Resource matching model training method, resource matching method, and electronic device

Abstract

Provided in the embodiments of the present disclosure are a resource matching model training method, a resource matching method, and an electronic device. The resource matching model training method comprises: executing a training operation on a machine learning model set until a preset ending condition is met; and determining a resource matching model according to the machine learning model set which is obtained when the preset ending condition is met. Executing a training operation on a machine learning model set until a preset ending condition is met comprises: using a training object attribute data set corresponding to the machine learning model set to perform at least one round of training on the machine learning model set, so as to obtain an intermediate machine learning model set corresponding to the machine learning model set; determining a model evaluation information set corresponding to the intermediate machine learning model set; and according to the model evaluation information set, determining a plurality of target intermediate machine learning models from the intermediate machine learning model set, and using same as a machine learning model set for the next round of training.

Description

Resource matching model training method, resource matching method and electronic equipment Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a resource matching model training method, a resource matching method, and an electronic device. Background Art

[0002] With the development of internet and artificial intelligence technologies, the application of resource matching models has gradually become more widespread. Resource matching models can determine the degree of matching between a first resource and a second resource based on the attribute information of the first resource and the attribute information of the second resource, such as the degree of matching between human resources and organizational resources.

[0003] Since the resource matching model has a low matching accuracy, it cannot accurately reflect the actual matching degree between the first object resource and the second comparison resource, thereby reducing the resource matching accuracy.

[0004] Summary of the Invention

[0005] According to a first aspect, the present disclosure provides a training method for a resource matching model, comprising: performing a training operation on a machine learning model set until a preset end condition is met; and determining a resource matching model based on the machine learning model set obtained when the preset end condition is met; wherein, performing a training operation on the machine learning model set until the preset end condition is met comprises: using a training object attribute data set corresponding to the machine learning model set to perform at least one round of training on the machine learning model set to obtain an intermediate machine learning model set corresponding to the machine learning model set; determining a model evaluation information set corresponding to the intermediate machine learning model set; based on the model evaluation information set, determining multiple target intermediate machine learning models from the intermediate machine learning model set as the machine learning model set for the next round of training; wherein, the training object attribute data set is a data set of a target application scenario, and the target application scenario characterizes matching a first object resource with a second object resource; wherein, the machine learning model set includes multiple machine learning models, the training object attribute data set includes multiple first training object attribute data subsets, and the machine learning model has a corresponding first training object attribute data subset.

[0006] For example, the model evaluation information set includes a model evaluation value corresponding to each of at least one intermediate machine learning models; wherein, based on the model evaluation information set, multiple target intermediate machine learning models are determined from the intermediate machine learning model set to obtain a machine learning model set for the next round of training, including: sorting the intermediate machine learning model set according to the model evaluation value corresponding to each of the at least one intermediate machine learning models to obtain an intermediate machine learning model sorting result; selecting a preset number of intermediate machine learning models from the intermediate machine learning model sorting results to obtain a target intermediate machine learning model; and obtaining a machine learning model set for the next round of training based on the target intermediate machine learning model.

[0007] For example, the above method also includes: dividing the training object attribute data set into multiple second training object attribute data subsets corresponding to multiple machine learning models; and determining the first training object attribute data subset corresponding to the machine learning model from each second training object attribute data subset.

[0008] For example, determining a first training object attribute data subset corresponding to a machine learning model from each second training object attribute data subset includes: randomly dividing the second training object attribute data subset into K second training object attribute data groups, where K is an integer greater than 1; randomly determining K-1 first training object attribute data groups from multiple second training object attribute data groups; and obtaining a first training object attribute data subset based on the K-1 first training object attribute data groups.

[0009] For example, dividing the training object attribute data set into second training object attribute data subsets corresponding to each of the multiple machine learning models includes one of the following: dividing the training object attribute data set into multiple second training object attribute data subsets in which object attribute data are non-repeated between subsets; dividing the training object data set into multiple second training object attribute data subsets in which object attribute data are repeated between subsets.

