WO2020168754A1 - 基于预测模型的绩效预测方法、装置及存储介质 - Google Patents

基于预测模型的绩效预测方法、装置及存储介质 Download PDF

Info

Publication number
WO2020168754A1
WO2020168754A1 PCT/CN2019/117787 CN2019117787W WO2020168754A1 WO 2020168754 A1 WO2020168754 A1 WO 2020168754A1 CN 2019117787 W CN2019117787 W CN 2019117787W WO 2020168754 A1 WO2020168754 A1 WO 2020168754A1
Authority
WO
WIPO (PCT)
Prior art keywords
category
discrete
factors
model
cluster
Prior art date
Application number
PCT/CN2019/117787
Other languages
English (en)
French (fr)
Inventor
徐亮
金戈
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2020168754A1 publication Critical patent/WO2020168754A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Definitions

  • This application relates to the field of deep learning, and in particular to a performance prediction method, device and storage medium based on a prediction model.
  • deep neural network models show better accuracy in the field of machine learning, many intelligent applications including computer vision, speech recognition and robotics have widely used deep neural network models.
  • the modeling of complex events is usually based on deep neural network models.
  • the deep neural network model is applied to performance prediction, its performance prediction accuracy is low.
  • the embodiments of the present application provide a performance prediction method, device and storage medium based on a prediction model, which can improve performance prediction accuracy.
  • an embodiment of the present application provides a performance prediction method based on a prediction model, including:
  • the performance preset model is obtained by training based on a training data set
  • the training data set includes a first cluster label and a continuous factor in a preset factor system
  • the first cluster label is based on the The discrete factor in the factor system is obtained.
  • an embodiment of the present application provides a performance prediction device based on a prediction model, including:
  • the classification unit is used to classify the discrete factors in the target performance prediction information to be predicted to obtain the discrete factors in different categories;
  • the first prediction unit is configured to use classification models corresponding to the discrete factors in the different categories to perform label prediction on the discrete factors in the different categories to obtain at least one classification label corresponding to each category;
  • the second prediction unit is used to input the at least one classification label corresponding to each category and the continuous factors in the target performance prediction information into a preset performance prediction model for performance prediction, and obtain the target performance prediction information corresponding The user’s performance prediction results; wherein the performance preset model is obtained by training according to a training data set; the training data set includes a cluster label and a continuous factor in a preset factor system, and the cluster label is According to the discrete factor in the factor system, the cluster label is the first cluster label or the second cluster label.
  • an embodiment of the present application provides an electronic device, including a processor, an input device, an output device, and a memory.
  • the processor, input device, output device, and memory are connected to each other, wherein the memory is used for storing A computer program, the computer program comprising program instructions, and the processor is configured to invoke the program instructions to perform the method as described in the first aspect.
  • an embodiment of the present application provides a computer non-volatile readable storage medium, the computer non-volatile readable storage medium stores a computer program, the computer program includes program instructions, and the program instructions When executed by a processor, the processor is caused to execute the method as described in the first aspect.
  • the electronic device can classify the discrete factors in the target performance prediction information to be predicted to obtain the discrete factors under different categories, and perform label prediction on the discrete factors under the different categories to obtain at least
  • a classification label combines the continuous factors in the target performance prediction information to perform performance prediction to obtain performance prediction results, which improves the accuracy of performance prediction.
  • FIG. 1 is a schematic flowchart of a performance prediction method based on a prediction model provided by an embodiment of the present application
  • FIG. 2 is a schematic flowchart of another performance prediction method based on a prediction model provided by an embodiment of the present application
  • FIG. 3 is a schematic structural diagram of a performance prediction device based on a prediction model provided by an embodiment of the present application
  • Fig. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.
  • FIG. 1 is a schematic flowchart of a performance prediction method based on a prediction model provided by an embodiment of this application.
  • This method can be applied to electronic devices.
  • the electronic device can be a terminal or a server.
  • the method may include:
  • the target performance prediction information refers to information used by the user for performance prediction, such as information on dimensions such as the user's personality and work.
  • the target performance prediction information may include continuous factors and discrete factors.
  • the continuous factor is used to represent continuous information.
  • the continuous information refers to information whose characteristic value is a numerical value.
  • the characteristic value can be understood as a value.
  • the continuous factor is the length of study (such as 1 hour) and the length of travel to work (such as 1 hour).
  • the discrete factor is used to represent discrete information.
  • the discrete information refers to information whose characteristic value is not a numerical value. If the discrete factor is absenteeism (such as yes or no), work attitude (such as positive or negative).
  • the electronic device may receive a classification instruction, and classify the discrete factors in the target performance prediction information according to the received classification instructions to obtain the discrete factors in different categories.
  • the discrete factors in the target performance prediction information can be classified into discrete factors under the job category, discrete factors under the personality category, and discrete factors under the learning category according to the received classification instruction.
  • the discrete factor under each category can correspond to a classification model.
  • the discrete factor under the work category corresponds to the first classification model
  • the discrete factor under the learning category corresponds to the second classification model
  • the discrete factor under the learning category corresponds to the third classification model.
  • the classification model may be obtained by training according to the first cluster label and the discrete factors in the preset factor system. Specifically, the classification model may be obtained by using a deep learning method and training according to the first cluster label and the discrete factors in the preset factor system.
  • the classification model may include a label prediction strategy obtained by training.
  • the first cluster label may be obtained according to the discrete factor in the factor system.
  • the first cluster label may be obtained by classifying the discrete factors in the factor system to obtain the discrete factors in different categories, and clustering the discrete factors in each category.
  • the first cluster label may be obtained by classifying the discrete factors in the factor system to obtain discrete factors in different categories, and clustering the discrete factors in each category through a preset model.
  • the preset model includes but is not limited to k-model and other models that can be used for cluster analysis.
  • the first clustering label can also be used to classify the discrete factors in the factor system to obtain discrete factors in different categories, and cluster the discrete factors in each category through a clustering algorithm. To the label setting instruction set.
  • the classification model may also be obtained by training according to the second cluster label and the discrete factors in the preset factor system.
  • the classification model may also be obtained by using a deep learning method and training according to the second cluster label and the discrete factors in the preset factor system.
  • the second cluster label may be obtained according to the number of adjusted cluster types and the discrete factor in the factor system.
  • the second cluster label may be obtained after clustering the discrete factors in each category again through a preset model according to the number of adjusted cluster types.
  • the preset model is a k-model
  • the number of types of clusters after adjustment is the adjusted k value.
  • the second clustering label may be set according to the received label setting instruction after clustering the discrete factors under each category again by using a clustering algorithm according to the number of adjusted cluster types.
  • the electronic device may use the classification models corresponding to the discrete factors in the different categories to perform label prediction on the discrete factors in the different categories to obtain at least one classification label corresponding to each category.
  • the embodiment of the present application can also obtain the importance value of each discrete factor by using the classification model, which is beneficial to the subsequent analysis of the contribution degree of each discrete factor to the performance prediction.
  • the electronic device uses the first classification model to predict the label of the discrete factor under the job category to obtain the label 1 corresponding to the discrete factor under the job category; the electronic device uses the second classification model to predict the label of the discrete factor under the personality category. Obtain label 2 and label 3 corresponding to the discrete factor under the personality category; the electronic device uses the third classification model to predict the label of the discrete factor under learning to obtain label 4 and label 5 corresponding to the discrete factor under the learning category.
  • the electronic device uses the classification models corresponding to the discrete factors in the different categories to perform label prediction on the discrete factors in the different categories to obtain at least one classification label corresponding to each category, which may include:
  • the discrete factors below are respectively input into the corresponding classification model to perform label prediction, and the classification labels of the discrete factors in the different categories are respectively output through the corresponding classification model.
  • the electronic device separately inputting the discrete factors under different categories into corresponding classification models to perform label prediction may include: inputting the characteristic values of the discrete factors under different categories into the corresponding classification models.
  • the electronic device may also characterize the characteristic value as a characteristic vector before inputting the characteristic values of the discrete factors in the different categories into the corresponding classification model.
  • the characteristic vector includes but is not limited to one-hot And other forms.
  • the above process of determining the feature vector of the eigenvalue can be as follows: if the absence is yes, that is, the feature value of the discrete factor is "yes”, then the feature vector corresponding to the eigenvalue can be For [0 1]. If the absence is no, that is, the characteristic value of the discrete factor is "No”, the characteristic vector corresponding to the characteristic value can be [1 0].
  • the performance prediction model may include a trained performance prediction strategy.
  • the performance preset model may be obtained by training according to the training data set.
