

[0024] Reduce the dimensions of the user dimension and the expected dimension respectively to obtain a low-dimensional historical user feature set and a low-dimensional historical expected feature set;

[0025] Discretize the low-dimensional historical user feature set according to the user class labels to obtain the user feature types;

[0026] Discretize the low-dimensional historical expected feature set according to the expected class labels to obtain the expected feature types.

[0027] Optionally, it is characterized in that the step of training a preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model includes:

[0028] Use the embedding layer in the preset recruitment model to perform positional encoding on the target training data set to obtain a training data time series vector, where the target training data set includes a user feature set and an expected feature set;

[0029] Use the encoding layer in the recruitment model to encode the training data time series vector to obtain a training data feature vector;

[0030] Use the decoding layer in the recruitment model to decode the training data feature vector to obtain a predicted expected feature set of the user feature set;

[0031] Use the loss function in the recruitment model to calculate the loss value between the predicted expected feature set and the true expected feature set, and adjust the parameters of the recruitment model according to the loss value until the loss value meets the preset conditions, and obtain the trained recruitment model.

[0032] Optionally, decoding the training data feature vector using the decoding layer in the recruitment model to obtain a predicted expected feature set of the user feature set, including:

[0033] Performing a mixed dot product operation on the training data feature vector using the multi-head attention mechanism layer in the decoding layer to obtain a training data associated feature vector;

[0034] Decoding the training data associated feature vector using the regularization layer in the decoding layer to obtain a decoded training data vector;

[0035] Inputting the decoded training data vector into an activation function in the decoding layer to output the predicted expected feature set of the user feature set.

[0036] Optionally, adding a temporal relationship to the historical user feature set and the historical expected feature set respectively to obtain a historical user temporal feature set and a historical expected temporal feature set, including:

[0037] Obtaining the collection time of each feature in the historical user feature set and the historical expected feature set, converting the collection time into a standard format to obtain a timestamp of each feature;

[0038] Adding the timestamp to the associated features with a temporal relationship between the historical user feature sets and between the historical expected feature sets respectively, and representing the associated features in order of the timestamps as the historical user temporal feature set and the historical expected temporal feature set respectively.

[0039] To solve the above problems, the present invention also provides a recruitment information acquisition device based on a temporal model, the device includes:

[0040] A user and expected feature classification module, configured to obtain a historical user feature set and a historical expected feature set, add a temporal relationship to the historical user feature set and the historical expected feature set respectively to obtain a historical user temporal feature set and a historical expected temporal feature set, and classify the historical user temporal feature set and the historical expected temporal feature set respectively to obtain a user feature type and an expected feature type;

[0041] A user and expected feature statistics module, configured to respectively count the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types;

[0042] A training data generation module, configured to use the historical user feature set and the historical expected feature set as an initial training data set, identify the association relationship between the historical user feature set and the historical expected feature set, and expand the initial training data set according to the association relationship, the user feature probability, and the expected feature probability to obtain a new training data set, and merge the initial training data set and the new training data set as a target training data set;

[0043] A recruitment model training module, configured to obtain a true expected feature set corresponding to the user feature set from the target training data set, and train a preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model;

[0044] A recruitment decision-making module, configured to obtain a candidate user feature set of a user to be recruited, input the candidate user feature set into the trained recruitment model to obtain a target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

[0045] To solve the above problems, the present invention also provides an electronic device, which includes:

[0046] A memory, storing at least one computer program; and

[0047] A processor, executing the computer program stored in the memory to implement the above-mentioned method for obtaining recruitment information based on a time series model.

[0048] To solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored, and the at least one computer program is executed by a processor in an electronic device to implement the above-mentioned method for obtaining recruitment information based on a time series model.

[0049] In the embodiments of the present invention, first, by adding the historical user feature set and the historical expected feature set with a time sequence relationship, and determining the feature time sequence of the user features and the expected features at the same time, the association between features is increased. Then, the historical user time sequence feature set and the historical expected time sequence feature set are classified, and the user feature probabilities of combinations between user feature types and the expected feature probabilities of combinations between expected feature types are statistically calculated. This can count different combinations between expected features, determine the influence degree of each expected feature on the subsequent recruitment decision, improve the quality of obtaining user information, and also facilitate improving the subsequent recruitment quality. Secondly, by identifying the association relationship between the historical user feature set and the combination of the historical expected feature set, the interaction between user features and multiple expected features is further strengthened. And according to the association relationship, the user feature probability, and the expected feature probability, the initial training data set is expanded, and the obtained expanded data and historical data are used as the training data set for training the recruitment model. While improving the accuracy of the model, the problem that multiple models need to be fused to obtain all the expected indicators corresponding to the candidates is solved, and the recruitment efficiency is improved. Finally, the target expected features matching the candidate users are output through the trained model, and the most matching candidate users can be screened out, improving the recruitment quality. Therefore, the recruitment information acquisition method, device, equipment, and storage medium based on the time sequence model proposed in the embodiments of the present invention can improve the recruitment efficiency and recruitment quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 FIG. is a schematic flowchart of a recruitment information acquisition method based on a time sequence model provided by an embodiment of the present invention;

[0051] Figure 2 FIG. is a detailed flowchart of a step in a recruitment information acquisition method based on a time sequence model provided by an embodiment of the present invention;

[0052] Figure 3 FIG. is a detailed flowchart of a step in a recruitment information acquisition method based on a time sequence model provided by an embodiment of the present invention;

[0053] Figure 4 FIG. is a module schematic diagram of a recruitment information acquisition device based on a time sequence model provided by an embodiment of the present invention;

[0054] Figure 5 FIG. is an internal structure schematic diagram of an electronic device for implementing a recruitment information acquisition method based on a time sequence model provided by an embodiment of the present invention.