[0010] For example, the above method also includes: determining an initial training object attribute data set from the object attribute data according to the ratio between the amount of training object attribute data and the total amount of object attribute data; and processing the initial training object attribute data set to obtain a training object attribute data set.

[0011] For example, processing the initial training object attribute dataset to obtain the training object attribute dataset includes: discretizing the initial training object attribute dataset to obtain a first intermediate training object attribute dataset; continuating the first intermediate training object attribute dataset to obtain a second intermediate training object attribute dataset; and obtaining the training object attribute dataset based on the second intermediate training object attribute dataset.

[0012] For example, discretizing the initial training object attribute dataset to obtain a first intermediate training object attribute dataset includes: binning the initial training object attribute dataset to obtain a first intermediate training object attribute dataset; wherein the first intermediate training object attribute dataset includes a plurality of first intermediate training object attribute dimension data subsets corresponding to each of the plurality of candidate dimensions, and the first intermediate training object attribute dimension data subsets include a plurality of first intermediate training object attribute dimension data.

[0013] For example, performing serial processing on a first intermediate training object attribute dataset to obtain a second intermediate training object attribute dataset includes: determining, for each candidate dimension of at least one candidate dimension and for each first intermediate training object attribute dimension data subset in a plurality of first intermediate training object attribute dimension data subsets corresponding to the candidate dimension, a first number and a second number corresponding to the first intermediate training object attribute dimension data subset, wherein the first number represents the number of first-category data included in the first intermediate training object attribute dimension data subset, and the second number represents the number of second-category data included in the first intermediate training object attribute dimension data subset, the first-category data representing data whose label corresponding to the first intermediate training object attribute dimension data is a first label, and the second-category data representing data whose label corresponding to the first intermediate training object attribute dimension data is a second label; determining, based on a third number, a fourth number, the first number, and the second number, second intermediate training object attribute data corresponding to the first intermediate training object attribute dimension data subset, wherein the third number represents the number of first-category data included in the first intermediate training object attribute dimension data set, and the fourth number represents the number of second-category data included in the first intermediate training object attribute dimension data set; and obtaining the second intermediate training object attribute dataset based on the plurality of second intermediate training object attribute data corresponding to each of the plurality of candidate dimensions.

[0014] For example, the second intermediate training object attribute data set includes multiple second intermediate training object attribute dimension data subsets corresponding to multiple candidate dimensions, and the second intermediate training object attribute dimension data subset includes multiple second intermediate training object attribute dimension data; wherein, according to the second intermediate training object attribute data set, the training object attribute data set is obtained, including: according to the multiple second intermediate training object attribute dimension data corresponding to the multiple candidate dimensions, determining the importance corresponding to each of the multiple candidate dimensions; according to the importance corresponding to each of the multiple candidate dimensions, determining multiple target dimensions from the multiple candidate dimensions; and obtaining the training object attribute data set according to the second intermediate training object attribute dimension data corresponding to each of the multiple target dimensions of the multiple training objects.

[0015] For example, multiple machine learning models have their own corresponding hyperparameter sets, and at least two of the multiple hyperparameter sets are different from each other.

[0016] For example, the multiple machine learning models include at least two model structures that are different from each other, and the machine learning model is a logistic regression model.

[0017] For example, determining the model evaluation information set corresponding to the intermediate machine learning model set includes: determining the test object attribute data set from the object attribute data set according to the ratio between the number of test object attribute data and the total number of object attribute data; and using the test object attribute data set to evaluate the model effect of the intermediate machine learning model set to obtain the model evaluation information set corresponding to the intermediate machine learning model.

[0018] For example, the preset end condition includes one of the following: the number of training rounds reaches a preset maximum number of rounds, and the loss function constructed according to the model training results converges.

[0019] For example, when there is one target intermediate machine learning model, the target intermediate machine learning model is cyclically trained using the first training object attribute data subset corresponding to the target intermediate machine learning model until the target intermediate machine learning model reaches the preset end condition.