  • the performance prediction model may be obtained by using a deep learning method and training according to a training data set.
  • the training data set may include cluster labels and continuous factors in a preset factor system.
  • the cluster labels are obtained according to discrete factors in the factor system.
  • the cluster labels are the first cluster label or the second cluster label. Class label.
  • the electronic device may input at least one classification label corresponding to each category and the continuous factor in the target performance prediction information into a preset performance prediction model for performance prediction, and output the target through the performance prediction model
  • the user's performance prediction result corresponding to the performance prediction information may include the predicted performance level, such as 1, 2, 3, or high, medium, or low. Or, the performance prediction result may also include a probability of being at a corresponding performance level, such as 90%.
  • the electronic device inputs the at least one classification label corresponding to each category and the continuous factors in the target performance prediction information into a preset performance prediction model to perform performance prediction, which may include: At least one classification label corresponding to the category and the characteristic value of the continuous factor in the target performance prediction information are input into a preset performance prediction model to perform performance prediction.
  • the electronic device may output the performance prediction result for display to relevant personnel.
  • the electronic device can classify the discrete factors in the target performance prediction information to be predicted to obtain the discrete factors in different categories, and use the classification model corresponding to the discrete factors in the different categories.
  • the prediction model is used for performance prediction, and the user's performance prediction result corresponding to the target performance prediction information is obtained, and the accuracy of the performance prediction is improved.
  • FIG. 2 is a schematic flowchart of a performance prediction method based on a prediction model provided by an embodiment of this application.
  • the method can be applied to an electronic device, and the electronic device can be a terminal or a server.
  • the method may include:
  • the electronic device can obtain the performance prediction information of at least one user to generate a factor system including the performance prediction information of the at least one user. Since the performance prediction information of the at least one user may include discrete factors and continuous factors, the factor system may include discrete factors and continuous factors.
  • the electronic device may obtain the stored performance prediction information of the at least one user. If the electronic device itself does not store the performance prediction information of the at least one user, the electronic device may obtain the performance prediction information of the at least one user from other devices.
  • the embodiment of the present application does not limit the method of obtaining the performance prediction information of the at least one user.
  • the electronic device can classify the discrete factors in the factor system to obtain discrete factors in different categories.
  • the factor system includes discrete factors x1, x2, x3, x4, x5, y1, y2, y3, y4, and y5.
  • Electronic devices can be divided into personality categories by x1, x2, x3, x4, and x5, and y1, y2, y3, y4, and y5 are divided into job categories.
  • the electronic device classifies the discrete factors in the factor system to obtain discrete factor electronic devices in different categories, which may include: the electronic device receives a classification instruction, and performs classification on the discrete factors in the factor system according to the received classification instruction. Factors are classified to obtain discrete factors in different categories. For example, the discrete factors in the factor system can be classified into discrete factors under the job category, discrete factors under the personality category, and discrete factors under the learning category according to the received classification instruction.
  • the electronic device classifies the discrete factors in the factor system to obtain discrete factor electronic devices in different categories. It may also include: the electronic device performs a calculation on different discrete factors according to the preset correspondence between the discrete factors and the categories. Classify, get the discrete factors of different classes. Dividing according to the corresponding relationship makes the classification process more automated.
  • the electronic device can cluster the discrete factors in each category through a preset model to obtain the cluster label corresponding to each category.
  • the preset model includes but is not limited to K-model and other models that can be used for cluster analysis.
  • K-model can use clustering algorithms, such as K-means clustering algorithm, to cluster the discrete factors under each category.
  • the embodiment of the present application can classify similar discrete factors in each category into the same category through clustering.
  • the electronic device clusters the discrete factors under each category through a preset model to obtain the first cluster label corresponding to each category, including: the electronic device uses the preset model to discretize each category Clustering of the factors to obtain at least one sub-category corresponding to each category; the electronic device sets a cluster label for each sub-category in the at least one sub-category to obtain at least one cluster label corresponding to each category; The electronic device uses the at least one cluster label corresponding to each subcategory as the first cluster label corresponding to each category.
  • the discrete factors under the personality category include x1, x2, x3, x4, and x5.
  • x1 and x2 are classified into the same category, such as the first category
  • X5 are classified into the same category, such as the second category.
  • at least one sub-category corresponding to the personality category includes the first category and the second category, and label 1 is set for the first category, and label 2 is set for the second category.
  • the first cluster labels corresponding to the personality category are label 1 and label 2.
  • the discrete factors under the job category include y1, y2, y3, y4, and y5.
  • y1, y2, and y3 are classified into the same category, such as the third Category
  • y4 and y5 are classified into the same category, such as the fourth category.
  • at least one subcategory corresponding to the work category is for the third category and the fourth category
  • label 3 is set for the third category
  • label 4 is set for the fourth category.
  • the first cluster labels corresponding to the job category are label 3 and label 4.
  • S204 Establish a performance prediction model according to the first cluster label corresponding to each category and the continuous factors in the factor system.
  • the electronic device can establish a performance prediction model according to the first cluster label corresponding to each category and the continuous factors in the factor system.
  • the process of establishing the performance prediction model can be implemented based on the SK-Learn library.
  • the electronic device establishes a performance prediction model according to the first cluster label corresponding to each category and the continuous factors in the factor system, which may include: the electronic device calculates the first cluster corresponding to each category A clustering label and each continuous factor in the factor system are input to a first gradient boosting decision tree (GBDT) model to train the first GBDT model; the electronic device will The GBDT model is used as a performance prediction model.
  • GBDT gradient boosting decision tree
  • the electronic device inputs the first cluster label corresponding to each category and each continuous factor in the factor system into the first GBDT model, which may include: the electronic device inputs the corresponding to each category The first cluster label and the characteristic value of each continuous factor in the factor system are input into the first GBDT model,
  • the embodiment of the present application may establish a performance prediction model whose prediction accuracy meets preset conditions by adjusting the number of cluster types.
  • the electronic device establishes a performance prediction model according to the first cluster label corresponding to each category and the continuous factors in the factor system, which may include: when the prediction accuracy of the performance prediction model does not meet a preset condition
  • the electronic device adjusts the number of cluster types; the electronic device performs clustering again on the discrete factors of each category through the preset model according to the adjusted number of cluster types to obtain the corresponding The second cluster label; the electronic device establishes a performance prediction model whose prediction accuracy meets preset conditions according to the second cluster label corresponding to each category and the continuous factors in the factor system.
  • the preset model is K-model
  • the number of types of the adjusted clusters is the adjusted k value.
  • the electronic device uses the discrete factors under each category and the first cluster label corresponding to each category to establish a corresponding classification model for the discrete factors under each category, which may include:
  • the discrete factors under each category and the first cluster label corresponding to each category are input into the second GBDT model corresponding to each category for training; the second electronic device will train the second GBDT model corresponding to each category after training.
  • 2. GBDT model as the classification model corresponding to the discrete factor under each category.
  • the second GBDT model and the first GBDT model can be the same or different, and can be set according to actual conditions.
  • the process of establishing the classification model can be implemented based on the SK-Learn library.
  • the electronic device separately inputting the discrete factor under each category and the first cluster label corresponding to each category into the second GBDT model corresponding to each category for training, which may include: The electronic device inputs the characteristic value of the discrete factor under each category and the cluster label corresponding to each category into the second GBDT model for training, and uses the trained second GBDT model as the classification model. Or, when determining the feature vector corresponding to the feature value, the feature vector corresponding to the feature value of the discrete factor under each category and the cluster label corresponding to each category may be input into the second GBDT model for training.
  • S206 Classify the discrete factors in the target performance prediction information to be predicted to obtain discrete factors in different categories;
  • S207 Use the classification models corresponding to the discrete factors in the different categories to perform label prediction on the discrete factors in the different categories to obtain at least one classification label corresponding to each category;
  • steps S206-S208 may refer to steps S101-S103 in the embodiment of FIG. 1, and details are not described in this embodiment of the present application.
  • the electronic device can classify the discrete factors in the constructed factor system to obtain the discrete factors in different categories, and cluster the discrete factors in different categories to obtain the corresponding to each category.
  • the performance prediction model can be established based on the first cluster label and the continuous factors in the factor system; and the electronic device can also be based on the discrete factor under each category and the first cluster corresponding to each category Labels, to establish a corresponding classification model for the discrete factors under each category, so as to realize the construction process of the performance prediction model and the classification model, and the model established by this method has good interpretability.