[0055] The realization, functional characteristics, and advantages of the objectives of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0057] An embodiment of the present invention provides a method for obtaining recruitment information based on a time series model. The execution subject of the method for obtaining recruitment information based on a time series model includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiments of the present application. In other words, the method for obtaining recruitment information based on a time series model can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0058] Referring to Figure 1 The flowchart of the method for obtaining recruitment information based on a time series model provided by an embodiment of the present invention shown in the figure, in the embodiment of the present invention, the method for obtaining recruitment information based on a time series model includes the following steps S1-S5:

[0059] S1. Obtain a historical user feature set and a historical expectation feature set, respectively add time series relationships to the historical user feature set and the historical expectation feature set to obtain a historical user time series feature set and a historical expectation time series feature set, and respectively classify the historical user time series feature set and the historical expectation time series feature set to obtain user feature types and expectation feature types.

[0060] In the embodiment of the present invention, the historical user feature set refers to the relevant personal information of the employees already recruited by the enterprise. For example, the names, genders, ages, household registration addresses, working years, reasons for leaving, and some interview information (family situation, salary expectations) of the already recruited insurance agents, etc.; the historical expectation feature set refers to the set of expectation indicators for the employees already recruited by the enterprise. For example, the number of loan disbursements, the number of effective loan disbursements, and the loan amount reached within a certain period of time (such as the third month, half a year) after the insurance agent joins the company.

[0061] In the embodiment of the present invention, the user feature type refers to the information type of different user features. For example, if a user feature is the working years, the corresponding user feature types are 1 year, 2 years, and 3 years, etc.; the expectation feature type refers to the information type of different expectation features. For example, if an expectation feature is the number of loan disbursements of an insurance agent, the corresponding expectation feature types are 5, 10, and 20, etc.

[0062] In the embodiments of the present invention, by adding a temporal relationship to the historical user feature set and the historical expected feature set, the feature time sequence between the user features and the expected features is determined, all expected indicators corresponding to the user features are obtained, the acquisition efficiency of user information is improved, the association between features can also be increased, and the historical user temporal feature set and the historical expected temporal feature set are classified, which is convenient for subsequently finding the expected feature set that best matches the user features.

[0063] As an embodiment of the present invention, adding a temporal relationship to the historical user feature set and the historical expected feature set respectively to obtain a historical user temporal feature set and a historical expected temporal feature set includes:

[0064] Obtain the collection time of each feature in the historical user feature set and the historical expected feature set, convert the collection time into a standard format to obtain the timestamp of each feature; add the timestamp to the associated features that have a time association between the historical user feature sets and between the historical expected feature sets respectively, and represent the associated features in order of the timestamp as the historical user temporal feature set and the historical expected temporal feature set respectively.

[0065] Wherein, the timestamp is time data generated using digital signature technology, and the objects of digital signature can include information such as the historical user feature set and the historical expected feature set, signature parameters, signature time, etc., and the local time is the time between the historical user feature sets stored locally and the historical expected feature sets; the standard format can be the year, month, day and time when the data is stored locally. For example, 2022.6.30.16:00.

[0066] In an embodiment of the present invention, adding the timestamp to the features that have a time association between the historical user feature sets and between the historical expected feature sets respectively can record the storage time of the feature set in real time, and can also associate the features in order of the timestamp.

[0067] For example, the historical user feature set is x1, x2,..., xn. By adding the timestamp, it can be clearly seen that x1 occurs before x3; there is a historical expected feature set y1, y2, y3. Then, through the timestamp order, we can clearly see that y1 occurs before y2 and y2 occurs before y3. That is, there is an expected feature y1 for the number of loan disbursements, y2 for the number of effective loan disbursements, and y3 for the effective loan amount. There is a time association between these features.

[0068] Further, classifying the historical user temporal feature set and the historical expected temporal feature set respectively to obtain a user feature type and an expected feature type includes:

[0069] Obtain the user dimension and user class labels of the historical user time series feature set, and obtain the expected dimension and expected class labels of the historical expected time series feature set; respectively perform dimensionality reduction on the user dimension and the expected dimension to obtain a low-dimensional historical user feature set and a low-dimensional historical expected feature set; discretize the low-dimensional historical user feature set according to the user class labels to obtain the user feature types; discretize the low-dimensional historical expected feature set according to the expected class labels to obtain the expected feature types.

[0070] Among them, the user dimension refers to the dimension of the historical user time series feature set; the expected dimension refers to the dimension of the historical expected time series feature set. The dimensionality reduction of the user dimension and the expected dimension can be performed through the locally linear embedding algorithm, so that the distance between different classes of data in the reduced low-dimensional space has good separability, and the weight relationship between the features in the high-dimensional space can also be maintained.

[0071] In an embodiment of the present invention, the user class label refers to the category label of each feature in the historical user time series feature set. For example, if there is a feature x1 in the user time series feature set, the corresponding user class label has a types; the expected class label refers to the category label of each feature in the historical expected time series feature set. For example, if there are continuous variables y1, y2, and y3 in the expected time series feature set, the corresponding expected class labels are that y1 has d types, y2 has e types, and y3 has f types.

[0072] In the embodiment of the present invention, since both the historical user time series feature set and the historical expected time series feature set are continuous variables, and there is an association relationship between different features, the features can be discretized through the ILLE-HD3 algorithm according to the actual scenario requirements to realize the classification of the features.

[0073] For example, the historical expected time series feature set includes the number of loan disbursements, the number of effective loan disbursements, and the loan amount reached within a certain period of time after an insurance agent joins the company. Through classification, it can be obtained that the expected feature d1 class corresponding to insurance agent A is within three months of joining the company, the e1 class has 100 loan disbursements, and the f1 class has 70 effective cases.

[0074] S2. Respectively count the user feature probabilities of the combinations between the user feature types and the expected feature probabilities of the combinations between the expected feature types.

[0075] In the embodiment of the present invention, the user feature probability refers to the combination probability between user features. The expected feature probability refers to the combination probability between expected features.

[0076] In the embodiments of the present invention, by separately counting the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types, the influence degree of each expected feature on the subsequent recruitment decision can be determined, the quality of obtaining user information can be improved, and it is also convenient to improve the subsequent recruitment quality.

