Method and device for predicting personnel stability based on a prediction model

Abstract

Embodiments of the present disclosure disclose a method and apparatus for predicting personnel stability based on a prediction model. The method includes: determining first feature data with a feature classification of static continuous type, second feature data with a feature classification of static discrete type, and third feature data with a feature classification of dynamic type according to the personnel information of the personnel to be evaluated; respectively performing data processing on at least one of the determined feature data in a data processing manner matching its feature classification; respectively generating a feature vector corresponding to the data processing result thereof in a feature vector generation manner matching its feature classification for at least one of the determined feature data; and predicting the stability of the personnel to be evaluated according to the generated feature vectors. Embodiments of the present disclosure can achieve the identification of personnel stability. The entire process requires little manual participation, thus having a low cost and high accuracy, and can better meet the actual needs.

Description

Technical Field

[0001] The present disclosure relates to the field of information processing technologies, and in particular, to a method and apparatus for predicting personnel stability based on a prediction model. Background Art

[0002] Personnel instability is a problem that every enterprise faces. An important part of enterprise human resource management is to identify the risk of personnel instability and manage personnel according to the identification results.

[0003] Existing identification schemes are mainly implemented manually, for example, based on evaluations between direct superiors and colleagues. Such a scheme has a high cost and low accuracy, and it is difficult to meet actual needs. Summary of the Invention

[0004] To solve the above technical problems, the present disclosure is proposed. Embodiments of the present disclosure provide a method and apparatus for predicting personnel stability based on a prediction model.

[0005] According to one aspect of the embodiments of the present disclosure, a method for predicting personnel stability based on a prediction model is provided, including:

[0006] Determine first feature data with a feature classification of static continuous type, second feature data with a feature classification of static discrete type, and third feature data with a feature classification of dynamic type according to the personnel information of the personnel to be evaluated;

[0007] Perform data processing on at least one of the first feature data, the second feature data, and the third feature data respectively in a data processing manner matching its feature classification;

[0008] Generate a feature vector corresponding to the data processing result of at least one of the first feature data, the second feature data, and the third feature data respectively in a feature vector generation manner matching its feature classification;

[0009] Perform stability prediction on the personnel to be evaluated according to the generated feature vectors.

[0010] In an optional example, the feature vector corresponding to the data processing result of the first feature data is a first feature vector, the feature vector corresponding to the data processing result of the second feature data is a second feature vector, and the feature vector corresponding to the data processing result of the third feature data is a third feature vector;

[0011] The first feature vector is obtained by vector transformation of the data processing result of the first feature data by the first sub-model in the personnel evaluation model. The second feature vector is obtained by dense vector transformation and feature vector extraction of the data processing result of the second feature data by the second sub-model in the personnel evaluation model. The third feature vector is obtained by sequence shaping, data normalization, and time series feature extraction of the data processing result of the third feature data by the third sub-model in the personnel evaluation model.

[0012] In an optional example, the personnel evaluation model further includes a hidden layer and a fully connected layer. The hidden layer is respectively connected to the first sub-model, the second sub-model, and the third sub-model. The fully connected layer is connected to the hidden layer. The hidden layer is used to splice the first feature vector, the second feature vector, and the third feature vector to obtain a spliced vector. The fully connected layer is used to generate a stability prediction result of the person to be evaluated according to the spliced vector.

[0013] In an optional example, for at least one of the first feature data, the second feature data, and the third feature data, data processing is respectively performed on it in a data processing manner matching its feature classification, including:

[0014] When there is data corresponding to the first personnel feature in the first feature data and the data corresponding to the first personnel feature satisfies the preset numerical anomaly truncation condition, determine the normalization processing result of the truncated feature value corresponding to the first personnel feature, and update the data corresponding to the first personnel feature to the normalization processing result;

[0015] When there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature satisfies the preset sparse category value filtering condition, update the data corresponding to the second personnel feature to the first preset category value;

[0016] When there is data corresponding to the third personnel feature in the third feature data, which is composed of multiple feature values arranged in chronological order, use a preset time decay factor to perform time decay processing on the multiple feature values respectively to obtain multiple decay values corresponding to the multiple feature values, and update the data corresponding to the third personnel feature to the multiple decay values.

[0017] In an optional example,

[0018] When performing stability prediction on the person to be evaluated based on the personnel evaluation model, the method further includes:

[0019] Obtain the personal information of each of the multiple reference persons used for training the person evaluation model;

[0020] For each of the multiple reference persons, select the category value that matches it from the preset set of category values composed of N category values corresponding to the second personal feature according to its personal information;

[0021] In the case where the ratio of the total selection times of M category values among the N category values to the total selection times of the N category values is greater than the preset ratio, determine each of the remaining N - M category values among the N category values as the sparse category value corresponding to the second personal feature;

[0022] The case where there is data corresponding to the second personal feature in the second feature data, and when the data corresponding to the second personal feature meets the preset sparse category value filtering condition, updating the data corresponding to the second personal feature to the first preset category value includes:

[0023] In the case where there is data corresponding to the second personal feature in the second feature data, and when the data corresponding to the second personal feature is any sparse category value corresponding to the second personal feature, updating the data corresponding to the second personal feature to the first preset category value.

[0024] In an optional example, the performing data processing on at least one of the first feature data, the second feature data, and the third feature data respectively in a data processing manner that matches its feature classification includes:

[0025] In the case where the data corresponding to the fourth personal feature is missing in the first feature data, determine the minimum feature value corresponding to the fourth personal feature, and add the minimum feature value as the data corresponding to the fourth personal feature to the first feature data;

[0026] In the case where the data corresponding to the fifth personal feature is missing in the second feature data, add the second preset category value as the data corresponding to the fifth personal feature to the second feature data;

[0027] In the case where the data corresponding to the sixth personal feature is missing in the third feature data, add the preset feature value as the data corresponding to the sixth personal feature to the third feature data.

[0028] In an optional example,

[0029] The performing stability prediction on the person to be evaluated according to the generated feature vector includes:

[0030] Determine the stable probability and unstable probability of the person to be evaluated according to the eigenvectors corresponding to the data processing results of the first characteristic data, the second characteristic data, and the third characteristic data respectively;

[0031] According to the ratio of the unstable probability to the stable probability, perform a logarithmic conversion process on the unstable probability to obtain an unstable score corresponding to the unstable probability;

[0032] The method further includes:

[0033] In the case where the unstable score is within a preset score range, manage the person to be evaluated according to the unstable score;

[0034] In the case where the unstable score is outside the preset score range, determine the score extreme value in the preset score range that is closest to the unstable score, and manage the person to be evaluated according to the determined score extreme value.

[0035] In an alternative example, the formula used to perform a logarithmic conversion process on the unstable probability according to the ratio of the unstable probability to the stable probability to obtain an unstable score corresponding to the unstable probability is:

[0036] Score’=Score+Bln(P)-Bln(P’)

[0037] Where Score’ is the unstable score, Score is the preset score, B is the preset coefficient, P is the preset ratio, and P’ is the ratio of the unstable probability to the stable probability.

[0038] According to another aspect of the embodiments of the present disclosure, there is provided a personnel stability prediction device based on a prediction model, including:

[0039] A first determination module, configured to determine first characteristic data with a characteristic classification of a static continuous type, second characteristic data with a characteristic classification of a static discrete type, and third characteristic data with a characteristic classification of a dynamic type according to the personnel information of the person to be evaluated;

[0040] A processing module, configured to perform data processing on at least one of the first characteristic data, the second characteristic data, and the third characteristic data respectively in a data processing manner matching its characteristic classification;

[0041] A generation module, configured to generate eigenvectors corresponding to the data processing results of at least one of the first characteristic data, the second characteristic data, and the third characteristic data respectively in an eigenvector generation manner matching its characteristic classification;

[0042] A prediction module for predicting the stability of the person to be evaluated based on the generated feature vectors.

[0043] In an optional example, the feature vector corresponding to the data processing result of the first feature data is the first feature vector, the feature vector corresponding to the data processing result of the second feature data is the second feature vector, and the feature vector corresponding to the data processing result of the third feature data is the third feature vector;

[0044] The first feature vector is obtained by vector transformation of the data processing result of the first feature data by the first sub-model in the personnel evaluation model. The second feature vector is obtained by dense vector transformation and feature vector extraction of the data processing result of the second feature data by the second sub-model in the personnel evaluation model. The third feature vector is obtained by sequence shaping, data normalization, and time series feature extraction of the data processing result of the third feature data by the third sub-model in the personnel evaluation model.