  • the electronic device can classify the discrete factors in the target performance prediction information to be predicted to obtain the discrete factors under different categories, and use the classification model corresponding to the discrete factors under the different categories to perform label prediction on the discrete factors under the different categories , To obtain at least one classification label corresponding to each category, to input the at least one classification label corresponding to each category and the continuous factors in the target performance prediction information into a preset performance prediction model to perform performance prediction to obtain the target performance The user's performance prediction result corresponding to the prediction information improves the accuracy of the performance prediction.
  • FIG. 3 is a schematic structural diagram of a performance prediction device based on a prediction model provided by an embodiment of this application.
  • the device can be applied to electronic equipment.
  • the electronic device can be a terminal or a server.
  • the device may include:
  • the classification unit 31 is used to classify the discrete factors in the target performance prediction information to be predicted to obtain discrete factors in different categories;
  • the first prediction unit 32 is configured to use the classification models corresponding to the discrete factors in the different categories to perform label prediction on the discrete factors in the different categories to obtain at least one classification label corresponding to each category;
  • the second prediction unit 33 is configured to input at least one classification label corresponding to each category and the continuous factor in the target performance prediction information into a preset performance prediction model to perform performance prediction, and obtain the target performance prediction information Corresponding user performance prediction results; wherein, the performance preset model is obtained by training according to a training data set; the training data set includes cluster labels and continuous factors in a preset factor system, and the cluster labels It is obtained according to the discrete factor in the factor system, and the cluster label is the first cluster label or the second cluster label.
  • the processing unit 34 is configured to construct a factor system; classify the discrete factors in the factor system to obtain discrete factors in different categories; use a preset model to determine the discrete factors in each category.
  • the discrete factors are clustered to obtain the first cluster label corresponding to each category; and the performance prediction model is established according to the first cluster label corresponding to each category and the continuous factors in the factor system.
  • the processing unit 34 clusters the discrete factors in each category through a preset model to obtain the first cluster label corresponding to each category. Specifically, the processing unit 34 performs clustering on each category through the preset model. Perform clustering on the discrete factor of the category to obtain at least one subcategory corresponding to each category; set a cluster label for each subcategory in the at least one subcategory to obtain at least one cluster label corresponding to each category ; Use the at least one cluster label corresponding to each subcategory as the first cluster label corresponding to each category.
  • the processing unit 34 establishes a performance prediction model according to the first cluster label corresponding to each category and the continuous factors in the factor system, and specifically corresponds to each category
  • the first cluster label of and each continuous factor in the factor system are input to the first gradient boosting decision tree GBDT model to train the first GBDT model; the first GBDT model after training is used as the performance prediction model.
  • the processing unit 34 is further configured to cluster the discrete factors under each category to obtain the first cluster label corresponding to each category, and then use each The discrete factor under the category and the first cluster label corresponding to each category establish a corresponding classification model for the discrete factor under each category for label prediction.
  • the processing unit 34 uses the discrete factor under each category and the first cluster label corresponding to each category to establish a corresponding classification model for the discrete factor under each category. , Specifically, input the discrete factor under each category and the first cluster label corresponding to each category into the second GBDT model corresponding to each category for training; each category after training corresponds to The second GBDT model is used as the classification model corresponding to the discrete factor under each category.
  • the processing unit 34 is further configured to adjust the number of types of clusters when the prediction accuracy of the performance prediction model does not meet the preset conditions; according to the adjusted number of types of clusters , Clustering the discrete factors under each category again through the preset model to obtain the second cluster label corresponding to each category; according to the second cluster label corresponding to each category and the The continuous factors in the factor system establish a performance prediction model whose prediction accuracy meets the preset conditions.
  • the electronic device can classify the discrete factors in the target performance prediction information to be predicted to obtain the discrete factors in different categories, and use the classification model corresponding to the discrete factors in the different categories.
  • the prediction model is used for performance prediction, and the user's performance prediction result corresponding to the target performance prediction information is obtained, and the accuracy of the performance prediction is improved.
  • FIG. 4 is a schematic structural diagram of an electronic device provided by an embodiment of this application.
  • the electronic device described in this embodiment may include: one or more processors 1000, one or more input devices 2000, one or more output devices 3000, and a memory 4000.
  • the processor 1000, the input device 2000, the output device 3000, and the memory 4000 may be connected by a bus.
  • the input device 2000 and the output device 3000 may be standard wired or wireless communication interfaces.
  • the processor 1000 may be a central processing module (Central Processing Unit, CPU), the processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (ASIC) , Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
  • the memory 4000 may be a high-speed RAM memory, or a non-volatile memory (non-volatile memory), such as a magnetic disk memory.
  • the memory 4000 is used to store a set of program codes, and the input device 2000, the output device 3000, and the processor 1000 can call the program codes stored in the memory 4000. specifically:
  • the processor 1000 is configured to classify the discrete factors in the target performance prediction information to be predicted to obtain the discrete factors in different categories; use the classification models corresponding to the discrete factors in the different categories to analyze the discrete factors in the different categories.
  • Factors perform label prediction to obtain at least one classification label corresponding to each category; input at least one classification label corresponding to each category and continuous factors in the target performance prediction information into a preset performance prediction model to perform performance prediction , Obtain the performance prediction result of the user corresponding to the target performance prediction information; wherein, the performance preset model is obtained by training according to a training data set; the training data set includes cluster labels and preset factor systems The continuous factor, the cluster label is obtained according to the discrete factor in the factor system, and the cluster label is the first cluster label or the second cluster label.
  • the processor 1000 is further configured to construct a factor system; classify the discrete factors in the factor system to obtain discrete factors in different categories; cluster the discrete factors in each category through a preset model , Obtain the first cluster label corresponding to each category; establish a performance prediction model according to the first cluster label corresponding to each category and the continuous factors in the factor system.
  • the processor 1000 clusters the discrete factors of each category through a preset model to obtain the first cluster label corresponding to each category, specifically, clusters the discrete factors of each category through the preset model.
  • Category obtain at least one subcategory corresponding to each category; set a cluster label for each subcategory in the at least one subcategory, and obtain at least one cluster label corresponding to each category; At least one cluster label corresponding to the category serves as the first cluster label corresponding to each category.
  • the processor 1000 establishes a performance prediction model according to the first cluster label corresponding to each category and the continuous factors in the factor system, and specifically includes the first cluster label corresponding to each category And each continuous factor in the factor system is input to the first gradient boosting decision tree GBDT model to train the first GBDT model; the first GBDT model after training is used as the performance prediction model.
  • the processor 1000 is further configured to cluster the discrete factors under each category to obtain the first cluster label corresponding to each category, and then use the discrete factors under each category and The first cluster label corresponding to each category establishes a corresponding classification model for the discrete factor under each category for label prediction.
  • the processor 1000 uses the discrete factor under each category and the first cluster label corresponding to each category to establish a corresponding classification model for the discrete factor under each category, specifically as follows:
  • the discrete factor under each category and the first cluster label corresponding to each category are respectively input to the second GBDT model corresponding to each category for training; the second GBDT model corresponding to each category after training, As the classification model corresponding to the discrete factor under each category.
  • the processor 1000 is further configured to adjust the number of cluster types when the prediction accuracy of the performance prediction model does not meet a preset condition; according to the adjusted number of cluster types, pass the preset The model clusters the discrete factors under each category again to obtain the second cluster label corresponding to each category; according to the second cluster label corresponding to each category and the continuous factor in the factor system , To establish a performance prediction model whose prediction accuracy meets the preset conditions.
  • the processor 1000, the input device 2000, and the output device 3000 described in the embodiments of the present application can perform the implementation manners described in the embodiments of FIGS. 1 to 2 and can also perform the implementation manners described in the embodiments of the present application. , I won’t repeat it here.
  • the functional modules in the various embodiments of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module.
  • the above-mentioned integrated modules can be implemented in the form of sampling hardware, or in the form of sampling software functional modules.
  • the program can be stored in a computer non-volatile readable storage medium.
  • the computer non-volatile readable storage medium may be a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于预测模型的绩效预测方法、装置及存储介质，其中，该方法可以包括：对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子(S101)；利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签(S102)；将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果(S103)。该方法可以提高绩效预测精度。