[0077] As an embodiment of the present invention, referring to Figure 2 shown, the separately counting the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types includes the following steps S21 - S22:

[0078] S21. Using a preset classifier to respectively perform user classification and expected classification on the user feature types and the expected feature types, so as to obtain the prior probabilities between the user feature types and the prior probabilities between the expected feature types;

[0079] S22. Determining the user feature probabilities of the categories to which the user feature types belong according to the prior probabilities between the user feature types, and then determining the expected feature probabilities of the categories to which the expected feature types belong according to the prior probabilities between the expected feature types.

[0080] Among them, the preset classifier can be a Bayesian classifier, and the prior represents the probability that an expected feature belongs to a certain expected feature category obtained according to experience.

[0081] For example, if the entry time y1 of different insurance agents has a types, the number of loan cases y2 has b types, and the number of effective loan cases y3 has c types, then the combined probability f of the expected types corresponding to each insurance agent can be counted as f = a * b * c.

[0082] S3. Using the historical user feature set and the historical expected feature set as the initial training data set, identifying the association relationship between the historical user feature set and the historical expected feature set, and expanding the initial training data set according to the association relationship, the user feature probabilities and the expected feature probabilities to obtain a new training data set, and combining the initial training data set and the new training data set as the target training data set.

[0083] In the embodiments of the present invention, the association relationship refers to the relationship between the expected features corresponding to the user features. For example, for a user, multiple features x1, x2, x3 have corresponding multiple expected features y1, y2, y3; the target training set refers to the set including the historical user feature set, the historical expected feature set, and all expanded user feature sets and expected feature sets.

[0084] In one embodiment of the present invention, the association relationship between the historical user feature set and the historical expected feature set can be identified through the Apriori algorithm.

[0085] In the embodiment of the present invention, by identifying the association relationship between the historical user feature set and the historical expected feature set, the interaction between user features and multiple expected features is further strengthened, and the data of the historical user feature set and the historical expected feature set is extended as the target training data according to the feature probability and the association relationship, that is, the associated user features and expected features are used as the data set for subsequent model training, which is convenient for improving the accuracy of the subsequent model.

[0086] As an embodiment of the present invention, referring to Figure 3 As shown, expanding the initial training data set according to the association relationship, the user feature probability, and the expected feature probability to obtain a new training data set includes the following steps S31-S34:

[0087] S31. Generate an initial new training data set according to the user feature probability and the expected feature probability;

[0088] S32. Traverse whether there are the same user features as those in the historical user feature set in the initial new training data set according to the association relationship;

[0089] S33. When there are no user features in the initial new training data set that are the same as those in the historical user feature set, calculate the similarity between the user feature and the historical user features in the historical user feature set, and find the expected features corresponding to the historical user features that match the user feature from the historical user feature set according to the similarity;

[0090] S34. Add the user feature and the expected feature corresponding to the user feature to the initial new training data set to obtain the new training data set.

[0091] Among them, since the weight of the new training data set is generated based on the user feature probability and the expected feature probability, the weight of the new training data can be maintained the same as that of the initial training data while expanding the new training data.

[0092] In one embodiment of the present invention, the similarity between the user feature and the historical user features in the historical user feature set can be calculated through the cosine similarity algorithm.

[0093] Specifically, for the new training data, by traversing whether there is a combination of x1, x2, … xn of user features in the historical user feature set in the initial new training data set, if it exists in the historical data samples, a sequence relationship with specific combinations of y1, y2, y3 can be obtained. If there is no user feature in the initial new training data set that is the same as the historical user feature set, by calculating the pairwise cosine similarity between the user features in the new training data and all historical user features, the closest x combination is found, and an association relationship is established with the corresponding historical expected feature y to achieve the expansion of the training data.

[0094] In an embodiment of the present invention, the data expansion of the historical expected feature set is the same as the above method, that is, by traversing all historical expected feature sets y1, y2,... yn, the corresponding relationship between the expected feature and the historical user feature set x is completed.

[0095] S4. Obtain the true expected feature set corresponding to the user feature set from the target training data set, and train a preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model.

[0096] In an embodiment of the present invention, the preset recruitment model may be a temporal Transformer model, which includes: an embedding layer, an encoding layer, a decoding layer, and a loss function.

[0097] In an embodiment of the present invention, by training a preset recruitment model according to the target training data set and the true expected feature set, the model can be trained through the user feature set and multiple expected features, which improves the accuracy of the model and solves the problem that multiple model fusions are required to obtain all expected indicators corresponding to candidates, thereby improving the recruitment efficiency.

[0098] As an embodiment of the present invention, the training of the preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model includes:

[0099] Using the embedding layer in the preset recruitment model to perform positional encoding on the target training data set to obtain a training data time series vector, where the target training data set includes a user feature set and an expected feature set; using the encoding layer in the recruitment model to encode the training data time series vector to obtain a training data feature vector; using the decoding layer in the recruitment model to decode the training data feature vector to obtain a predicted expected feature set of the user feature set; using the loss function in the recruitment model to calculate the loss value between the predicted expected feature set and the true expected feature set, and adjusting the parameters of the recruitment model according to the loss value until the loss value meets a preset condition, then obtaining the trained recruitment model.

[0100] Among them, performing positional encoding on the target training dataset can obtain a positional encoding vector containing data time series information, and the positional encoding can be implemented by the following formula:

[0101]

[0102]

[0103] Among them, PE(t, 2i) represents the positional encoding vector of the i-th dimension at time t in the input target training dataset, and d model represents the dimension of positional encoding of the target training dataset, t represents the time of inputting the target training dataset, and i represents the length of the input target training dataset.

[0104] In one embodiment of the present invention, the encoding layer can further extract important information of the training data time series vector, making the model more accurate during the training process; the loss function can be a cross-entropy function; adjusting the parameters of the augmentation model can be achieved by the stochastic gradient descent method.