[0045] In an optional example, the personnel evaluation model further includes a hidden layer and a fully connected layer. The hidden layer is respectively connected to the first sub-model, the second sub-model, and the third sub-model. The fully connected layer is connected to the hidden layer. The hidden layer is used to splice the first feature vector, the second feature vector, and the third feature vector to obtain a spliced vector. The fully connected layer is used to generate the stability prediction result of the person to be evaluated according to the spliced vector.

[0046] In an optional example, the processing module includes:

[0047] A first processing sub-module for determining the normalized processing result of the truncated feature value corresponding to the first personnel feature and updating the data corresponding to the first personnel feature to the normalized processing result when there is data corresponding to the first personnel feature in the first feature data and the data corresponding to the first personnel feature meets the preset numerical anomaly truncation condition;

[0048] A second processing sub-module for updating the data corresponding to the second personnel feature to the first preset category value when there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature meets the preset sparse category value filtering condition;

[0049] A third processing sub-module, configured to, when there is data corresponding to a third personnel feature in the third feature data, the data being composed of a plurality of feature values arranged in chronological order, perform time decay processing on the plurality of feature values respectively by using a preset time decay factor to obtain a plurality of decay values corresponding to the plurality of feature values, and update the data corresponding to the third personnel feature to the plurality of decay values.

[0050] In an optional example,

[0051] The apparatus further includes:

[0052] An acquisition module, configured to, when performing stability prediction on the to-be-evaluated personnel based on a personnel evaluation model, acquire the personnel information of each of a plurality of reference personnel used for training the personnel evaluation model;

[0053] A selection module, configured to, for each of the plurality of reference personnel, select a category value that matches it from a preset category value set corresponding to the second personnel feature and composed of N category values according to its personnel information;

[0054] A second determination module, configured to, when the ratio of the total selection times of M category values among the N category values to the total selection times of the N category values is greater than a preset ratio, determine each of the remaining N - M category values among the N category values as the sparse category values corresponding to the second personnel feature;

[0055] The second processing sub-module is specifically configured to:

[0056] When there is data corresponding to the second personnel feature in the second feature data, and the data corresponding to the second personnel feature is any sparse category value corresponding to the second personnel feature, update the data corresponding to the second personnel feature to a first preset category value.

[0057] In an optional example, the processing module includes:

[0058] A fourth processing sub-module, configured to, when data corresponding to a fourth personnel feature is missing in the first feature data, determine the minimum feature value corresponding to the fourth personnel feature, and add the minimum feature value as the data corresponding to the fourth personnel feature to the first feature data;

[0059] A fifth processing sub-module, configured to, when data corresponding to a fifth personnel feature is missing in the second feature data, add a second preset category value as the data corresponding to the fifth personnel feature to the second feature data;

[0060] The sixth processing sub-module is configured to add the data corresponding to the sixth personnel feature as the preset feature value to the third feature data when the data corresponding to the sixth personnel feature is missing in the third feature data.

[0061] In an alternative example,

[0062] The prediction module includes:

[0063] A determination sub-module, configured to determine the stable probability and the unstable probability of the to-be-evaluated personnel according to the feature vectors corresponding to the data processing results of the first feature data, the second feature data, and the third feature data respectively;

[0064] An acquisition sub-module, configured to perform a logarithmic conversion process on the unstable probability according to the ratio of the unstable probability to the stable probability to obtain an unstable score corresponding to the unstable probability;

[0065] The apparatus further includes:

[0066] A first management module, configured to manage the to-be-evaluated personnel according to the unstable score when the unstable score is within a preset score range;

[0067] A second management module, configured to determine the score extreme value in the preset score range that is closest to the unstable score when the unstable score is outside the preset score range, and manage the to-be-evaluated personnel according to the determined score extreme value.

[0068] In an alternative example, the formula used by the acquisition sub-module to perform a logarithmic conversion process on the unstable probability according to the ratio of the unstable probability to the stable probability to obtain an unstable score corresponding to the unstable probability is:

[0069] Score’ = Score + Bln(P) - Bln(P’)

[0070] where Score’ is the unstable score, Score is a preset score, B is a preset coefficient, P is a preset ratio, and P’ is the ratio of the unstable probability to the stable probability.

[0071] According to yet another aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium storing a computer program for executing the above-mentioned method for predicting personnel stability based on a prediction model.

[0072] According to still another aspect of the embodiments of the present disclosure, there is provided an electronic device, including:

[0073] A processor;

[0074] A memory for storing the processor-executable instructions;

[0075] The processor is configured to read the executable instructions from the memory and execute the instructions to implement the above-mentioned personnel stability prediction method based on a prediction model.

[0076] In an embodiment of the present disclosure, for any enterprise employee whose stability needs to be evaluated, they can be regarded as the person to be evaluated, and the personnel information of the person to be evaluated is obtained. Then, according to the personnel information of the person to be evaluated, the first feature data with the feature classification of static continuous type, the second feature data with the feature classification of static discrete type, and the third feature data with the feature classification of dynamic type are determined. The first feature data, the second feature data, and the third feature data can effectively represent the static continuous features, static discrete features, and dynamic features of the person to be evaluated. Next, for at least one of the first feature data, the second feature data, and the third feature data, data processing is respectively performed on it in a data processing manner matching its feature classification, which is beneficial to ensuring the data processing effect of the first feature data, the second feature data, and the third feature data. After that, for at least one of the first feature data, the second feature data, and the third feature data, a feature vector generation method matching its feature classification is respectively used to generate the feature vector corresponding to the data processing result of it, which is beneficial to ensuring the reliability of each generated feature vector. Then, according to the generated feature vectors, the stability of the person to be evaluated is predicted, which can better ensure the prediction accuracy. It can be seen that in the embodiment of the present disclosure, through the determination of feature data with different feature classifications, the data processing of each feature data, and the feature vector generation processing, the stability identification of personnel can be realized. The whole process requires little manual participation, has low cost and high accuracy, and can better meet the actual needs.

[0077] Next, through the drawings and embodiments, the technical solutions of the present disclosure will be further described in detail. Description of the Drawings

[0078] By describing the embodiments of the present disclosure in more detail in conjunction with the drawings, the above-mentioned and other objects, features, and advantages of the present disclosure will become more obvious. The drawings are used to provide a further understanding of the embodiments of the present disclosure, and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the present disclosure, and do not constitute a limitation to the present disclosure. In the drawings, the same reference numerals generally represent the same components or steps.

[0079] Figure 1 It is a flowchart of a personnel stability prediction method based on a prediction model provided by an exemplary embodiment of the present disclosure.

[0080] Figure 2 It is a schematic diagram of the normal distribution of the instability tendency in an exemplary embodiment of the present disclosure.

[0081] Figure 3 It is a schematic flowchart of a personnel stability prediction method based on a prediction model provided in another exemplary embodiment of the present disclosure.

[0082] Figure 4 It is a schematic structural diagram of a personnel stability prediction device based on a prediction model provided in an exemplary embodiment of the present disclosure.

[0083] Figure 5 It is a structural diagram of an electronic device provided in an exemplary embodiment of the present disclosure. Detailed implementation manners

[0084] Next, exemplary embodiments according to the present disclosure will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments of the present disclosure. It should be understood that the present disclosure is not limited by the exemplary embodiments described herein.

[0085] It should be noted that: Unless otherwise specifically stated, the relative arrangements, numerical expressions, and numerical values of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0086] Those skilled in the art can understand that terms such as "first" and "second" in the embodiments of the present disclosure are only used to distinguish different steps, devices, or modules, etc., and neither represent any specific technical meaning nor indicate an inevitable logical order between them.

[0087] It should also be understood that in the embodiments of the present disclosure, "a plurality of" may refer to two or more, and "at least one" may refer to one, two, or more.

[0088] It should also be understood that for any component, data, or structure mentioned in the embodiments of the present disclosure, without clear limitation or contrary indication in the context, it can generally be understood as one or more.