Description

基于预测模型的绩效预测方法、装置及存储介质
本申请要求于2019年02月18日提交中国专利局、申请号为2019101232206、申请名称为“基于预测模型的绩效预测方法、装置及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及深度学习领域，尤其涉及一种基于预测模型的绩效预测方法、装置及存储介质。
背景技术
由于深度神经网络模型在机器学习领域表现出了较佳的准确度，包括计算机视觉、语音识别和机器人在内的诸多智能应用已广泛地使用了深度神经网络模型。目前，针对复杂事件的建模通常以深度神经网络模型为主，然而当将深度神经网络模型应用于绩效预测时，其绩效预测精度较低。
发明内容
本申请实施例提供了一种基于预测模型的绩效预测方法、装置及存储介质，可以提高绩效预测精度。
第一方面，本申请实施例提供了一种基于预测模型的绩效预测方法，包括：
对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；
利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签；
将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果；
其中，所述绩效预设模型是根据训练数据集训练得到的，所述训练数据集包括第一聚类标签和预设的因子体系中的连续因子，所述第一聚类标签是根据所述因子体系中的离散因子得到的。
第二方面，本申请实施例提供了一种基于预测模型的绩效预测装置，包括：
分类单元，用于对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；
第一预测单元，用于利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签；
第二预测单元，用于将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果；其中，所述绩效预设模型是根据训练数据集训练得到的；所述训练数据集包括聚类标签和预设的因子体系中的连续因子，所述聚类标签是根据所述因子体系中的离散因子得到的，所述聚类标签为第一聚类标签或第二聚类标签。
第三方面，本申请实施例提供了一种电子设备，包括处理器、输入设备、输出设备和存储器，所述处理器、输入设备、输出设备和存储器相互连接，其中，所述存储器用于存储计算机程序，所述计算机程序包括程序指令，所述处理器被配置用于调用所述程序指令，执行如第一方面所述的方法。
第四方面，本申请实施例提供了一种计算机非易失性可读存储介质，所述计算机非易失性可读存储介质存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被处理器执行时使所述处理器执行如第一方面所述的方法。
综上所述，电子设备可以对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子，并对该不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签，以结合该目标绩效预测信息中的连续因子进行绩效预测，得到绩效预测结果，提高了绩效预测的精度。
附图说明
图1是本申请实施例提供的一种基于预测模型的绩效预测方法的流程示意图；
图2是本申请实施例提供的另一种基于预测模型的绩效预测方法的流程示意图；
图3是本申请实施例提供的一种基于预测模型的绩效预测装置的结构示意图；
图4是本申请实施例提供的一种电子设备的结构示意图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行描述。
请参阅图1，为本申请实施例提供的一种基于预测模型的绩效预测方法的流程示意图。该方法可以应用于电子设备。该电子设备可以为终端或服务器。具体地，该方法可以包括：
S101、对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子。
该目标绩效预测信息是指用户用于绩效预测的信息，如用户的性格、工作等维度的信息，该目标绩效预测信息可以包括连续因子和离散因子。该连续因子用于表示连续型信息。在一个实施例中，该连续型信息是指特征值为数值的信息。该特征值可以理解为取值。如，该连续因子为学习时长(如1小时)、上班乘车时长(如1小时)。离散因子用于表示离散型信息。在一个实施例中，该离散型信息是指特征值不为数值的信息。如该离散因子为缺勤(如是或否)，工作态度(如积极或消极)。
在一个实施例中，电子设备可以接收分类指令，并根据接收到的分类指令对目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子。例如，可以根据接收到的分类指令将该目标绩效预测信息中的离散因子，分类为工作类别下的离散因子、性格类别下的离散因子、学习类别下的离散因子。
S102、利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签。
其中，每个类别下的离散因子可以对应一个分类模型。例如，工作类别下的离散因子对应第一分类模型，学习类别下的离散因子对应第二分类模型、学习类别下的离散因子对应第三分类模型。
该分类模型可以是根据第一聚类标签以及预设的因子体系中的离散因子 训练得到的。具体地，该分类模型可以是采用深度学习方法，根据第一聚类标签以及预设的因子体系中的离散因子训练得到的。该分类模型可以包括训练得到的标签预测策略。其中，该第一聚类标签可以是根据该因子体系中的离散因子得到的。在一个实施例中，该第一聚类标签可以是对该因子体系中的离散因子进行分类，得到不同类别下的离散因子，并对每个类别下的离散因子进行聚类后得到的。例如，该第一聚类标签可以是对该因子体系中的离散因子进行分类，得到不同类别下的离散因子，并通过预设模型对每个类别下的离散因子进行聚类后得到的。该预设模型包括但不限于k-model等可以用于聚类分析的模型。或，该第一聚类标签还可以是对该因子体系中的离散因子进行分类，得到不同类别下的离散因子，并通过聚类算法对每个类别下的离散因子进行聚类后，根据接收到的标签设置指令设置的。
或者，该分类模型还可以是根据第二聚类标签以及预设的因子体系中的离散因子训练得到的。具体地，该分类模型还可以是采用深度学习方法，根据第二聚类标签以及预设的因子体系中的离散因子训练得到的。该第二聚类标签可以是根据调整后的聚类的种类的数量，以及该因子体系中的离散因子得到的。在一个实施例中，该第二聚类标签可以是根据调整后的聚类的种类的数量，通过预设模型对每个类别下的离散因子再次进行聚类后得到的。当该预设模型为k-model时，该调整后的聚类的种类的数量为调整后的k值。或该第二聚类标签可以是根据调整后的聚类的种类的数量，采用聚类算法对每个类别下的离散因子再次进行聚类后，根据接收到的标签设置指令设置的。
本申请实施例中，电子设备可以利用该不同类别下的离散因子对应的分类模型，对该不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签。本申请实施例通过使用分类模型，还可以得到各离散因子的重要程度值，有利于后续分析各离散因子对绩效预测的贡献程度。
例如，电子设备通过第一分类模型对工作类别下的离散因子进行标签预测，得到工作类别下的离散因子对应的标签1；电子设备通过第二分类模型对性格类别下的离散因子进行标签预测，得到性格类别下的离散因子对应的标签2和标签3；电子设备通过第三分类模型对学习下的离散因子进行标签预测，得到学习类别下的离散因子对应的标签4、标签5。
具体地，电子设备利用该不同类别下的离散因子对应的分类模型，对该不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签，可以包括：电子设备将该不同类别下的离散因子分别输入对应的分类模型，以进行标签预测，并通过该对应的分类模型分别输出该不同类别下的离散因子的分类标签。
在一个实施例中，电子设备将该不同类别下的离散因子分别输入对应的分类模型，以进行标签预测，可以包括：将该不同类别下的离散因子的特征值分别输入对应的分类模型。
在一个实施例中，电子设备在将该不同类别下的离散因子的特征值分别输入对应的分类模型之前，还可以将该特征值表征为特征向量，该特征向量包括但不限于以one-hot等形式表示。在确定该特征值对应的特征向量后，上述将在将该不同类别下的离散因子的特征值分别输入对应的分类模型，可以为将该不同类别下的离散因子的特征值对应的特征向量分别输入对应的分类模型。
以离散因子为“缺勤”为例来说，上述确定特征值的特征向量的过程可以如下：如果缺勤为是，也即离散因子的特征值为“是”，则该特征值对应的特征向量可以为[0 1]。如果缺勤为否，也即离散因子的特征值为“否”，则该特征值对应的特征向量可以为[1 0]。
S103、将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果。
其中，该绩效预测模型可以包括训练出的绩效预测策略。绩效预设模型可以是根据训练数据集训练得到的。具体地，该绩效预测模型可以是采用深度学习方法，根据训练数据集训练得到的。该训练数据集可以包括聚类标签和预设的因子体系中的连续因子，该聚类标签是根据该因子体系中的离散因子得到的，该聚类标签为第一聚类标签或第二聚类标签。
本申请实施例中，电子设备可以将该每个类别对应的至少一个分类标签、该目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，并通过绩效预测模型输出该目标绩效预测信息对应的用户的绩效预测结果。该绩效预测结果可以包括预测出的绩效等级，如1、2、3，或高、中、低。或， 该绩效预测结果还可以包括处于相应绩效等级的概率，如90％。