[0105] Furthermore, decoding the training data feature vector by using the decoding layer in the augmentation model to obtain the predicted expected feature set of the user feature set includes:

[0106] Performing a dot product operation on the training data feature vector by using the multi-head attention mechanism layer in the decoding layer to obtain a training data correlation feature vector; decoding the training data correlation feature vector by using the regularization layer in the decoding layer to obtain a decoded training data vector; inputting the decoded training data vector into the activation function in the decoding layer to output the predicted expected feature set of the user feature set.

[0107] Among them, the dot product operation pays more attention to local features to ensure that important features will not be lost in subsequent operations, and due to the role of the multi-head attention mechanism, the model can also obtain attention information of the same input at different positions in different representation subspaces, and the multi-head attention mechanism can perform parallel calculations, thereby improving the performance and training speed of the model.

[0108] In one embodiment of the present invention, when connecting layer by layer, using a regularization layer to replace the general residual connection layer can improve the convergence speed of the model, and can further improve the prediction result of the model during the decoding process.

[0109] In an embodiment of the present invention, the activation function can be a Softmax activation function.

[0110] S5. Obtain the candidate user feature set of the user to be recruited, input the candidate user feature set into the trained recruitment model, obtain the target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

[0111] In the embodiment of the present invention, the candidate set of users to be recruited refers to the relevant personal information of the employees to be recruited by an enterprise, including name, gender, age, household registration address, reason for leaving the job, whether engaged in the relevant industry, etc.

[0112] In an embodiment of the present invention, by inputting the candidate user feature set into the trained recruitment model, multiple expected features that match can be obtained.

[0113] For example, there is an insurance agent to be recruited who has been engaged in the relevant industry for three years and is 28 years old. Then, through the model, it can be estimated that the target expected features obtained after the agent joins the job can be that the number of loan cases per month can reach 80, and the loan amount can also reach a relatively high business level. This agent can be recruited.

[0114] In the embodiment of the present invention, first, by adding the time series relationship between the historical user feature set and the historical expected feature set, and determining the feature time series of the user features and the expected features at the same time, the association between the features is increased, and the historical user time series feature set and the historical expected time series feature set are classified, and the user feature probabilities of the combinations between the user feature types and the expected feature probabilities of the combinations between the expected feature types are statistically calculated, which can statistically calculate different combinations between the expected features, determine the influence degree of each expected feature on the subsequent recruitment decision, improve the quality of obtaining user information, and also facilitate improving the subsequent recruitment quality; secondly, by identifying the association relationship between the historical user feature set and the combination of the historical expected feature set, the interaction between the user features and multiple expected features is further strengthened, and the initial training data set is expanded according to the association relationship, the user feature probability and the expected feature probability, and the obtained expanded data and historical data are used as the training data set for training the recruitment model, which improves the accuracy of the model while solving the problem that multiple model fusions are required to obtain all the expected indicators corresponding to the candidates, and improves the recruitment efficiency; finally, the target expected features that match the candidate users are output through the trained model, and the most matching candidate users can be screened out, improving the recruitment quality. Therefore, the recruitment information acquisition method based on the time series model proposed in the embodiment of the present invention can improve the recruitment efficiency and recruitment quality.

[0115] The recruitment information acquisition device 100 based on the time series model according to the present invention can be installed in an electronic device. According to the implemented functions, the recruitment information acquisition device based on the time series model may include a user and expected feature classification module 101, a user and expected feature statistics module 102, a training data generation module 103, a recruitment model training module 104, and a recruitment decision module 105. The modules in the present invention may also be referred to as units, which refer to a series of computer program segments that can be executed by the processor of the electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0116] In this embodiment, the functions of each module / unit are as follows:

[0117] The user and expected feature classification module 101 is used to obtain the historical user feature set and the historical expected feature set, add time series relationships to the historical user feature set and the historical expected feature set respectively to obtain the historical user time series feature set and the historical expected time series feature set, and classify the historical user time series feature set and the historical expected time series feature set respectively to obtain the user feature type and the expected feature type.

[0118] In the embodiment of the present invention, the historical user feature set refers to the relevant personal information of the recruited personnel of the enterprise. For example, the names, genders, ages, household registration addresses, working years, reasons for leaving, and some interview information (family situation, salary expectations) of the recruited insurance agents, etc.; the historical expected feature set refers to the set of expected indicators of the enterprise for the recruited personnel. For example, the number of loan disbursements, the number of effective loan disbursements, and the loan amount reached within a certain period of time (such as the third month, half a year) after the insurance agent joins the company.

[0119] In the embodiment of the present invention, the user feature type refers to the information type of different user features. For example, if a user feature is the working years, the corresponding user feature types are 1 year, 2 years, 3 years, etc.; the expected feature type refers to the information type of different expected features. For example, if an expected feature is the number of loan disbursements of an insurance agent, the corresponding expected feature types are 5 pieces, 10 pieces, 20 pieces, etc.

[0120] In the embodiment of the present invention, by adding time series relationships to the historical user feature set and the historical expected feature set, the feature time series of the user features and the expected features are determined, all expected indicators corresponding to the user features are obtained, the acquisition efficiency of user information is improved, the association between features can also be increased, and the historical user time series feature set and the historical expected time series feature set are classified to facilitate finding the most matching expected feature set for the user features subsequently.

[0121] As an embodiment of the present invention, the user and the expected feature classification module 101 respectively add a temporal relationship to the historical user feature set and the historical expected feature set by performing the following operations, obtaining a historical user temporal feature set and a historical expected temporal feature set, including:

[0122] Obtain the collection time of each feature in the historical user feature set and the historical expected feature set, convert the collection time into a standard format, and obtain the timestamp of each feature;

[0123] Add the timestamp to the associated features that have a time association between the historical user feature sets and between the historical expected feature sets respectively, and represent the associated features as the historical user temporal feature set and the historical expected temporal feature set respectively in the order of the timestamps.

[0124] Among them, the timestamp is time data generated using digital signature technology. The objects of digital signature can include information such as the historical user feature set and the historical expected feature set, signature parameters, and signature time. The local time is the time between the historical user feature sets and the historical expected feature sets stored locally; the standard format can be the year, month, day, and time when the data is stored locally. For example, 2022.6.30.16:00.