[0089] In addition, the term "and / or" in the present disclosure is only a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present disclosure generally represents an "or" relationship between the associated objects before and after.

[0090] It should also be understood that the present disclosure emphasizes the differences between the various embodiments. Their similarities or similarities can be referred to each other. For the sake of brevity, they will not be elaborated one by one.

[0091] Meanwhile, it should be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationship.

[0092] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way a limitation on the present disclosure, its application, or use.

[0093] Techniques, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods, and devices should be considered as part of the specification.

[0094] It should be noted that like reference numerals and letters denote like items in the following drawings, and thus, once an item is defined in one drawing, further discussion thereof is not required in subsequent drawings.

[0095] Embodiments of the present disclosure can be applied to electronic devices such as terminal devices, computer systems, servers, etc., which can operate together with numerous other general-purpose or special-purpose computing system environments or configurations. Examples of well-known terminal devices, computing systems, environments, and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, servers, etc. include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above systems, and so on.

[0096] Electronic devices such as terminal devices, computer systems, servers, etc. can be described in the general context of computer system-executable instructions (such as program modules) executed by a computer system. Generally, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. The computer system / server can be implemented in a distributed cloud computing environment where tasks are executed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can be located on local or remote computing system storage media including storage devices.

[0097] Exemplary method

[0098] Figure 1 It is a flowchart showing the method for predicting personnel stability based on a prediction model provided by an exemplary embodiment of the present disclosure. Figure 1 The method shown may include step 101, step 102, step 103, and step 104, and each step will be described separately below.

[0099] Step 101: Determine the first feature data with a feature classification of static continuous type, the second feature data with a feature classification of static discrete type, and the third feature data with a feature classification of dynamic type according to the personnel information of the person to be evaluated.

[0100] Here, the person to be evaluated can be any enterprise employee whose stability needs to be evaluated, such as any real estate company employee whose stability needs to be evaluated, any insurance company employee whose stability needs to be evaluated, etc.; among them, the real estate company employee can be a real estate broker, the insurance company employee can be an insurance broker, and the stability of any enterprise employee can be characterized by the strength or likelihood of the enterprise employee's willingness to continue working in the current enterprise. For the sake of easy understanding, the case where the person to be evaluated is a real estate broker will be taken as an example for illustration in the following text.

[0101] In step 101, the personnel information of the real estate broker who is the person to be evaluated can be obtained from the back-end database of the real estate company. The personnel information of the real estate broker includes, but is not limited to, the broker's personal basic information, the broker's operation behavior information, and the performance information of the store, team, and city where the broker is located; among them, the broker's personal basic information includes, but is not limited to, the city information, credit score information, recruitment source information, job level information, business opportunity conversion ability information, etc.; the broker's operation behavior information includes, but is not limited to, the number of house viewings, the number of house listings, the number of on-site inspections of houses, the number of business opportunities, etc.; the performance information of the store, team, and city where the broker is located includes, but is not limited to, the total number of house sales in the current month, the number of house sales per day in the recent period, etc.

[0102] After obtaining the personnel information of the real estate broker who is the person to be evaluated, the first feature data with a feature classification of static continuous type, the second feature data with a feature classification of static discrete type, and the third feature data with a feature classification of dynamic type can be determined based on the obtained personnel information.

[0103] It should be noted that generally speaking, the feature classifications of the first feature data and the second feature data both belong to the static type. Therefore, the data in the first feature data and the data in the second feature data will not change with time in the short term. However, individually speaking, the feature classification of the first feature data is specifically the static continuous type, which means that there can theoretically be infinitely many possibilities for the data in the first feature data, while the feature classification of the second feature data is specifically the static discrete type, which means that there are only a finite number of possibilities for the data in the second feature data. In addition, the feature classification of the third feature data is the dynamic type. Therefore, the data in the third feature data is likely to change with time. For example, the data in the third feature data may change every day, and the third feature data can also be called time series data.

[0104] Optionally, the first feature data includes but is not limited to credit score information, business opportunity conversion ability information, etc.; the second feature data includes but is not limited to city information, constellation information, recruitment source information, etc.; the third feature data includes but is not limited to the number of property viewings, the number of property listings entered, the number of property site inspections, etc.

[0105] Step 102, for at least one of the first feature data, the second feature data, and the third feature data, perform data processing on it respectively with a data processing method that matches its feature classification.

[0106] Here, the corresponding relationship between the feature classification and the data processing method can be preset, and the data processing method corresponding to any feature classification is the data processing method that matches this feature classification.

[0107] In this way, according to the preset corresponding relationship, it is possible to conveniently and reliably determine the first data processing method that matches the static continuous type, the second data processing method that matches the static discrete type, and the third data processing method that matches the dynamic type. Next, the first feature data can be processed with the first data processing method to obtain the data processing result of the first feature data; and / or, the second feature data can be processed with the second data processing method to obtain the data processing result of the second feature data; and / or, the third feature data can be processed with the third data processing method to obtain the data processing result of the third feature data.

[0108] Step 103, for at least one of the first feature data, the second feature data, and the third feature data, generate a feature vector corresponding to its data processing result respectively with a feature vector generation method that matches its feature classification.

[0109] Here, the corresponding relationship between the feature classification and the feature vector generation method can be preset, and the feature vector generation method corresponding to any feature classification is the feature vector generation method that matches this feature classification.

[0110] Here, according to the preset corresponding relationship, it is possible to conveniently and reliably determine the first feature vector generation method corresponding to the static continuous type, the second feature vector generation method corresponding to the static discrete type, and the third feature vector generation method corresponding to the dynamic type. Next, the feature vector corresponding to the data processing result of the first feature data can be generated with the first feature vector generation method; and / or, the feature vector corresponding to the data processing result of the second feature data can be generated with the second feature vector generation method; and / or, the feature vector corresponding to the data processing result of the third feature data can be generated with the third feature vector generation method.

[0111] Step 104: Perform stability prediction on the personnel to be evaluated according to the generated feature vectors.

[0112] Here, by performing stability prediction on the personnel to be evaluated according to the generated feature vectors, the stable probability (i.e., the probability of continuing to work in the current enterprise) and / or the unstable probability (i.e., the probability of not continuing to work in the current enterprise) of the personnel to be evaluated can be determined. Subsequently, based on the stability prediction results of the personnel to be evaluated, some intervention measures can be taken. For example, when the stable probability of the personnel to be evaluated is less than the set stable probability, their needs can be communicated with them or more resources can be obtained for them.

[0113] In the embodiments of the present disclosure, for any enterprise employee whose stability needs to be evaluated, they can be regarded as the personnel to be evaluated, and their personnel information can be obtained. Then, according to the personnel information of the personnel to be evaluated, the first feature data with the feature classification of static continuous type, the second feature data with the feature classification of static discrete type, and the third feature data with the feature classification of dynamic type can be determined. The first feature data, the second feature data, and the third feature data can effectively represent the static continuous features, static discrete features, and dynamic features of the personnel to be evaluated. Next, for at least one of the first feature data, the second feature data, and the third feature data, data processing can be performed on it respectively in a data processing manner matching its feature classification, which is conducive to ensuring the data processing effect of the first feature data, the second feature data, and the third feature data. Subsequently, for at least one of the first feature data, the second feature data, and the third feature data, a feature vector generation method matching its feature classification can be used to generate the feature vector corresponding to its data processing result, which is conducive to ensuring the reliability of each generated feature vector. Then, performing stability prediction on the personnel to be evaluated according to the generated feature vectors can better ensure the prediction accuracy. It can be seen that in the embodiments of the present disclosure, through the determination of feature data with different feature classifications, the data processing of each feature data, and the feature vector generation process, the stability identification of personnel can be realized. The whole process requires little manual participation, has low costs, high accuracy, and can better meet the actual needs.

[0114] In an optional example, the feature vector corresponding to the data processing result of the first feature data is the first feature vector, the feature vector corresponding to the data processing result of the second feature data is the second feature vector, and the feature vector corresponding to the data processing result of the third feature data is the third feature vector;

[0115] The first feature vector is obtained by performing vector transformation on the data processing result of the first feature data by the first sub-model in the personnel evaluation model. The second feature vector is obtained by performing dense vector transformation and feature vector extraction on the data processing result of the second feature data by the second sub-model in the personnel evaluation model. The third feature vector is obtained by performing sequence shaping, data normalization, and time series feature extraction on the data processing result of the third feature data by the third sub-model in the personnel evaluation model.