在一个实施例中，电子设备将该每个类别对应的至少一个分类标签、该目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，可以包括：电子设备将该每个类别对应的至少一个分类标签、该目标绩效预测信息中的连续因子的特征值输入预设的绩效预测模型以进行绩效预测。
在一个实施例中，电子设备在获取该绩效预测结果后，可以输出该绩效预测结果，以展示给相关人员。
可见，图1所示的实施例中，电子设备可以对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子，并利用该不同类别下的离散因子对应的分类模型，对该不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签，以将该每个类别对应的至少一个分类标签、该目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得该目标绩效预测信息对应的用户的绩效预测结果，提高了绩效预测的精度。
请参阅图2，为本申请实施例提供的一种基于预测模型的绩效预测方法的流程示意图。该方法可以应用于电子设备，该电子设备可以为终端或服务器。具体地，该方法可以包括：
S201、构建因子体系。
本申请实施例中，电子设备可以获取至少一个用户的绩效预测信息，以生成包括该至少一个用户的绩效预测信息的因子体系。由于该至少一个用户的绩效预测信息可以包括离散因子和连续因子，因此该因子体系可以包括离散因子和连续因子。
在一个实施例中，若电子设备自身存储了该至少一个用户的绩效预测信息，则电子设备可以获取存储的该至少一个用户的绩效预测信息。若电子设备自身未存储该至少一个用户的绩效预测信息，则电子设备可以从其他设备获取该至少一个用户的绩效预测信息。本申请实施例对该至少一个用户的绩效预测信息的获取方式不做限制。
S202、对所述因子体系中的离散因子进行分类，得到不同类别下的离散因子。
本申请实施例中，电子设备可以对因子体系中的离散因子进行分类，得到不同类别下的离散因子。
例如，因子体系包括离散因子x1、x2、x3、x4、x5、y1、y2、y3、y4、y5，电子设备可以x1、x2、x3、x4、x5分到性格类别下，将y1、y2、y3、y4、y5分到工作类别下。
在一个实施例中，电子设备对因子体系中的离散因子进行分类，得到不同类别下的离散因子电子设备，可以包括：电子设备接收分类指令，并根据接收到的分类指令对因子体系中的离散因子进行分类，得到不同类别下的离散因子。例如，可以根据接收到的分类指令将因子体系中的离散因子，分类为工作类别下的离散因子、性格类别下的离散因子、学习类别下的离散因子。
在一个实施例中，电子设备对因子体系中的离散因子进行分类，得到不同类别下的离散因子电子设备，还可以包括：电子根据预设的离散因子与类别的对应关系，对不同离散因子进行类别划分，得到不同类别的离散因子。根据对应关系划分，使得分类过程更加自动化。
S203、通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签。
本申请实施例，电子设备可以通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的聚类标签。其中，该预设模型包括但不限于K-model等可以用于聚类分析的模型。K-model可以采用聚类算法，如K-means聚类算法等算法对每个类别下的离散因子进行聚类。本申请实施例通过聚类可以将每个类别下相似的离散因子分类到同一类别。
在一个实施例中，电子设备通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签，包括：电子设备通过预设模型对每个类别的离散因子进行聚类，得到每个类别对应划分出的至少一个子类别；电子设备为所述至少一个子类别中每个子类别设置聚类标签，得到所述每个类别对应的至少一个聚类标签；电子设备将所述每个子类别对应的至少一个聚类标签，作为所述每个类别对应的第一聚类标签。
例如，性格类别下的离散因子包括x1、x2、x3、x4、x5，通过预设模型将五个离散因子进行聚类后，x1、x2被划分至同一类别，如第一类别，x3、 x4、x5被划分至同一类别，如第二类别。此时性格类别对应的至少一个子类别包括第一类别和第二类别，对第一类别设置标签1，对第二类别设置标签2。性格类别对应的第一聚类标签为标签1和标签2。同样的，工作类别下的离散因子包括y1、y2、y3、y4、y5，通过预设模型对工作下的各个离散因子进行聚类后，y1、y2、y3被划分至同一类别，如第三类别，y4、y5被划分至同一类别，如第四类别。此时工作类别对应的至少一个子类别对第三类别和第四类别，对第三类别设置标签3，对第四类别设置标签4。工作类别对应的第一聚类标签为标签3和标签4。
S204、根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型。
本申请实施例，电子设备可以根据每个类别对应的第一聚类标签以及因子体系中的连续因子，建立绩效预测模型。在一个实施例中，该绩效预测模型的建立过程可以基于SK-Learn库实现。
在一个实施例中，电子设备根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型，可以包括：电子设备将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子输入第一梯度提升决策树(Gradient Boosting Decision Tree，GBDT)模型，以对所述第一GBDT模型进行训练；电子设备将训练后的第一GBDT模型作为绩效预测模型。
在一个实施例中，电子设备将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子输入第一GBDT模型，可以包括：电子设备将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子的特征值输入第一GBDT模型，
为了使绩效预测模型可以达到较高的精度，因此，本申请实施例可以通过调整聚类的种类的数量来建立预测精度符合预设条件的绩效预测模型。具体地，电子设备根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型，可以包括：当所述绩效预测模型的预测精度不符合预设条件时，电子设备调整聚类的种类的数量；电子设备根据调整的聚类的种类的数量，通过所述预设模型对每个类别下的离散因子再次进行聚类，得到所述每个类别对应的第二聚类标签；电子设备根据所述每个类别对应的第二聚类 标签以及所述因子体系中的连续因子，建立预测精度符合预设条件的绩效预测模型。当该预设模型为K-model时，该调整后的聚类的种类的数量为调整后的k值。
S205、利用所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，以用于标签预测。
具体地，电子设备利用所述每个类别下的离散因子以及每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，可以包括：电子设备将所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，分别输入到每个类别对应的第二GBDT模型进行训练；电子设备二将训练后的每个类别对应的第二GBDT模型，作为所述每个类别下的离散因子对应的分类模型。其中，该第二GBDT模型与第一GBDT模型可以相同或不同，可根据实际情况进行设置。在一个实施例中，该分类模型的建立过程可以基于SK-Learn库实现。
在一个实施例中，电子设备将所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，分别输入到每个类别对应的第二GBDT模型进行训练，可以包括：电子设备将每个类别下的离散因子的特征值以及所述每个类别对应的聚类标签输入第二GBDT模型进行训练，将训练后的第二GBDT模型作为分类模型。或，当确定特征值对应的特征向量时，还可以将将每个类别下的离散因子的特征值对应的特征向量以及所述每个类别对应的聚类标签输入第二GBDT模型进行训练。
S206、对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；
S207、利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签；
S208、将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果。
其中，步骤S206-S208可以参见图1实施例中的步骤S101-S103，本申请实施例在此不做赘述。
可见，图2所示的实施例中，电子设备可以对构建的因子体系中的离散因子，通过分类得到不同类别下的离散因子，并对不同类别下的离散因子进行聚类得到每个类别对应的第一聚类标签，从而根据第一聚类标签以及因子体系中的连续因子建立绩效预测模型；并且，电子设备还可以根据每个类别下的离散因子以及每个类别对应的第一聚类标签，为每个类别下的离散因子建立对应的分类模型，从而实现绩效预测模型和分类模型的构建过程，通过此类方式建立的模型具备较好的可解释性。电子设备可以对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子，并利用该不同类别下的离散因子对应的分类模型，对该不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签，以将该每个类别对应的至少一个分类标签、该目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得该目标绩效预测信息对应的用户的绩效预测结果，提高了绩效预测的精度。