[0125] In an embodiment of the present invention, adding the timestamp to the features that have a time association between the historical user feature sets and between the historical expected feature sets respectively can record the storage time of the feature sets in real time and can also associate the features in the order of the timestamps.

[0126] For example, the historical user feature set is x1, x2,..., xn. By adding the timestamp, it can be clearly seen that x1 occurs before x3. There is a historical expected feature set y1, y2, y3. Then, through the order of the timestamps, we can clearly see that y1 occurs before y2 before y3. That is, there is an expected feature y1 for the number of loans, y2 for the number of effective loans, and y3 for the effective loan amount. There is a time association between these features.

[0127] Further, classifying the historical user temporal feature set and the historical expected temporal feature set respectively to obtain a user feature type and an expected feature type, including:

[0128] Obtain the user dimension and user class labels of the historical user time series feature set, and obtain the expectation dimension and expectation class labels of the historical expectation time series feature set; respectively perform dimensionality reduction on the user dimension and the expectation dimension to obtain a low-dimensional historical user feature set and a low-dimensional historical expectation feature set; discretize the low-dimensional historical user feature set according to the user class labels to obtain the user feature types; discretize the low-dimensional historical expectation feature set according to the expectation class labels to obtain the expectation feature types.

[0129] Wherein, the user dimension refers to the dimension of the historical user time series feature set; the expectation dimension refers to the dimension of the historical expectation time series feature set. Dimensionality reduction can be performed on the user dimension and the expectation dimension through the locally linear embedding algorithm, so that the distance between different classes of data in the reduced low-dimensional space has good separability, and the weight relationship between the features in the high-dimensional space can also be maintained.

[0130] In an embodiment of the present invention, the user class label refers to the class label of each feature in the historical user time series feature set. For example, if there is a feature x1 in the user time series feature set, the corresponding user class label has a types; the expectation class label refers to the class label of each feature in the historical expectation time series feature set. For example, if there are continuous variables y1, y2, and y3 in the expectation time series feature set, the corresponding expectation class labels are that y1 has d types, y2 has e types, and y3 has f types.

[0131] In an embodiment of the present invention, since both the historical user time series feature set and the historical expectation time series feature set are continuous variables, and there is an association relationship between different features, the features can be discretized through the ILLE-HD3 algorithm according to the actual scenario requirements to realize feature classification.

[0132] For example, the historical expectation time series feature set includes the number of loan disbursements, the number of effective loan disbursements, and the loan amount reached within a certain period of time after an insurance agent joins the company. Through classification, it can be obtained that the expectation feature d1 class corresponding to insurance agent A is within three months of joining the company, the e1 class of the number of loan disbursements reached is 100 pieces, and the f1 class of the number of effective pieces is 70 pieces.

[0133] The user and expectation feature statistics module 102 is used to respectively count the user feature probabilities of the combinations between the user feature types and the expectation feature probabilities of the combinations between the expectation feature types.

[0134] In an embodiment of the present invention, the user feature probability refers to the combination probability between user features. The expectation feature probability refers to the combination probability between expectation features.

[0135] In the embodiment of the present invention, by separately counting the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types, the influence degree of each expected feature on the subsequent recruitment decision can be determined, the quality of obtaining user information can be improved, and it is also convenient to improve the subsequent recruitment quality.

[0136] As an embodiment of the present invention, the user and expected feature statistics module 102 separately counts the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types by performing the following operations, including:

[0137] Using a preset classifier to perform user classification and expected classification on the user feature types and the expected feature types respectively, to obtain the prior probabilities between the user feature types and the prior probabilities between the expected feature types;

[0138] Determine the user feature probabilities of the categories to which the user feature types belong according to the prior probabilities between the user feature types, and then determine the expected feature probabilities of the categories to which the expected feature types belong according to the prior probabilities between the expected feature types.

[0139] Among them, the preset classifier can be a Bayesian classifier, and the prior represents the probability that an expected feature belongs to a certain expected feature category obtained according to experience.

[0140] For example, if the entry time y1 of different insurance agents has a types, the number of loan disbursements y2 has b types, and the number of effective loan disbursements y3 has c types, then the combined probability of the expected types corresponding to each insurance agent can be statistically calculated as f = a * b * c.

[0141] The training data generation module 103 is used to use the historical user feature set and the historical expected feature set as the initial training data set, identify the association relationship between the historical user feature set and the historical expected feature set, and expand the initial training data set according to the association relationship, the user feature probability, and the expected feature probability to obtain a new training data set, and merge the initial training data set and the new training data set as the target training data set.

[0142] In the embodiment of the present invention, the association relationship refers to the relationship between the expected features corresponding to the user features. For example, for a user, multiple features x1, x2, x3 have corresponding multiple expected features y1, y2, y3; the target training set refers to the set including the historical user feature set, the historical expected feature set, and all expanded user feature sets and expected feature sets.

[0143] In an embodiment of the present invention, the Apriori algorithm can be used to identify the association relationship between the historical user feature set and the historical expected feature set.

[0144] In the embodiment of the present invention, by identifying the association relationship between the historical user feature set and the historical expected feature set, the interaction between user features and multiple expected features is further strengthened, and the data of the historical user feature set and the historical expected feature set is extended according to the feature probability and the association relationship as the target training data, that is, the associated user features and expected features are used as the data set for subsequent model training, which is convenient for improving the accuracy of the subsequent model.

[0145] As an embodiment of the present invention, the training data generation module 103 expands the initial training data set according to the association relationship, the user feature probability, and the expected feature probability by performing the following operations to obtain a new training data set, including:

[0146] Generate an initial new training data set according to the user feature probability and the expected feature probability;

[0147] Traverse the initial new training data set according to the association relationship to check whether there are user features identical to those in the historical user feature set;

[0148] When there are no user features in the initial new training data set that are identical to those in the historical user feature set, calculate the similarity between the user feature and the historical user features in the historical user feature set, and find the expected features corresponding to the historical user features that match the user feature from the historical user feature set according to the similarity;

[0149] Add the user feature and the expected feature corresponding to the user feature to the initial new training data set to obtain the new training data set.