[0116] Optionally, the personnel evaluation model further includes a Hidden Layer and a Full Connection Layer. The Hidden Layer is respectively connected to the first sub-model, the second sub-model, and the third sub-model. The Full Connection Layer is connected to the Hidden Layer. The Hidden Layer is used to splice the first feature vector, the second feature vector, and the third feature vector to obtain a spliced vector. The Full Connection Layer is used to generate a stability prediction result of the person to be evaluated according to the spliced vector.

[0117] In an embodiment of the present disclosure, a personnel evaluation model belonging to an end-to-end model can be pre-trained through a machine learning algorithm. The personnel evaluation model can be specifically trained based on the personnel information of multiple reference personnel. Among them, the personnel evaluation model can include a first sub-model, a second sub-model, a third sub-model, a Hidden Layer respectively connected to the first sub-model, the second sub-model, and the third sub-model, and a Full Connection Layer connected to the Hidden Layer.

[0118] The data processing result of the first feature data can be provided to the first sub-model. The first sub-model can perform vector transformation on the data processing result of the first feature data. For example, directly convert the data processing result of the first feature data into Dense Features (i.e., dense features). The converted Dense Features can be used as the first feature vector. The first sub-model can also output the first feature vector to the Hidden Layer.

[0119] The data processing result of the second feature data can be provided to the second sub-model. The second sub-model can perform dense vector transformation and feature vector extraction on the data processing result of the second feature data. For example, first convert the data processing result of the second feature data into a Dense vector (dense vector) through an Embedding layer (which is a way to convert discrete variables into continuous variables), and then extract the feature vector through a Gated Recurrent Unit (GRU) neural network structure to obtain the second feature vector. The second sub-model can also output the second feature vector to the Hidden Layer.

[0120] The data processing result of the third feature data can be provided to the third sub-model. The third sub-model can perform sequence shaping, data normalization, and time series feature extraction on the data processing result of the third feature data. For example, sequence shaping can be performed first, and then the data can be normalized through batch normalization. After that, time series features can be extracted through a long short-term memory (LSTM) network structure to obtain a third feature vector. The third sub-model can also output the third feature vector to the hidden layer.

[0121] After obtaining the first feature vector, the second feature vector, and the third feature vector, the hidden layer can concatenate the first feature vector, the second feature vector, and the third feature vector to form a complete feature vector, which can be used as the concatenated vector. The hidden layer can output the concatenated vector to the fully connected layer.

[0122] There can be several layers in the fully connected layer. The fully connected layer is used to process the concatenated vector, with ReLU as the activation function. Then, binary classification is performed through the Sigmoid function to obtain the probabilities of the person to be evaluated being stable and unstable (i.e., the stable probability and the unstable probability mentioned above), and thus the stability prediction result of the person to be evaluated can be obtained.

[0123] In the embodiments of the present disclosure, for the data processing results of the first feature data, the second feature data, and the third feature data respectively, the corresponding sub-models in the personnel evaluation model trained through machine learning algorithms can be used to generate the corresponding feature vectors in a suitable manner. In addition, by using the Hidden Layer and the fully connected layer in the personnel evaluation model, the generation of the stability prediction result can be conveniently and reliably achieved, and the objectivity of the stability prediction result can be better guaranteed. In this way, the problem of strong subjectivity in personnel stability identification based on manual implementation can be better avoided, thus better ensuring the accuracy and reliability of the identification result.

[0124] It should be noted that there are various specific implementation manners for step 102, which are introduced by way of example below.

[0125] In a specific implementation manner, step 102 includes:

[0126] When there is data corresponding to the first personnel feature in the first feature data and the data corresponding to the first personnel feature meets the preset numerical anomaly truncation condition, determine the normalization processing result of the truncated feature value corresponding to the first personnel feature, and update the data corresponding to the first personnel feature to the normalization processing result;

[0127] When there is data corresponding to the second personal feature in the second feature data, and the data corresponding to the second personal feature meets the preset sparse category value filtering condition, update the data corresponding to the second personal feature to the first preset category value;

[0128] When there is data corresponding to the third personal feature in the third feature data, and the data consists of multiple feature values arranged in chronological order, use the preset time decay factor to perform time decay processing on the multiple feature values respectively to obtain multiple decay values corresponding to the multiple feature values, and update the data corresponding to the third personal feature to the multiple decay values.

[0129] When there is data corresponding to the first personal feature in the first feature data, it can be determined whether the data corresponding to the first personal feature meets the abnormal truncation condition. When the determination result is that it meets, the truncation feature value corresponding to the first personal feature can be obtained, and the truncation feature value corresponding to the first personal feature is normalized to obtain the normalization result, and the data corresponding to the first personal feature is updated to the normalization result.

[0130] It should be noted that there are two mainstream normalization methods currently, namely MinMaxScaler and StandardScaler; among them, MinMaxScaler scales the data to the range of 0 - 1 according to the maximum and minimum values of the feature; StandardScaler scales the data to an interval with a mean of 0 according to the mean and variance of the feature. Since StandardScaler has a stronger tolerance for outliers in the data, StandardScaler can be used in the embodiments of the present disclosure to perform the normalization of the truncation feature values.

[0131] In a specific example, if the first personal feature is the credit score feature, then the data corresponding to the first personal feature is specifically the credit score information, and the credit score information can specifically be the credit score value.

[0132] In order to implement the judgment related to the preset numerical abnormal truncation condition, the truncation feature value corresponding to the credit score feature can be preset. Assume that the personnel evaluation model is specifically trained based on the personnel information of 10,000 reference personnel (each reference personnel is a real estate agent), then based on the personnel information of these 10,000 reference personnel respectively, 10,000 credit score values corresponding to these 10,000 reference personnel can be obtained, and these 10,000 credit score values are arranged in ascending order to obtain a credit score value sequence. Next, the credit score value ranked at a specified position (such as the 95% position, 98% position, etc.) in the credit score value sequence can be selected, and the selected credit score value is used as the truncation feature value corresponding to the credit score feature.

[0133] When there is a credit score value that is the data corresponding to the credit score feature in the first feature data, the credit score value can be compared with the truncated feature value corresponding to the credit score feature. When the credit score value is less than or equal to the truncated feature value corresponding to the credit score feature, it can be determined that the credit score value does not meet the preset numerical anomaly truncation condition; otherwise, it can be determined that the credit score value meets the preset numerical anomaly truncation condition. When the credit score value meets the preset numerical anomaly truncation condition, the truncated feature value corresponding to the credit score feature can be normalized to obtain a normalization result, and the credit score value can be updated to the normalization result.

[0134] When there is data corresponding to the second personnel feature in the second feature data, it can be determined whether the data corresponding to the second personnel feature meets the preset sparse category value filtering condition. When the determination result is satisfied, the data corresponding to the second personnel feature can be updated to the first preset category value.

[0135] Optionally, when performing stability prediction on the to-be-evaluated person based on the personnel evaluation model, the method further includes:

[0136] Obtaining the personnel information of each of the multiple reference persons used to train the personnel evaluation model;

[0137] For each of the multiple reference persons, according to their personnel information, select the category value that matches them from the preset category value set composed of N category values corresponding to the second personnel feature;

[0138] When the ratio of the total selection times of M category values among the N category values to the total selection times of the N category values is greater than the preset ratio, each of the remaining N - M category values among the N category values is determined as the sparse category value corresponding to the second personnel feature;

[0139] When there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature meets the preset sparse category value filtering condition, updating the data corresponding to the second personnel feature to the first preset category value includes:

[0140] When there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature is any sparse category value corresponding to the second personnel feature, the data corresponding to the second personnel feature is updated to the first preset category value.

[0141] In a specific example, the second personnel feature is the recruitment source feature, then the data corresponding to the second personnel feature is specifically the recruitment source information, and the recruitment source information can specifically be a recruitment source option (which can be in the form of a field).