[0020] According to the second aspect, the present disclosure also provides a resource matching method, including: obtaining target object attribute data; and inputting the target object attribute data into a resource matching model to obtain target matching information; wherein the resource matching model is trained using the above-mentioned resource matching model training method.

[0021] For example, the resource matching model includes at least one target machine learning model; wherein, inputting the target object attribute data into the resource matching model to obtain target matching information includes: inputting the target object attribute data into at least one target machine learning model to obtain candidate matching information corresponding to each of the at least one target machine learning model; and obtaining target matching information based on at least one candidate matching information.

[0022] According to the third aspect, the present disclosure also provides an electronic device, including: one or more processors; a storage device for storing one or more programs, wherein, when the one or more programs are executed by one or more processors, the one or more processors execute the above-mentioned resource matching model training method and resource matching method.

[0023] According to a fourth aspect, the present disclosure further provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the above-mentioned resource matching model training method and resource matching method.

[0024] According to the fifth aspect, the present disclosure also provides a computer program product, including a computer program, which implements the above-mentioned resource matching model training method and resource matching method when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The above and other features of the embodiments of the present disclosure will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which:

[0026] FIG1 is a system architecture diagram of a resource matching model training method and a resource matching method according to an embodiment of the present disclosure;

[0027] FIG2 is a flow chart of a method for training a resource matching model according to an embodiment of the present disclosure;

[0028] FIG3 is a schematic diagram of a training process of a resource matching model according to another embodiment of the present disclosure;

[0029] FIG4 is a flowchart of dynamic optimization during resource matching model training according to an embodiment of the present disclosure;

[0030] FIG5 is a schematic diagram of an output result of a resource matching model according to an embodiment of the present disclosure;

[0031] FIG6A is an evaluation result obtained by evaluating a resource matching model according to an embodiment of the present disclosure;

[0032] FIG6B is an accuracy level standard of a resource matching model according to an embodiment of the present disclosure;

[0033] FIG6C is a diagram illustrating a discrimination level standard of a resource matching model according to an embodiment of the present disclosure;

[0034] FIG7 is a flow chart of a resource matching method according to an embodiment of the present disclosure;

[0035] FIG8 is a block diagram of an electronic device suitable for implementing a resource matching model training method and a resource matching method according to an embodiment of the present disclosure.

[0036] In the drawings, the same or similar structures are marked with the same or similar reference numerals. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the described embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present disclosure. It should be noted that throughout the drawings, the same elements are represented by the same or similar figure marks. In the following description, some specific embodiments are only for descriptive purposes and should not be understood as any limitation to the present disclosure, but are only examples of the embodiments of the present disclosure. Conventional structures or configurations will be omitted when they may cause confusion in the understanding of the present disclosure. It should be noted that the shapes and sizes of the components in the figures do not reflect the actual size and proportion, but only illustrate the contents of the embodiments of the present disclosure.

[0038] It should be noted that in the drawings, the sizes and relative sizes of elements may be exaggerated for clarity and / or descriptive purposes. Thus, the sizes and relative sizes of the individual elements are not necessarily limited to those shown in the drawings. In the specification and drawings, the same or similar reference numerals indicate the same or similar parts.

[0039] Unless otherwise defined, the technical or scientific terms used in the embodiments of the present disclosure should have the ordinary meanings understood by those of ordinary skill in the art. The terms "first," "second," and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are simply used to distinguish different components. The terms "include" or "comprising" and similar terms mean that the elements or objects preceding the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0040] Unless otherwise specified, directional terms such as "upper," "lower," "left," "right," "inner," and "outer" are used herein to indicate positions or relationships based on the figures shown. These terms are intended solely to facilitate the description of the present disclosure and are not intended to indicate or imply that the devices, components, or parts referred to must have, be constructed, or operate in a specific orientation. It should be understood that when the absolute positions of the objects being described change, the relative positions they represent may also change accordingly. Therefore, these directional terms should not be construed as limiting the present disclosure.