[0150] Among them, since the weight of the new training data set is generated based on the user feature probability and the expected feature probability, the weight of the new training data can be maintained the same as that of the initial training data while expanding the new training data.

[0151] In an embodiment of the present invention, the similarity between the user feature and the historical user features in the historical user feature set can be calculated by the cosine similarity algorithm.

[0152] Specifically, for the new training data, by traversing whether there is a combination of x1, x2,... xn of user features in the initial new training data set that is the same as that in the historical user feature set, if it exists in the historical data sample, a sequence relationship of a specific combination of y1, y2, y3 can be obtained. If there are no user features in the initial new training data set that are the same as those in the historical user feature set, by calculating the one-to-one cosine similarity between the user feature in the new training data and all historical user features, the closest x combination is found, and an association relationship is established with the corresponding historical expected feature y to achieve the expansion of the training data.

[0153] In one embodiment of the present invention, the data expansion of the historical expected feature set is the same as the above method, that is, all historical expected feature sets y1, y2,... yn are traversed to complete the correspondence between the expected features and the historical user feature set x.

[0154] The recruitment model training module 104 is used to obtain the true expected feature set corresponding to the user feature set from the target training data set, and train a preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model.

[0155] In an embodiment of the present invention, the preset recruitment model may be a temporal Transformer model, which includes: an embedding layer, an encoding layer, a decoding layer, and a loss function.

[0156] In the embodiment of the present invention, by training the preset recruitment model according to the target training data set and the true expected feature set, the model can be trained through the user feature set and multiple expected features, which improves the accuracy of the model and solves the problem that multiple model fusions are required to obtain all expected indicators corresponding to the candidates, thereby improving the recruitment efficiency.

[0157] As an embodiment of the present invention, the recruitment model training module 104 trains the preset recruitment model according to the target training data set and the true expected feature set by performing the following operations to obtain a trained recruitment model, including:

[0158] Using the embedding layer in the preset recruitment model to perform positional encoding on the target training data set to obtain a training data time series vector, where the target training data set includes a user feature set and an expected feature set;

[0159] Using the encoding layer in the recruitment model to encode the training data time series vector to obtain a training data feature vector;

[0160] Using the decoding layer in the recruitment model to decode the training data feature vector to obtain a predicted expected feature set of the user feature set;

[0161] Using the loss function in the recruitment model to calculate the loss value between the predicted expected feature set and the true expected feature set, and adjusting the parameters of the recruitment model according to the loss value until the loss value meets the preset conditions, then obtaining the trained recruitment model.

[0162] Among them, performing positional encoding on the target training data set can obtain a positional encoding vector containing data time series information, and the positional encoding can be implemented by the following formula:

[0163]

[0164]

[0165] Among them, PE(t, 2i) represents the position encoding vector of the i-th dimension at time t in the input target training dataset, and d model represents the dimension for position encoding of the target training dataset, t represents the time of the input target training dataset, and i represents the length of the input target training dataset.

[0166] In an embodiment of the present invention, the encoding layer can further extract important information of the training data time series vector, making the model more accurate during the training process; the loss function can be a cross-entropy function; the adjustment of the parameters of the recruitment model can be achieved by the stochastic gradient descent method.

[0167] Further, using the decoding layer in the recruitment model to decode the training data feature vector to obtain the predicted expected feature set of the user feature set includes: using the multi-head attention mechanism layer in the decoding layer to perform a dot product operation on the training data feature vector to obtain a training data associated feature vector; using the regularization layer in the decoding layer to decode the training data associated feature vector to obtain a decoded training data vector; inputting the decoded training data vector into the activation function in the decoding layer to output the predicted expected feature set of the user feature set.

[0168] Among them, the dot product operation pays more attention to local features to ensure that important features will not be lost in subsequent operations, and due to the role of the multi-head attention mechanism, the model can also obtain attention information of the same input at different positions in different representation subspaces, and this multi-head attention mechanism can be calculated in parallel, thereby improving the performance and training speed of the model.

[0169] In an embodiment of the present invention, in the connection between layers, using a regularization layer to replace the general residual connection layer can improve the convergence speed of the model and further improve the prediction result of the model during the decoding process.

[0170] In an embodiment of the present invention, the activation function can be a Softmax activation function.

[0171] The recruitment decision module 105 is used to obtain the candidate user feature set of the user to be recruited, input the candidate user feature set into the trained recruitment model to obtain the target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

[0172] In an embodiment of the present invention, the candidate user set to be recruited refers to the relevant personal information of the employees to be recruited by an enterprise, including name, gender, age, household registration address, reason for leaving the job, whether engaged in the relevant industry, etc.

[0173] In an embodiment of the present invention, by inputting the candidate user feature set into the trained recruitment model, multiple expected features that match can be obtained.

[0174] For example, there is a candidate insurance agent to be recruited who has been engaged in the relevant industry for three years and is 28 years old. Then, through the model, it can be estimated that the target expected features obtained after the agent joins the company can be that the number of loan disbursements per month can reach 80, and the loan disbursement amount can also reach a relatively high business level. This agent can be recruited.

[0175] In an embodiment of the present invention, first, by adding the historical user feature set and the historical expected feature set in a time series relationship, and at the same time determining the feature time series of the user features and the expected features to increase the association between the features, and classifying the historical user time series feature set and the historical expected time series feature set, the user feature probabilities of the combinations between the user feature types and the expected feature probabilities of the combinations between the expected feature types are statistically calculated, which can statistically calculate different combinations between the expected features, determine the influence degree of each expected feature on the subsequent recruitment decision, improve the quality of obtaining user information, and also facilitate improving the subsequent recruitment quality; secondly, by identifying the association relationship between the historical user feature set and the combination of the historical expected feature set, the interaction between the user features and multiple expected features is further strengthened, and the initial training data set is expanded according to the association relationship, the user feature probability, and the expected feature probability, and the obtained expanded data and historical data are used as the training data set for training the recruitment model, which improves the accuracy of the model while solving the problem that multiple model fusions are required to obtain all the expected indicators corresponding to the candidates, and improves the recruitment efficiency; finally, the target expected features that match the candidate users are output through the trained model, and the most matching candidate users can be screened out, improving the recruitment quality. Therefore, the recruitment information acquisition device based on the time series model proposed in the embodiment of the present invention can improve the recruitment efficiency and the recruitment quality.