[0142] After obtaining the personal information of each of the multiple reference persons used to train the personnel evaluation model, for example, after obtaining the personal information of 10,000 reference persons, for each reference person, according to their personal information, from the set of recruitment source options corresponding to the recruitment source feature, which consists of N recruitment source options, select the recruitment source option that matches it. Suppose the value of N is specifically 100, but when selecting the recruitment source option, in most cases, 5 specific recruitment source options out of the 100 recruitment source options are selected. For example, a total of 10,000 selections are made, and among them, the specific 5 recruitment source options are selected 8,000 times. The ratio of 8,000 to 10,000 is 0.8. Suppose the preset ratio is 0.7. Obviously, 0.8 is greater than 0.7. At this time, the remaining 95 recruitment source options among the 100 recruitment source options can be determined as the sparse category values corresponding to the recruitment source feature.

[0143] In the case where there is a recruitment source option that is the data corresponding to the second personnel feature in the second personnel feature, it can be determined whether the recruitment source option is any of the sparse category values corresponding to the recruitment source feature. If the judgment result is yes, it can be considered that the recruitment source option belongs to a relatively sparse recruitment source option. At this time, the recruitment source option can be updated to the first preset category value, for example, updated to other recruitment source options. In this way, all sparse recruitment source options are combined into a new recruitment source option.

[0144] In the case where there are multiple feature values arranged in chronological order corresponding to the third personnel feature in the third feature data, a preset time decay factor can be obtained. The time decay factor can specifically be 0.9, 0.85, 0.80, etc., and will not be listed one by one here. Next, the preset time decay factor can be used to perform time decay processing on the multiple feature data respectively to obtain multiple decay values corresponding to the multiple feature values. There can be a one-to-one correspondence between the multiple decay values and the multiple feature values. After that, the data corresponding to the third personnel feature can be updated from multiple feature values to multiple decay values.

[0145] In a specific example, the third personnel feature is the number of property viewings feature, then the data corresponding to the third personnel feature can be composed of N feature values corresponding to N consecutive days, and each feature value can specifically be a viewing number value. Suppose N is 15, then the data corresponding to the third personnel feature is specifically composed of 15 viewing number values. Suppose a certain viewing number value among the 15 viewing number values is C, the number of days from the date corresponding to the viewing number value to the current date is t, and the preset time decay factor is 0.9, then 0.9 t *C can be calculated, and 0.9 t*C is the decay value corresponding to the viewing times value of C. In a similar manner, 14 decay values corresponding to the other 14 viewing times values can be obtained, thus finally obtaining 15 decay values. After that, the data corresponding to the third personnel feature can be updated from 15 viewing times values to 15 decay values.

[0146] In this implementation manner, based on the preset numerical anomaly truncation condition, the situation where the data in the first feature data is abnormal can be accurately identified. Combining the normalization processing result of the truncated feature values and the update processing of the abnormal data can not only ensure the accuracy and reliability of the data processing result of the first feature data, but also make the data dimensions consistent, avoiding the impact of inconsistent data dimensions on the operation effect of the personnel evaluation model. Based on the preset sparse category value filtering condition, the situation where the data in the second feature data belongs to sparse data can be accurately identified. By merging all the sparse data, new discrete feature data can be recombined, which can better solve the problem of the long-tail effect of the data. In addition, by using the time decay factor to perform time decay processing on the feature values in the third feature data and updating the feature values to decay values, different weights can be given to the feature values at different distances from the current date, thus effectively ensuring the reliability of the data processing result of the third feature data.

[0147] In another specific implementation manner, step 102 includes:

[0148] In the case where the data corresponding to the fourth personnel feature is missing in the first feature data, determine the minimum feature value corresponding to the fourth personnel feature, and add the minimum feature value as the data corresponding to the fourth personnel feature to the first feature data;

[0149] In the case where the data corresponding to the fifth personnel feature is missing in the second feature data, add the second preset category value as the data corresponding to the fifth personnel feature to the second feature data;

[0150] In the case where the data corresponding to the sixth personnel feature is missing in the third feature data, add the preset feature value as the data corresponding to the sixth personnel feature to the third feature data.

[0151] In a specific example, the data corresponding to the fourth personnel feature is missing in the first feature data, and the fourth personnel feature is the credit score feature. Then, the credit score value is specifically missing in the first feature data. At this time, the minimum feature value corresponding to the credit score feature can be determined.

[0152] Assume that the personnel evaluation model is specifically trained based on the personnel information of 10,000 reference personnel. Then, based on the personnel information of each of these 10,000 reference personnel, 10,000 credit scores corresponding to these 10,000 reference personnel can be obtained, and the credit score with the smallest value can be selected from these 10,000 credit scores. The selected credit score can be used as the minimum characteristic value corresponding to the credit score feature.

[0153] After determining the minimum characteristic value corresponding to the credit score feature, the determined minimum characteristic data can be added to the first feature data as the data corresponding to the fourth personnel feature.

[0154] In another specific example, the data corresponding to the fifth personnel feature is missing in the second feature data, and the fifth personnel feature is the recruitment source feature. Then, the recruitment source options are specifically missing in the second feature data. At this time, the second preset category value can be added to the second feature data as the data corresponding to the fifth personnel feature, and the second preset category value can specifically be other recruitment source options.

[0155] In yet another specific example, the data corresponding to the sixth personnel feature is missing in the third feature data, and the sixth personnel feature is the number of property viewings feature. Then, the number of property viewings values are specifically missing in the third feature data. At this time, the preset feature value can be added to the third feature data as the data corresponding to the sixth personnel feature, and the preset feature value can specifically be 0.

[0156] In this implementation manner, for the case where data is missing in the first feature data, data completion can be achieved by adding the minimum characteristic value. For the case where data is missing in the second feature data, data completion can be achieved by adding the second preset category value. For the case where data is missing in the third feature data, data completion can be achieved by adding the preset feature value. That is, whether data is missing in the first feature data, the second feature data, or the third feature data, data completion can be achieved by an appropriate method, thereby better ensuring the reliability of the data processing result of the feature data.

[0157] It can be seen that in the embodiments of the present disclosure, when data processing is performed on the first feature data, the second feature data, and the third feature data respectively, data processing can be performed from different processing perspectives or from the same processing perspective in different ways to better ensure the final data processing effect.

[0158] In an optional example, step 104 includes:

[0159] Determine the stable probability and unstable probability of the personnel to be evaluated according to the feature vectors corresponding to the data processing results of the first feature data, the second feature data, and the third feature data respectively;

[0160] Perform a logarithmic transformation on the instability probability according to the ratio of the instability probability to the stability probability to obtain an instability score corresponding to the instability probability;

[0161] The method further includes:

[0162] In the case where the instability score is within a preset score range, manage the person to be evaluated according to the instability score;

[0163] In the case where the instability score is outside the preset score range, determine the score extreme value in the preset score range that is closest to the instability score, and manage the person to be evaluated according to the determined score extreme value.

[0164] In the embodiments of the present disclosure, according to the feature vectors corresponding to the data processing results of the first feature data, the second feature data, and the third feature data respectively, and in combination with the use of the personnel evaluation model, a stability prediction result of the person to be evaluated can be generated. The stability prediction result of the person to be evaluated may specifically include the instability probability of the person to be evaluated. By subtracting 1 from the instability probability of the person to be evaluated, the stability probability of the person to be evaluated can be obtained.

[0165] Next, according to the ratio of the instability probability to the stability probability (this ratio can also be referred to as the target probability ratio), perform a logarithmic transformation on the instability probability to obtain an instability score corresponding to the instability probability. In a specific implementation manner, the formula used to perform a logarithmic transformation on the instability probability according to the ratio of the instability probability to the stability probability to obtain an instability score corresponding to the instability probability is:

[0166] Score’=Score+Bln(P)-Bln(P’)

[0167] Where Score’ is the instability score, Score is the preset score, B is the preset coefficient, P is the preset ratio, and P’ is the ratio of the instability probability to the stability probability.

[0168] Here, Score can be 65, B can be (-4) / ln2, and P can be 1 / 60. Then, if the target probability ratio is 1 / 60, the instability score is 65 points, and for every 1-fold increase in the target probability ratio, the instability score increases by 4 points.

[0169] In this implementation manner, by substituting the target probability value into the above formula for calculation, the instability score corresponding to the instability probability can be determined very conveniently and reliably.