[0041] In addition, in the description of the embodiments of the present disclosure, the term "connected" or "connected to" may refer to a direct connection between two components or a connection between two components via one or more other components. In addition, the two components may be connected or coupled via a wired or wireless manner.

[0042] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure and application of the data involved (including but not limited to user personal information) comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0043] FIG1 is a system architecture diagram of a resource matching model training method and a resource matching method according to an embodiment of the present disclosure.

[0044] It should be noted that FIG1 is only an example of a system architecture to which the embodiments of the present disclosure can be applied, to help those skilled in the art understand the technical content of the present disclosure, but does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios. For example, in another embodiment, the system architecture to which the training method and resource matching method of the resource matching model can be applied may include a terminal device, but the terminal device can implement the training method and resource matching method of the resource matching model provided in the embodiments of the present disclosure without interacting with the server.

[0045] As shown in FIG1 , the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used as a medium for providing communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links.

[0046] A user can use at least one of the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103. For example, at least one of a resource matching application, an application with a resource matching model, a web browser application, a search application, an instant messaging tool, an email client, and a social platform software.

[0047] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, for example, at least one of a smart phone, a tablet computer, a laptop computer, and a desktop computer.

[0048] Server 105 can be any type of server that provides various services. For example, server 105 can be a cloud server, also known as a cloud computing server or cloud host, a host product within a cloud computing service system that addresses the management difficulties and poor scalability of traditional physical hosts and VPS (Virtual Private Server) services. Server 105 can also be a server for a distributed system or a server integrated with blockchain.

[0049] It should be noted that the resource matching model training method provided in the embodiments of the present disclosure can generally be executed by the server 105. The resource matching model training method provided in the embodiments of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0050] Alternatively, the resource matching model training method provided in the embodiments of the present disclosure may also generally be executed by the terminal device 101 , 102 , or 103 .

[0051] It should be noted that the resource matching method provided in the embodiments of the present disclosure can generally be executed by the server 105. The resource matching method provided in the embodiments of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0052] Alternatively, the resource matching method provided in the embodiment of the present disclosure may also generally be executed by the first terminal device 101 , the second terminal device 102 , or the third terminal device 103 .

[0053] It should be understood that the number of terminal devices, networks, and servers in Figure 1 is merely illustrative and any number of terminal devices, networks, and servers may be provided as required.

[0054] It should be noted that the sequence numbers of the operations in the following method are only used to indicate the operation for the purpose of description, and should not be regarded as indicating the order in which the operations should be performed. Unless explicitly stated, the method does not need to be performed in the order shown.

[0055] FIG2 is a flowchart of a method for training a resource matching model according to an embodiment of the present disclosure.

[0056] As shown in FIG. 2 , the method includes operations S210 to S220 .

[0057] In operation S210, a training operation is performed on the machine learning model set until a preset end condition is met.

[0058] In operation S220, a resource matching model is determined based on the machine learning model set obtained when a preset end condition is met.

[0059] According to an embodiment of the present disclosure, performing a training operation on a machine learning model set until a preset end condition is met may include the following operations: using a training object attribute data set corresponding to the machine learning model set to perform at least one round of training on the machine learning model set to obtain an intermediate machine learning model set corresponding to the machine learning model set; determining a model evaluation information set corresponding to the intermediate machine learning model set; and determining multiple target intermediate machine learning models from the intermediate machine learning model set based on the model evaluation information set as the machine learning model set for the next round of training.

[0060] According to an embodiment of the present disclosure, the training object attribute data set is a data set of a target application scenario, and the target application scenario characterization matches a first object resource with a second object resource; wherein, the machine learning model set includes multiple machine learning models, and the training object attribute data set includes multiple first training object attribute data subsets, and the machine learning model has a corresponding first training object attribute data subset.

[0061] According to an embodiment of the present disclosure, the target application scenario may be a scenario representing resource matching, such as a resource matching scenario between human resources and organizational resources. The first object resource may refer to human resources, and the second object resource may refer to organizational resources.