[0176] As Figure 5 shown, it is a schematic structural diagram of an electronic device for implementing the recruitment information acquisition method based on the time series model of the present invention.

[0177] The electronic device may include a processor 10, a memory 11, a communication bus 12, and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a recruitment information acquisition program based on the time series model.

[0178] Among them, the memory 11 includes at least one type of medium, which includes flash memory, external hard drive, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, local disk, optical disc, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device, such as the external hard drive of the electronic device. In some other embodiments, the memory 11 can also be an external storage device of the electronic device, such as a plug-in external hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. equipped on the electronic device. Further, the memory 11 can also include both the internal storage unit and the external storage device of the electronic device. The memory 11 can be used not only to store application software installed on the electronic device and various types of data, such as the code of the employee recruitment information acquisition program based on the time series model, etc., but also to temporarily store the data that has been output or will be output.

[0179] In some embodiments, the processor 10 can be composed of integrated circuits. For example, it can be composed of a single packaged integrated circuit, or can be composed of multiple integrated circuits with the same or different functions packaged together, including the combination of one or more Central Processing Units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, connecting various components of the entire electronic device through various interfaces and lines, and by running or executing the programs or modules stored in the memory 11 (such as the employee recruitment information acquisition program based on the time series model, etc.), and calling the data stored in the memory 11, to execute various functions of the electronic device and process data.

[0180] The communication bus 12 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The communication bus 12 is set to realize the connection and communication between the memory 11 and at least one processor 10, etc. For the sake of simplicity of representation, only a thick line is shown in the figure, but it does not mean that there is only one bus or one type of bus.

[0181] Figure 5 Only the electronic device with components is shown. Those skilled in the art can understand that Figure 5The structures shown do not constitute a limitation on the electronic device, which may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0182] For example, although not shown, the electronic device may further include a power source (such as a battery) for powering each component. Preferably, the power source can be logically connected to the at least one processor 10 through a power management device, so as to implement functions such as charging management, discharging management, and power consumption management through the power management device. The power source may also include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or inverter, a power status indicator, etc. The electronic device may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0183] Optionally, the communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between this electronic device and other electronic devices.

[0184] Optionally, the communication interface 13 may further include a user interface. The user interface may be a display, an input unit (such as a keyboard), and optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the electronic device and to display a visual user interface.

[0185] It should be understood that the embodiments are only for illustration purposes and are not limited by this structure in the scope of the patent application.

[0186] The program for obtaining increment information based on a timing model stored in the memory 11 in the electronic device is a combination of multiple computer programs. When running in the processor 10, it can implement:

[0187] Obtain a historical user feature set and a historical expected feature set, respectively add a timing relationship to the historical user feature set and the historical expected feature set to obtain a historical user timing feature set and a historical expected timing feature set, and respectively classify the historical user timing feature set and the historical expected timing feature set to obtain a user feature type and an expected feature type;

[0188] Respectively count the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types;

[0189] Use the historical user feature set and the historical expected feature set as the initial training data set, identify the association relationship between the historical user feature set and the historical expected feature set, and expand the initial training data set according to the association relationship, the user feature probability, and the expected feature probability to obtain a new training data set. Combine the initial training data set and the new training data set as the target training data set;

[0190] Obtain the true expected feature set corresponding to the user feature set from the target training data set, and train a preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model;

[0191] Obtain the candidate user feature set of the user to be recruited, input the candidate user feature set into the trained recruitment model to obtain the target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

[0192] Specifically, for the specific implementation method of the above computer program by the processor 10, reference may be made to Figure 1 the description of the relevant steps in the corresponding embodiment, which will not be elaborated here.

[0193] Further, if the modules / units integrated in the electronic device are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable medium. The computer-readable medium can be non-volatile or volatile. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disc, a computer memory, a read-only memory (ROM, Read-Only Memory).

[0194] An embodiment of the present invention can also provide a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor of an electronic device, it can implement:

[0195] Obtain a historical user feature set and a historical expected feature set, respectively add a time series relationship to the historical user feature set and the historical expected feature set to obtain a historical user time series feature set and a historical expected time series feature set, and respectively classify the historical user time series feature set and the historical expected time series feature set to obtain a user feature type and an expected feature type;

[0196] Respectively count the user feature probability of combinations between the user feature types and the expected feature probability of combinations between the expected feature types;

[0197] Use the historical user feature set and the historical expected feature set as the initial training data set, identify the association relationship between the historical user feature set and the historical expected feature set, and expand the initial training data set according to the association relationship, the user feature probability, and the expected feature probability to obtain a new training data set. Combine the initial training data set and the new training data set as the target training data set;

[0198] Obtain the true expected feature set corresponding to the user feature set from the target training data set, and train a preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model;

[0199] Obtain the candidate user feature set of the user to be recruited, input the candidate user feature set into the trained recruitment model to obtain the target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

[0200] Furthermore, the computer-readable storage medium may mainly include a storage program area and a storage data area. Among them, the storage program area may store an operating system, application programs required for at least one function, etc.; the storage data area may store data created according to the use of the blockchain node, etc.

[0201] In several embodiments provided by the present invention, it should be understood that the disclosed medium, device, apparatus, and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0202] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0203] In addition, each functional module in various embodiments of the present invention may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit. The above integrated unit can be implemented in the form of hardware, or in the form of a hardware plus a software functional module.

[0204] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0205] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced by the present invention. Any reference signs in the claims should not be construed as limiting the claims involved.