请参阅图3，为本申请实施例提供的一种基于预测模型的绩效预测装置的结构示意图。该装置可以应用于电子设备。该电子设备可以为终端或服务器。具体地，该装置可以包括：
分类单元31，用于对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；
第一预测单元32，用于利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签；
第二预测单元33，用于将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果；其中，所述绩效预设模型是根据训练数据集训练得到的；所述训练数据集包括聚类标签和预设的因子体系中的连续因子，所述聚类标签是根据所述因子体系中的离散因子得到的，所述聚类标签为第一聚类标签或第二聚类标签。
在一种可选的实施方式中，处理单元34，用于构建因子体系；对所述因子体系中的离散因子进行分类，得到不同类别下的离散因子；通过预设模型对 每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签；根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型。
在一种可选的实施方式中，处理单元34通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签，具体为通过预设模型对每个类别的离散因子进行聚类，得到每个类别对应划分出的至少一个子类别；为所述至少一个子类别中每个子类别设置聚类标签，得到所述每个类别对应的至少一个聚类标签；将所述每个子类别对应的至少一个聚类标签，作为所述每个类别对应的第一聚类标签。
在一种可选的实施方式中，处理单元34根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型，具体为将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子输入第一梯度提升决策树GBDT模型，以对所述第一GBDT模型进行训练；将训练后的第一GBDT模型作为绩效预测模型。
在一种可选的实施方式中，处理单元34，还用于在对所述每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签后，利用所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，以用于标签预测。
在一种可选的实施方式中，处理单元34利用所述每个类别下的离散因子以及每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，具体为将所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，分别输入到每个类别对应的第二GBDT模型进行训练；将训练后的每个类别对应的第二GBDT模型，作为所述每个类别下的离散因子对应的分类模型。
在一种可选的实施方式中，处理单元34，还用于当所述绩效预测模型的预测精度不符合预设条件时，调整聚类的种类的数量；根据调整的聚类的种类的数量，通过所述预设模型对每个类别下的离散因子再次进行聚类，得到所述每个类别对应的第二聚类标签；根据所述每个类别对应的第二聚类标签以及所述因子体系中的连续因子，建立预测精度符合预设条件的绩效预测模型。
可见，图3所示的实施例中，电子设备可以对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子，并利用该不同类别下的离散因子对应的分类模型，对该不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签，以将该每个类别对应的至少一个分类标签、该目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得该目标绩效预测信息对应的用户的绩效预测结果，提高了绩效预测的精度。
请参阅图4，为本申请实施例提供的一种电子设备的结构示意图。其中，本实施例中所描述的电子设备可以包括：一个或多个处理器1000，一个或多个输入设备2000，一个或多个输出设备3000和存储器4000。处理器1000、输入设备2000、输出设备3000和存储器4000可以通过总线连接。
输入设备2000、输出设备3000可以是标准的有线或无线通信接口。
处理器1000可以是中央处理模块(Central Processing Unit，CPU)，该处理器还可以是其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现成可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
存储器4000可以是高速RAM存储器，也可为非不稳定的存储器(non-volatile memory)，例如磁盘存储器。存储器4000用于存储一组程序代码，输入设备2000、输出设备3000和处理器1000可以调用存储器4000中存储的程序代码。具体地：
处理器1000，用于对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签；将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果；其中，所述绩效预设模型是根据训练数据集训练得到的；所述训练数据集包括聚类标签和预设的因子体系中的连续因子， 所述聚类标签是根据所述因子体系中的离散因子得到的，所述聚类标签为第一聚类标签或第二聚类标签。
可选地，处理器1000，还用于构建因子体系；对所述因子体系中的离散因子进行分类，得到不同类别下的离散因子；通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签；根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型。
可选地，处理器1000通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签，具体为通过预设模型对每个类别的离散因子进行聚类，得到每个类别对应划分出的至少一个子类别；为所述至少一个子类别中每个子类别设置聚类标签，得到所述每个类别对应的至少一个聚类标签；将所述每个子类别对应的至少一个聚类标签，作为所述每个类别对应的第一聚类标签。
可选地，处理器1000根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型，具体为将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子输入第一梯度提升决策树GBDT模型，以对所述第一GBDT模型进行训练；将训练后的第一GBDT模型作为绩效预测模型。
可选地，处理器1000，还用于在对所述每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签后，利用所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，以用于标签预测。
可选地，处理器1000利用所述每个类别下的离散因子以及每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，具体为将所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，分别输入到每个类别对应的第二GBDT模型进行训练；将训练后的每个类别对应的第二GBDT模型，作为所述每个类别下的离散因子对应的分类模型。
可选地，处理器1000，还用于当所述绩效预测模型的预测精度不符合预设条件时，调整聚类的种类的数量；根据调整的聚类的种类的数量，通过所述预设模型对每个类别下的离散因子再次进行聚类，得到所述每个类别对应的第 二聚类标签；根据所述每个类别对应的第二聚类标签以及所述因子体系中的连续因子，建立预测精度符合预设条件的绩效预测模型。
具体实现中，本申请实施例中所描述的处理器1000、输入设备2000、输出设备3000可执行图1-图2实施例所描述的实现方式，也可执行本申请实施例所描述的实现方式，在此不再赘述。
在本申请各个实施例中的各功能模块可以集成在一个处理模块中，也可以是各个模块单独物理存在，也可以是两个或两个以上模块集成在一个模块中。上述集成的模块既可以采样硬件的形式实现，也可以采样软件功能模块的形式实现。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机非易失性可读存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的计算机非易失性可读存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)或随机存储记忆体(Random Access Memory，RAM)等。
以上所揭露的仅为本申请一种较佳实施例而已，当然不能以此来限定本申请之权利范围，本领域普通技术人员可以理解实现上述实施例的全部或部分流程，并依本申请权利要求所作的等同变化，仍属于本申请所涵盖的范围。