[0170] After obtaining the unstable score, it can be determined whether the unstable score is within the preset score range; wherein the preset score range can be 0 to 100 points. If the result of the judgment is within, the person to be evaluated can be managed directly according to the unstable score. For example, if the unstable score is greater than the set score (such as 50 points, 60 points, etc.), some intervention measures can be taken, such as communicating with the person to be evaluated about their needs or striving for more resources for the person to be evaluated. If the result of the judgment is not within, since the preset score range of 0 to 100 points has two extreme score values, the extreme score value closest to the unstable score in these two extreme score values ​​can be determined. Assuming that the unstable score is 105 points, the determined extreme score value is 100 points. Since 100 points is greater than the set score, some intervention measures can be taken to achieve the management of the person to be evaluated.

[0171] It can be seen that in the embodiments of the present disclosure, by utilizing the logarithmic transformation logic and combining it with the preset score range, a score value within the preset score range and conforming to the normal distribution can be finally obtained, which can more intuitively characterize the instability tendency of a person and make the instability tendency in the population conform to the normal distribution (for details, see Figure 2 The distribution shown in Figure ).

[0172] In an alternative example, Figure 3 As shown, the personnel information of the real estate broker who is to be evaluated can be obtained as the original data. The original data includes but is not limited to the broker's personal basic information, broker's work behavior information, and the broker's store, team, and city performance information.

[0173] Next, feature screening can be performed on the original data. Optionally, when performing feature screening on the original data, starting from the feature composition structure, static information and time series data can be specifically screened out. Static information refers to information that will not change over time in the short term, including but not limited to information about the city, credit score information, job level information, business opportunity conversion ability information, etc. Time series data refers to the operation data generated every day, including but not limited to the number of house viewings, the number of house entry information, the number of house inspections, business opportunity volume information, transaction performance information, etc.; or, when performing feature screening on the original data, starting from the business classification, basic broker information, viewing-related information, house-related information, customer source-related information, communication-related information, booth purchase-related information, cooperation transaction-related information, conversion transaction-related information, operation preference-related information, performance-related information, negative information, etc. can be specifically screened out.

[0174] After filtering out static information and time series data, data cleaning, data association, and wide table construction can be performed to obtain static continuous data (which is equivalent to the first feature data with the feature classification of static continuous type in the above text), static categorical data (which is equivalent to the second feature data with the feature classification of static discrete type in the above text), and time series data (which is equivalent to the third feature data with the feature classification of dynamic type in the above text), so as to enter the feature engineering stage based on this.

[0175] It should be noted that there may be about 500 feature-related data in the original data. Not all of these features are beneficial to the personnel evaluation model, and there may be correlations and collinearity among the features. To ensure the efficiency of training and deployment of the personnel evaluation model and reduce the interference of invalid features, methods such as IV value, chi-square test, correlation, and collinearity can be used in the feature screening stage to finally retain only nearly 200 features. In this way, the total number of personnel features involved in the first feature data, the second feature data, and the third feature data can be only about 200.

[0176] In the feature engineering stage, the main processing steps include the following aspects:

[0177] Missing value processing: First, analyze the missing values of the features, analyze the proportion of their missing and zero values, and perform different ways of missing value filling according to the type of the feature (discrete / continuous), the proportion of missing values, and the different business meanings. Specifically, for the missing values of operation data, 0 can be directly filled, that is, it is defaulted that there is no operation; for credit scores and other score-type data, the minimum value is filled; for discrete data, a separate "other" type is newly added for filling (corresponding to the latter specific implementation method in step 102 in the above text).

[0178] Outlier processing: Through exploratory data analysis (EDA) of numerical features, it is found that there are outliers in some features. Since outliers will seriously affect the result of data normalization and reduce the discrimination of the personnel evaluation model, the outliers can be truncated at the 98% or 95% position to eliminate the outliers on the premise of retaining most of the information (corresponding to the relevant processing of the first feature data in the former specific implementation method in step 102 in the above text).

[0179] Data normalization: Since the effect of StandardScaler is better in engineering practice, StandardScaler can be selected for data normalization.

[0180] Low-frequency filtering: Through EDA analysis of categorical features, it is found that there is a long-tail effect in the enumerated values of some features. For example, in the recruitment source field, there are nearly 100 enumerated values in total, but 5 enumerated values account for more than 80% of the total samples, and the remaining enumerated values are very sparse. For this situation, we combine all the sparse enumerated values and recombine them into new discrete features (corresponding to the relevant processing of the second feature data in the previous specific implementation manner of step 102 in the above text).

[0181] Time series step size: The step size of time series data is a key issue worthy of consideration. If the step size is set too long, the details of the broker's operations will be lost, which will inevitably affect the model effect of the personnel evaluation model. If it is set too short, such as in days, there will be a large amount of noisy data, and it will be difficult for the personnel evaluation model to fit and converge, and it cannot capture the essential laws of the broker's operations. In addition, a too short step size will also lead to a too large capacity of the training data. Considering various factors, it is finally chosen to use 15 days as a cycle (equivalent to N taking the value of 15 in the above text).

[0182] Through the feature engineering stage, static continuous data after normalization processing (which is equivalent to the data processing result of the first feature data in the above text), static categorical data after low-frequency filtering processing (which is equivalent to the data processing result of the second feature data in the above text), and time series data after time decay processing (which is equivalent to the data processing result of the third feature data in the above text) can be obtained.

[0183] After that, the first sub-model in the personnel evaluation model can convert the static continuous data after normalization processing into Dense Features to obtain the first feature vector; the second sub-model in the personnel evaluation model first converts the static categorical data after low-frequency filtering processing through the Embedding layer (which is beneficial to fully explore the correlation between discrete features), and then extracts the feature vector through GRU to obtain the second feature vector; the third sub-model in the personnel evaluation model first shapes the time series data after time decay processing, then normalizes the data through the Batch Normalization method, and finally extracts the time series features through the LTSM network to obtain the third feature vector, and the third feature vector can effectively reflect the dynamic features of the personnel to be evaluated.

[0184] After that, the Hidden Layer in the personnel evaluation model splices the first feature vector, the second feature vector, and the third feature vector, and provides the spliced vector in the personnel evaluation model to the Full Connection Layer. The Full Connection Layer can finally obtain the instability tendency of the personnel to be evaluated, and the obtained instability tendency can be used for subsequent business calls to achieve the management of the personnel to be evaluated.

[0185] It can be seen that by combining and uniformly training the first sub-model, the second sub-model, and the third sub-model through the end-to-end model, the entire personnel evaluation model can be regarded as a feature extractor, which can extract features from data with different structures and different dimensions, perform data dimensionality reduction operations, and finally reduce the original personnel information to one dimension after layer-by-layer extraction, that is, the final stability prediction result.

[0186] Optionally, the personnel evaluation model can also be optimized, for example, by adjusting the network structure, learning rate, L2 regularization, Batch Size, and the number of iterations, etc., to improve the discrimination and generalization performance of the personnel evaluation model.

[0187] In summary, in the embodiments of the present disclosure, relying on big data technology, quantitatively evaluating stability through machine learning algorithms can accurately, quickly, and comprehensively evaluate the instability tendency of employees, with low costs, and can also be promoted among different cities and industries.

[0188] Any of the personnel stability prediction methods based on the prediction model provided by the embodiments of the present disclosure can be executed by any suitable device with data processing capabilities, including but not limited to: terminal devices and servers, etc. Or, any of the personnel stability prediction methods based on the prediction model provided by the embodiments of the present disclosure can be executed by a processor. For example, the processor executes any of the personnel stability prediction methods mentioned in the embodiments of the present disclosure by calling the corresponding instructions stored in the memory. This will not be elaborated below.

[0189] Exemplary device

[0190] Figure 4 It is a schematic structural diagram of a personnel stability prediction device provided by an exemplary embodiment of the present disclosure. Figure 4 The shown device includes a first determination module 401, a processing module 402, a generation module 403, and a prediction module 404.

[0191] The first determination module 401 is configured to determine first feature data with a feature classification of static continuous type, second feature data with a feature classification of static discrete type, and third feature data with a feature classification of dynamic type according to the personnel information of the personnel to be evaluated.