[0206] The blockchain referred to in the present invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm. Blockchain, in essence, is a decentralized database, a series of data blocks generated by using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of the information (anti-counterfeiting) and generate the next block. The blockchain can include the blockchain underlying platform, the platform product service layer, and the application service layer, etc.

[0207] In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. Words such as "second" are used to denote names and do not denote any particular order.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for obtaining recruitment information based on a time series model, characterized in that, The method includes: Obtain a historical user feature set and a historical expected feature set, respectively add temporal relationships to the historical user feature set and the historical expected feature set to obtain a historical user temporal feature set and a historical expected temporal feature set, and classify the historical user temporal feature set and the historical expected temporal feature set respectively to obtain user feature types and expected feature types; Statistically calculate the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types respectively; Use the historical user feature set and the historical expected feature set as an initial training data set, identify the association relationship between the historical user feature set and the historical expected feature set, and expand the initial training data set according to the association relationship, the user feature probability and the expected feature probability to obtain a new training data set, and merge the initial training data set and the new training data set as a target training data set; Obtain the true expected feature set corresponding to the user feature set from the target training data set, use the embedding layer in the preset recruitment model to perform positional encoding on the target training data set to obtain a training data temporal vector, where the target training data set includes a user feature set and an expected feature set, use the encoding layer in the recruitment model to encode the training data temporal vector to obtain a training data feature vector, use the multi-head attention mechanism layer in the decoding layer of the recruitment model to perform a dot product operation on the training data feature vector to obtain a training data association feature vector, use the regularization layer in the decoding layer to decode the training data association feature vector to obtain a decoded training data vector, input the decoded training data vector into the activation function in the decoding layer to output the predicted expected feature set of the user feature set, use the loss function in the recruitment model to calculate the loss value between the predicted expected feature set and the true expected feature set, and adjust the parameters of the recruitment model according to the loss value until the loss value meets the preset conditions to obtain a trained recruitment model; Obtain the candidate user feature set of the user to be recruited, input the candidate user feature set into the trained recruitment model to obtain the target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

2. The method for obtaining recruitment information based on a time series model according to claim 1, characterized in that, The expanding the initial training data set according to the association relationship, the user feature probability and the expected feature probability to obtain a new training data set includes: Generate an initial new training data set according to the user feature probability and the expected feature probability; Traverse whether there are user features in the initial new training data set that are the same as those in the historical user feature set according to the association relationship; When there is no user feature in the initial new training data set that is the same as the user features in the historical user feature set, calculate the similarity between the user feature and the historical user features in the historical user feature set, and find the expected feature corresponding to the historical user feature that matches the user feature from the historical user feature set according to the similarity; Add the user feature and the expected feature corresponding to the user feature to the initial new training data set to obtain the new training data set.

3. The method for obtaining recruitment information based on a time series model according to claim 1, characterized in that, The separately counting the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types includes: Use a preset classifier to perform user classification and expected classification on the user feature types and the expected feature types respectively, to obtain the prior probabilities between the user feature types and the prior probabilities between the expected feature types; Determine the user feature probabilities of the categories to which the user feature types belong according to the prior probabilities between the user feature types, and then determine the expected feature probabilities of the categories to which the expected feature types belong according to the prior probabilities between the expected feature types.

4. The method for obtaining recruitment information based on a time series model according to claim 1, characterized in that, The separately classifying the historical user time series feature set and the historical expected time series feature set to obtain the user feature type and the expected feature type includes: Obtain the user dimension and user class label of the historical user time series feature set, and obtain the expected dimension and expected class label of the historical expected time series feature set; Separate the user dimension and the expected dimension are dimensionally reduced to obtain a low-dimensional historical user feature set and a low-dimensional historical expected feature set; Discretize the low-dimensional historical user feature set according to the user class label to obtain the user feature type; Discretize the low-dimensional historical expected feature set according to the expected class label to obtain the expected feature type.

5. The method for obtaining recruitment information based on a time series model according to any one of claims 1 to 4, characterized in that, The separately adding time series relationships to the historical user feature set and the historical expected feature set to obtain a historical user time series feature set and a historical expected time series feature set includes: Obtain the collection time of each feature in the historical user feature set and the historical expected feature set, convert the collection time to a standard format to obtain the time stamp of each feature; Add the time stamp to the associated features with time associations between the historical user feature sets and between the historical expected feature sets respectively, and represent the associated features in order of the time stamp as the historical user time series feature set and the historical expected time series feature set respectively.

6. A device for obtaining recruitment information based on a time series model, used to implement the method for obtaining recruitment information based on a time series model according to any one of claims 1 to 5, characterized in that, The device includes: A user and expected feature classification module, configured to obtain a historical user feature set and a historical expected feature set, separately add time series relationships to the historical user feature set and the historical expected feature set to obtain a historical user time series feature set and a historical expected time series feature set, and separately classify the historical user time series feature set and the historical expected time series feature set to obtain a user feature type and an expected feature type; A user and expected feature statistics module, configured to separately count the user feature probabilities of combinations between the user feature types and the expected feature probabilities of combinations between the expected feature types; A training data generation module, configured to use the historical user feature set and the historical expected feature set as an initial training data set, identify the association relationship between the historical user feature set and the historical expected feature set, and expand the initial training data set according to the association relationship, the user feature probability, and the expected feature probability to obtain a new training data set, and merge the initial training data set and the new training data set as a target training data set; A recruitment model training module, configured to obtain a true expected feature set corresponding to the user feature set from the target training data set, and train a preset recruitment model according to the target training data set and the true expected feature set to obtain a trained recruitment model; A recruitment decision module, configured to obtain a candidate user feature set of the user to be recruited, input the candidate user feature set into the trained recruitment model to obtain a target expected feature set corresponding to the candidate user feature set, and make a recruitment decision on the user to be recruited according to the target expected feature set to obtain a recruitment result.

7. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the recruitment information acquisition method based on a time series model according to any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the recruitment information acquisition method based on a time series model according to any one of claims 1 to 5.

Images