Claims (20)

  1. 一种基于预测模型的绩效预测方法，其特征在于，包括：
    对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；
    利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签；
    将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果；
    其中，所述绩效预设模型是根据训练数据集训练得到的；所述训练数据集包括聚类标签和预设的因子体系中的连续因子，所述聚类标签是根据所述因子体系中的离散因子得到的，所述聚类标签为第一聚类标签或第二聚类标签。
  2. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    构建因子体系；
    对所述因子体系中的离散因子进行分类，得到不同类别下的离散因子；
    通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签；
    根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型。
  3. 根据权利要求2所述的方法，其特征在于，所述通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签，包括：
    通过预设模型对每个类别的离散因子进行聚类，得到每个类别对应划分出的至少一个子类别；
    为所述至少一个子类别中每个子类别设置聚类标签，得到所述每个类别对应的至少一个聚类标签；
    将所述每个子类别对应的至少一个聚类标签，作为所述每个类别对应的第一聚类标签。
  4. 根据权利要求2所述的方法，其特征在于，所述根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型，包括：
    将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子输入第一梯度提升决策树GBDT模型，以对所述第一GBDT模型进行训练；
    将训练后的第一GBDT模型作为绩效预测模型。
  5. 根据权利要求2-4任意一项所述的方法，其特征在于，所述对所述每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签后，所述方法还包括：
    利用所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，以用于标签预测。
  6. 根据权利要求5所述的方法，其特征在于，所述利用所述每个类别下的离散因子以及每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，包括：
    将所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，分别输入到每个类别对应的第二GBDT模型进行训练；
    将训练后的每个类别对应的第二GBDT模型，作为所述每个类别下的离散因子对应的分类模型。
  7. 根据权利要求2所述的方法，其特征在于，所述方法还包括：
    当所述绩效预测模型的预测精度不符合预设条件时，调整聚类的种类的数量；
    根据调整的聚类的种类的数量，通过所述预设模型对每个类别下的离散因子再次进行聚类，得到所述每个类别对应的第二聚类标签；
    根据所述每个类别对应的第二聚类标签以及所述因子体系中的连续因子，建立预测精度符合预设条件的绩效预测模型。
  8. 一种基于预测模型的绩效预测装置，其特征在于，包括：
    分类单元，用于对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；
    第一预测单元，用于利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类 标签；
    第二预测单元，用于将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果；其中，所述绩效预设模型是根据训练数据集训练得到的；所述训练数据集包括聚类标签和预设的因子体系中的连续因子，所述聚类标签是根据所述因子体系中的离散因子得到的，所述聚类标签为第一聚类标签或第二聚类标签。
  9. 根据权利要求8所述的装置，其特征在于，所述装置还包括处理单元，
    所述处理单元，用于构建因子体系；对所述因子体系中的离散因子进行分类，得到不同类别下的离散因子；通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签；根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型。
  10. 根据权利要求9所述的装置，其特征在于，所述处理单元通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签，具体为通过预设模型对每个类别的离散因子进行聚类，得到每个类别对应划分出的至少一个子类别；为所述至少一个子类别中每个子类别设置聚类标签，得到所述每个类别对应的至少一个聚类标签；将所述每个子类别对应的至少一个聚类标签，作为所述每个类别对应的第一聚类标签。
  11. 根据权利要求9所述的装置，其特征在于，所述处理单元根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型，具体为将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子输入第一梯度提升决策树GBDT模型，以对所述第一GBDT模型进行训练；将训练后的第一GBDT模型作为绩效预测模型。
  12. 根据权利要求9-11任意一项所述的装置，其特征在于，所述处理单元，还用于在对所述每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签后，利用所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，以用于标签预测。
  13. 根据权利要求12所述的装置，其特征在于，所述处理单元利用所述 每个类别下的离散因子以及每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，具体为将所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，分别输入到每个类别对应的第二GBDT模型进行训练；将训练后的每个类别对应的第二GBDT模型，作为所述每个类别下的离散因子对应的分类模型。
  14. 根据权利要求9所述的装置，其特征在于，所述处理单元，还用于当所述绩效预测模型的预测精度不符合预设条件时，调整聚类的种类的数量；根据调整的聚类的种类的数量，通过所述预设模型对每个类别下的离散因子再次进行聚类，得到所述每个类别对应的第二聚类标签；根据所述每个类别对应的第二聚类标签以及所述因子体系中的连续因子，建立预测精度符合预设条件的绩效预测模型。
  15. 一种电子设备，其特征在于，包括处理器、输入设备、输出设备和存储器，所述处理器、输入设备、输出设备和存储器相互连接，其中，所述存储器用于存储计算机程序，所述计算机程序包括程序指令，所述处理器被配置用于调用所述程序指令，执行：
    对待预测的目标绩效预测信息中的离散因子进行分类，得到不同类别下的离散因子；
    利用所述不同类别下的离散因子对应的分类模型，对所述不同类别下的离散因子进行标签预测，得到每个类别对应的至少一个分类标签；
    将所述每个类别对应的至少一个分类标签、所述目标绩效预测信息中的连续因子输入预设的绩效预测模型以进行绩效预测，获得所述目标绩效预测信息对应的用户的绩效预测结果；
    其中，所述绩效预设模型是根据训练数据集训练得到的；所述训练数据集包括聚类标签和预设的因子体系中的连续因子，所述聚类标签是根据所述因子体系中的离散因子得到的，所述聚类标签为第一聚类标签或第二聚类标签。
  16. 根据权利要求15所述的电子设备，其特征在于，所述处理器，还用于构建因子体系；对所述因子体系中的离散因子进行分类，得到不同类别下的离散因子；通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签；根据所述每个类别对应的第一聚类标签以及所述因子体系 中的连续因子，建立绩效预测模型。
  17. 根据权利要求16所述的电子设备，其特征在于，所述处理器通过预设模型对每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签，具体为通过预设模型对每个类别的离散因子进行聚类，得到每个类别对应划分出的至少一个子类别；为所述至少一个子类别中每个子类别设置聚类标签，得到所述每个类别对应的至少一个聚类标签；将所述每个子类别对应的至少一个聚类标签，作为所述每个类别对应的第一聚类标签。
  18. 根据权利要求16所述的电子设备，其特征在于，所述处理器根据所述每个类别对应的第一聚类标签以及所述因子体系中的连续因子，建立绩效预测模型，具体为将所述每个类别对应的第一聚类标签以及所述因子体系中的各连续因子输入第一梯度提升决策树GBDT模型，以对所述第一GBDT模型进行训练；将训练后的第一GBDT模型作为绩效预测模型。
  19. 根据权利要求16-18任意一项所述的电子设备，其特征在于，所述处理器在对所述每个类别下的离散因子进行聚类，得到每个类别对应的第一聚类标签后，利用所述每个类别下的离散因子以及所述每个类别对应的第一聚类标签，为所述每个类别下的离散因子建立对应的分类模型，以用于标签预测。
  20. 一种计算机非易失性可读存储介质，其特征在于，所述计算机非易失性可读存储介质存储有计算机程序，所述计算机程序包括程序指令，所述程序指令当被处理器执行时使所述处理器执行如权利要求1-7任一项所述的方法。
PCT/CN2019/117787 2019-02-18 2019-11-13 基于预测模型的绩效预测方法、装置及存储介质 WO2020168754A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910123220.6 2019-02-18
CN201910123220.6A CN109934389A (zh) 2019-02-18 2019-02-18 基于预测模型的绩效预测方法、装置及存储介质