[0192] A processing module 402, configured to perform data processing on at least one of the first feature data, the second feature data, and the third feature data respectively in a data processing manner matching its feature classification;

[0193] A generation module 403, configured to generate a feature vector corresponding to the data processing result thereof respectively in a feature vector generation manner matching its feature classification for at least one of the first feature data, the second feature data, and the third feature data;

[0194] A prediction module 404, configured to perform a stability prediction on the person to be evaluated according to the generated feature vector.

[0195] In an optional example, the feature vector corresponding to the data processing result of the first feature data is a first feature vector, the feature vector corresponding to the data processing result of the second feature data is a second feature vector, and the feature vector corresponding to the data processing result of the third feature data is a third feature vector;

[0196] The first feature vector is obtained by vector transformation of the data processing result of the first feature data by a first sub-model in the personnel evaluation model, the second feature vector is obtained by dense vector transformation and feature vector extraction of the data processing result of the second feature data by a second sub-model in the personnel evaluation model, and the third feature vector is obtained by sequence shaping, data normalization, and time series feature extraction of the data processing result of the third feature data by a third sub-model in the personnel evaluation model.

[0197] In an optional example, the personnel evaluation model further includes a hidden layer and a fully connected layer. The hidden layer is respectively connected to the first sub-model, the second sub-model, and the third sub-model, and the fully connected layer is connected to the hidden layer. The hidden layer is configured to splice the first feature vector, the second feature vector, and the third feature vector to obtain a spliced vector, and the fully connected layer is configured to generate a stability prediction result of the person to be evaluated according to the spliced vector.

[0198] In an optional example, the processing module 402 includes:

[0199] A first processing sub-module, configured to determine a normalized processing result of the truncated feature value corresponding to the first personnel feature and update the data corresponding to the first personnel feature to the normalized processing result when there is data corresponding to the first personnel feature in the first feature data and the data corresponding to the first personnel feature satisfies a preset numerical anomaly truncation condition;

[0200] A second processing sub-module, configured to update the data corresponding to the second personnel feature to a first preset class value when there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature meets a preset sparse class value filtering condition;

[0201] A third processing sub-module, configured to, when there is data corresponding to a third personnel feature in the third feature data, which is composed of a plurality of feature values arranged in chronological order, perform time decay processing on the plurality of feature values respectively by using a preset time decay factor to obtain a plurality of decay values corresponding to the plurality of feature values, and update the data corresponding to the third personnel feature to the plurality of decay values.

[0202] In an optional example,

[0203] The apparatus further includes:

[0204] An acquisition module, configured to acquire the personnel information of each of a plurality of reference personnel used for training the personnel evaluation model when performing stability prediction on a to-be-evaluated person based on the personnel evaluation model;

[0205] A selection module, configured to, for each of the plurality of reference personnel, select a class value that matches it from a preset class value set corresponding to the second personnel feature and composed of N class values according to its personnel information;

[0206] A second determination module, configured to determine each of the remaining N - M class values among the N class values as the sparse class value corresponding to the second personnel feature when the ratio of the total selection times of M class values among the N class values to the total selection times of the N class values is greater than a preset ratio;

[0207] The second processing sub-module is specifically configured to:

[0208] When there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature is any sparse class value corresponding to the second personnel feature, update the data corresponding to the second personnel feature to the first preset class value.

[0209] In an optional example, the processing module 402 includes:

[0210] A fourth processing sub-module, configured to determine the minimum feature value corresponding to the fourth personnel feature and add the minimum feature value as the data corresponding to the fourth personnel feature to the first feature data when the data corresponding to the fourth personnel feature is missing in the first feature data;

[0211] The fifth processing sub-module is used to add the second preset category value as the data corresponding to the fifth personnel feature to the second feature data when the data corresponding to the fifth personnel feature is missing in the second feature data;

[0212] The sixth processing sub-module is used to add the preset feature value as the data corresponding to the sixth personnel feature to the third feature data when the data corresponding to the sixth personnel feature is missing in the third feature data.

[0213] In an optional example,

[0214] The prediction module 404 includes:

[0215] The determination sub-module is used to determine the stable probability and the unstable probability of the person to be evaluated according to the feature vectors corresponding to the data processing results of the first feature data, the second feature data, and the third feature data respectively;

[0216] The acquisition sub-module is used to perform a logarithmic conversion process on the unstable probability according to the ratio of the unstable probability to the stable probability to obtain an unstable score corresponding to the unstable probability;

[0217] The device further includes:

[0218] The first management module is used to manage the person to be evaluated according to the unstable score when the unstable score is within the preset score range;

[0219] The second management module is used to determine the score extreme value in the preset score range that is closest to the unstable score when the unstable score is outside the preset score range, and manage the person to be evaluated according to the determined score extreme value.

[0220] In an optional example, the formula used by the acquisition sub-module to perform a logarithmic conversion process on the unstable probability according to the ratio of the unstable probability to the stable probability to obtain an unstable score corresponding to the unstable probability is:

[0221] Score’ = Score + Bln(P) - Bln(P’)

[0222] where Score’ is the unstable score, Score is the preset score, B is the preset coefficient, P is the preset ratio, and P’ is the ratio of the unstable probability to the stable probability.

[0223] Exemplary electronic device

[0224] Next, refer to Figure 5Describe an electronic device according to an embodiment of the present disclosure. The electronic device may be either or both of the first device and the second device, or a stand-alone device independent of them, and the stand-alone device may communicate with the first device and the second device to receive the collected input signals from them.

[0225] Figure 5 FIG. illustrates a block diagram of an electronic device 500 according to an embodiment of the present disclosure.

[0226] As Figure 5 shown, the electronic device 500 includes one or more processors 501 and a memory 502.

[0227] The processor 501 may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 500 to perform desired functions.

[0228] The memory 502 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 501 may run the program instructions to implement the personnel stability prediction method based on the prediction model and / or other desired functions of the various embodiments of the present disclosure described above. Various contents such as input signals, signal components, noise components, etc. may also be stored in the computer-readable storage medium.

[0229] In one example, the electronic device 500 may further include: an input device 503 and an output device 504, and these components are interconnected through a bus system and / or other forms of connection mechanisms (not shown).

[0230] For example, when the electronic device 500 is the first device or the second device, the input device 503 may be a microphone or a microphone array. When the electronic device 500 is a stand-alone device, the input device 503 may be a communication network connector for receiving the collected input signals from the first device and the second device.

[0231] In addition, the input device 503 may further include, for example, a keyboard, a mouse, and so on.

[0232] The output device 504 can output various information to the outside. The output device 504 can include, for example, a display, a speaker, a printer, a communication network, and a remote output device connected thereto, and so on.

[0233] Of course, for simplicity, Figure 5 only some of the components of the electronic device 500 related to the present disclosure are shown in [the figure], and components such as a bus, an input / output interface, etc. are omitted. In addition, according to specific application scenarios, the electronic device 500 may further include any other appropriate components.

[0234] Exemplary computer program product and computer-readable storage medium

[0235] In addition to the above methods and devices, an embodiment of the present disclosure may also be a computer program product, which includes computer program instructions that, when run by a processor, cause the processor to execute the steps in the prediction model-based personnel stability prediction method according to various embodiments of the present disclosure described in the "Exemplary Method" section above of this specification.

[0236] The computer program product can be written in any combination of one or more programming languages to write program code for performing the operations of the embodiments of the present disclosure. The programming languages include object-oriented programming languages such as Java, C++, etc., and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0237] Furthermore, an embodiment of the present disclosure may also be a computer-readable storage medium, on which computer program instructions are stored, and the computer program instructions, when run by a processor, cause the processor to execute the steps in the prediction model-based personnel stability prediction method according to various embodiments of the present disclosure described in the "Exemplary Method" section above of this specification.

[0238] The computer-readable storage medium may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0239] The basic principles of the present disclosure have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, benefits, effects, etc. mentioned in the present disclosure are only examples and not limitations, and it cannot be considered that these advantages, benefits, effects, etc. are essential for each embodiment of the present disclosure. Additionally, the above-described specific details are only for illustrative and facilitating understanding purposes and are not limitations. The above details do not limit the present disclosure to necessarily adopt the above specific details for implementation.

[0240] Each embodiment in this specification is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For system embodiments, since they basically correspond to method embodiments, the description is relatively simple, and reference can be made to the relevant parts of the method embodiments for the relevant content.