Publications (1)

Publication Number Publication Date
WO2020168754A1 true WO2020168754A1 (zh) 2020-08-27

Family

ID=66985780

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/117787 WO2020168754A1 (zh) 2019-02-18 2019-11-13 基于预测模型的绩效预测方法、装置及存储介质

Country Status (2)

Country Link
CN (1) CN109934389A (zh)
WO (1) WO2020168754A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109934389A (zh) * 2019-02-18 2019-06-25 平安科技（深圳）有限公司 基于预测模型的绩效预测方法、装置及存储介质
JP7310459B2 (ja) * 2019-09-03 2023-07-19 トヨタ自動車株式会社 情報処理方法および情報処理装置
CN112418822B (zh) * 2020-12-10 2021-09-14 浙江辉驿网络科技有限公司 一种线下门店的人资管理辅助系统
CN116504665A (zh) * 2023-06-28 2023-07-28 全芯智造技术有限公司 Cmp工艺预测、预测模型训练方法及装置、计算设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170286569A1 (en) * 2016-03-29 2017-10-05 Xerox Corporation System and method of modeling irregularly sampled temporal data using kalman filters
CN109002900A (zh) * 2018-06-05 2018-12-14 平安科技（深圳）有限公司 预测投保风险概率的方法、装置、计算机设备和存储介质
CN109034880A (zh) * 2018-07-09 2018-12-18 麒麟合盛网络技术股份有限公司 收入预测方法及装置
CN109934389A (zh) * 2019-02-18 2019-06-25 平安科技（深圳）有限公司 基于预测模型的绩效预测方法、装置及存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170286569A1 (en) * 2016-03-29 2017-10-05 Xerox Corporation System and method of modeling irregularly sampled temporal data using kalman filters
CN109002900A (zh) * 2018-06-05 2018-12-14 平安科技（深圳）有限公司 预测投保风险概率的方法、装置、计算机设备和存储介质
CN109034880A (zh) * 2018-07-09 2018-12-18 麒麟合盛网络技术股份有限公司 收入预测方法及装置
CN109934389A (zh) * 2019-02-18 2019-06-25 平安科技（深圳）有限公司 基于预测模型的绩效预测方法、装置及存储介质

Also Published As

Publication number Publication date
CN109934389A (zh) 2019-06-25

Similar Documents

Publication Publication Date Title
WO2020168754A1 (zh) 基于预测模型的绩效预测方法、装置及存储介质
US10262059B2 (en) Method, apparatus, and storage medium for text information processing
CN106897746B (zh) 数据分类模型训练方法和装置
WO2022213465A1 (zh) 基于神经网络的图像识别方法、装置、电子设备及介质
KR20210110823A (ko) 이미지 인식 방법, 인식 모델의 트레이닝 방법 및 관련 장치, 기기
JP7266674B2 (ja) 画像分類モデルの訓練方法、画像処理方法及び装置
WO2020244336A1 (zh) 一种告警分类方法、装置、电子设备及存储介质
WO2020186887A1 (zh) 一种连续小样本图像的目标检测方法、装置及设备
CN110889446A (zh) 人脸图像识别模型训练及人脸图像识别方法和装置
WO2020114108A1 (zh) 聚类结果的解释方法和装置
CN110364185B (zh) 一种基于语音数据的情绪识别方法、终端设备及介质
US10210424B2 (en) Method and system for preprocessing images
EP3620982B1 (en) Sample processing method and device
US10769808B2 (en) Apparatus and methods of automated tracking and counting of objects on a resource-constrained device
WO2023272852A1 (zh) 通过决策树模型对用户进行分类的方法、装置、设备和存储介质
CN113313053B (zh) 图像处理方法、装置、设备、介质及程序产品
CN113222942A (zh) 多标签分类模型的训练方法和预测标签的方法
WO2021008037A1 (zh) 基于A-BiLSTM神经网络的文本分类方法、存储介质及计算机设备
CN113657483A (zh) 模型训练方法、目标检测方法、装置、设备以及存储介质
EP4343616A1 (en) Image classification method, model training method, device, storage medium, and computer program
WO2019095587A1 (zh) 人脸识别方法、应用服务器及计算机可读存储介质
CN112668482A (zh) 人脸识别训练方法、装置、计算机设备及存储介质
WO2020143303A1 (zh) 深度学习模型训练方法、装置、计算机设备及存储介质
WO2021174814A1 (zh) 众包任务的答案验证方法、装置、计算机设备及存储介质
WO2022194049A1 (zh) 一种对象处理方法及装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19916175

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19916175

Country of ref document: EP

Kind code of ref document: A1