[0241] The block diagrams of the devices, apparatuses, equipment, and systems involved in the present disclosure are only illustrative examples and do not intend to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open-ended terms meaning "including but not limited to" and can be used interchangeably with each other. The word "or" and "and" used herein refer to the word "and / or" and can be used interchangeably with each other, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to" and can be used interchangeably with each other.

[0242] The methods and apparatuses of the present disclosure can be implemented in many ways. For example, the methods and apparatuses of the present disclosure can be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of the steps for the method is for illustration only. The steps of the method of the present disclosure are not limited to the specific order described above, unless otherwise specifically stated. In addition, in some embodiments, the present disclosure can also be implemented as a program recorded in a recording medium, and these programs include machine-readable instructions for implementing the method according to the present disclosure. Therefore, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

[0243] It should also be noted that in the apparatuses, devices, and methods of the present disclosure, each component or each step can be decomposed and / or recombined. These decompositions and / or recombinations should be regarded as equivalent solutions of the present disclosure.

[0244] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0245] The above description has been given for purposes of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although several example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and subcombinations thereof.

Claims

1. A method for predicting personnel stability based on a prediction model, characterized in that, it includes: Determine the first feature data with a feature classification of static continuous type, the second feature data with a feature classification of static discrete type, and the third feature data with a feature classification of dynamic type according to the personnel information of the personnel to be evaluated. The first feature data is static continuous data, the second feature data is static categorical data, and the third feature data is time series data; For at least one of the first feature data, the second feature data, and the third feature data, perform data processing on it respectively with a data processing method matching its feature classification to obtain a data processing result; among them, the data processing method matching the feature classification of the first feature data is normalization processing, and the obtained data processing result is normalized static continuous data; the data processing method matching the feature classification of the second feature data is low-frequency filtering, and the obtained data processing result is static categorical data obtained by low-frequency filtering. The data processing method matching the feature classification of the third feature data is to perform time decay processing on multiple feature values arranged in chronological order included in the third feature data respectively based on a preset time decay factor, and the obtained data processing result is time series data obtained by time decay processing; For at least one of the first feature data, the second feature data, and the third feature data, respectively determine the feature vector generation method matching its feature classification according to the correspondence between the preset feature classification and the feature vector generation method, and generate the feature vector corresponding to its data processing result with the feature vector generation method matching its feature classification; Perform stability prediction on the personnel to be evaluated according to the generated feature vector.

2. The method according to claim 1, characterized in that, The feature vector corresponding to the data processing result of the first feature data is the first feature vector, the feature vector corresponding to the data processing result of the second feature data is the second feature vector, and the feature vector corresponding to the data processing result of the third feature data is the third feature vector; The first feature vector is obtained by vector transformation of the data processing result of the first feature data by the first sub-model in the personnel evaluation model, the second feature vector is obtained by dense vector transformation and feature vector extraction of the data processing result of the second feature data by the second sub-model in the personnel evaluation model, and the third feature vector is obtained by sequence shaping, data normalization, and time series feature extraction of the data processing result of the third feature data by the third sub-model in the personnel evaluation model.

3. The method according to claim 2, characterized in that, The personnel evaluation model further includes a hidden layer and a fully connected layer. The hidden layer is respectively connected to the first sub-model, the second sub-model, and the third sub-model. The fully connected layer is connected to the hidden layer. The hidden layer is used to splice the first feature vector, the second feature vector, and the third feature vector to obtain a spliced vector. The fully connected layer is used to generate a stability prediction result of the person to be evaluated according to the spliced vector.

4. According to the method described in claim 1, wherein, the data processing of at least one of the first feature data, the second feature data, and the third feature data respectively by a data processing method matching its feature classification includes: when there is data corresponding to a first personnel feature in the first feature data and the data corresponding to the first personnel feature meets a preset numerical anomaly truncation condition, determining a normalization result of the truncated feature value corresponding to the first personnel feature, and updating the data corresponding to the first personnel feature to the normalization result; when there is data corresponding to a second personnel feature in the second feature data and the data corresponding to the second personnel feature meets a preset sparse category value filtering condition, updating the data corresponding to the second personnel feature to a first preset category value; when there is data corresponding to a third personnel feature in the third feature data, which consists of a plurality of feature values arranged in chronological order, using a preset time decay factor to perform time decay processing on the plurality of feature values respectively to obtain a plurality of decay values corresponding to the plurality of feature values, and updating the data corresponding to the third personnel feature to the plurality of decay values.

5. According to the method described in claim 4, wherein, when performing stability prediction on the person to be evaluated based on the personnel evaluation model, the method further includes: obtaining the personnel information of each of a plurality of reference persons used for training the personnel evaluation model; for each of the plurality of reference persons, selecting a category value matching it from a preset category value set composed of N category values corresponding to the second personnel feature according to its personnel information; when the ratio of the total selection times of M category values among the N category values to the total selection times of the N category values is greater than a preset ratio, determining each of the remaining N - M category values among the N category values as the sparse category value corresponding to the second personnel feature; the step of, when there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature meets the preset sparse category value filtering condition, updating the data corresponding to the second personnel feature to the first preset category value, includes: when there is data corresponding to the second personnel feature in the second feature data and the data corresponding to the second personnel feature is any sparse category value corresponding to the second personnel feature, updating the data corresponding to the second personnel feature to the first preset category value.

6. The method according to claim 1, wherein, the stability prediction of the person to be evaluated based on the generated feature vectors includes: determining the stable probability and the unstable probability of the person to be evaluated according to the feature vectors corresponding to the data processing results of the first feature data, the second feature data, and the third feature data respectively; performing logarithmic conversion processing on the unstable probability according to the ratio of the unstable probability to the stable probability to obtain an unstable score corresponding to the unstable probability; the method further includes: when the unstable score is within a preset score range, managing the person to be evaluated according to the unstable score; when the unstable score is outside the preset score range, determining the score extreme value in the preset score range that is closest to the unstable score, and managing the person to be evaluated according to the determined score extreme value.

7. The method according to claim 6, wherein, performing logarithmic conversion processing on the unstable probability according to the ratio of the unstable probability to the stable probability to obtain an unstable score corresponding to the unstable probability, and the formula used is: Score’ = Score + Bln(P) - Bln(P’) where Score’ is the unstable score, Score is a preset score, B is a preset coefficient, P is a preset ratio, and P’ is the ratio of the unstable probability to the stable probability.

8. A personnel stability prediction device based on a prediction model, wherein, it includes: a first determination module, configured to determine first feature data with a feature classification of static continuous type, second feature data with a feature classification of static discrete type, and third feature data with a feature classification of dynamic type according to the personnel information of the person to be evaluated, where the first feature data is static continuous data, the second feature data is static category data, and the third feature data is time series data; a processing module, configured to perform data processing on at least one of the first feature data, the second feature data, and the third feature data respectively in a data processing manner matching its feature classification to obtain a data processing result; wherein, the data processing manner matching the feature classification of the first feature data is normalization processing, and the obtained data processing result is normalized static continuous data; the data processing manner matching the feature classification of the second feature data is low-frequency filtering, and the obtained data processing result is static category data obtained by low-frequency filtering, and the data processing manner matching the feature classification of the third feature data is to perform time decay processing on multiple feature values included in the third feature data arranged in chronological order based on a preset time decay factor, and the obtained data processing result is time series data obtained by time decay processing; A generation module, configured to, for at least one of the first feature data, the second feature data, and the third feature data, respectively determine a feature vector generation method that matches its feature classification according to the correspondence between the preset feature classification and the feature vector generation method, and generate a feature vector corresponding to the data processing result thereof in a feature vector generation method that matches its feature classification; A prediction module, configured to perform a stability prediction on the person to be evaluated according to the generated feature vector.

9. A computer-readable storage medium, storing a computer program, characterized in that, the computer program is used to execute the method for predicting the stability of a person based on a prediction model according to any one of claims 1 to 7 above.

10. An electronic device, characterized in that, it includes: a processor; a memory for storing executable instructions that can be executed by the processor; the processor is configured to read the executable instructions from the memory and execute the instructions to implement the method for predicting the stability of a person based on a prediction model according to any one of claims 1 to 7 above.

Images