[0062] According to an embodiment of the present disclosure, the training object attribute data set may include attribute data of multiple dimensions of the first object resource, such as data representing the basic attributes of the first object, data representing the capability attributes of the first object, data representing the evaluation attributes of the first object, etc. Basic attributes may include the type of work, position, and projects participated in by the first object. Capability attributes may include the contribution of the first object in the type of work, position, and projects participated in. Evaluation attributes can be understood as reference data that can be used to evaluate the capabilities of the first object, such as the number of projects participated in, the level of the projects participated in, etc.

[0063] According to an embodiment of the present disclosure, a machine learning model set may include multiple machine learning models, and the model structure of each machine learning model and the structure of each machine learning model may be different, thereby reflecting the diversity of the model structure. Specifically, the machine learning model may include a logistic regression model, and each logistic regression model may be constructed based on a linear regression function, a Sigmoid function, and a quantization function. Among these three functions, the parameters of the linear regression function are different, or the parameters of the Sigmoid function are different, or the parameters of the quantization function are different, all of which can make the model structure of the logistic regression model different, and the machine learning model includes a logistic regression model, and the model structure of the logistic regression model is different, which will make the model structures between the machine learning model and the machine model also different from each other.

[0064] According to an embodiment of the present disclosure, the first training object attribute data subset can be split from the training object attribute data set and used to train one or more machine learning models. Specifically, one machine learning model can correspond to one first training object attribute data subset; one machine learning model can also correspond to multiple first training object attribute data subsets. Similarly, one first training object attribute data subset can also train multiple machine learning models. By providing multiple first training object attribute data subsets, the diversity of the training data set can be reflected.

[0065] According to an embodiment of the present disclosure, each training round may include multiple rounds of training. For example, a round of training may include 1 round of training or 10 rounds of training, and the specific number of rounds may be adaptively adjusted according to actual needs. It is understandable that the number of training rounds included in the next round may be the same as or different from the number of training rounds included in the previous round. The intermediate machine learning model may be obtained by the machine learning model during each round of training. For example, in a round of training, the machine learning model is trained for 10 rounds, and the intermediate machine learning model may be the model obtained by the machine learning model after training for 10 rounds.

[0066] According to an embodiment of the present disclosure, a model evaluation information set may be information obtained by evaluating and testing an intermediate machine learning model, and used to obtain the performance of the intermediate machine learning model. The model evaluation information set may include the accuracy, discrimination, etc. of the intermediate machine learning model. For example, in an embodiment, 10 machine learning models participate in the first round of training. After the first round of training, 10 intermediate machine learning models corresponding to the 10 machine learning models can be obtained. By analyzing the model evaluation information sets of these 10 intermediate machine learning models, it can be determined that there are currently 5 intermediate machine learning models with better performance. These 5 machine learning models with better performance can be used as 5 target machine learning models to participate in the next round of training, and the machine learning models with better model performance are selected to enter the next round of training. The remaining 5 intermediate machine learning models can stop the next round of training. It is understandable that the first training object attribute data subset used by the intermediate machine learning model entering the next round of training can be the same as the first training object attribute data subset used in the previous round of training.

[0067] Optionally, when there is one target intermediate machine learning model, the target intermediate machine learning model can be cyclically trained using only the first training object attribute data subset corresponding to the target intermediate machine learning model. There is no need to evaluate the model results of the target intermediate machine learning model and select the model with better performance. The target intermediate machine learning model can be trained until the target intermediate machine learning model reaches the preset end conditions.

[0068] According to embodiments of the present disclosure, the preset termination conditions may include at least one of the following: the number of training rounds reaches a preset maximum number of rounds, or the target loss function value constructed based on the model training results converges. The target loss function value can be determined based on the matching output value and matching label value of the training object attribute data. The form of the target loss function can be configured according to actual business needs and is not limited here. For example, the target loss function may include a cross-entropy loss function.

[0069] According to an embodiment of the present disclosure, a resource matching model may be a set of machine learning models that meet preset end conditions, and the resource matching model may include at least one machine learning model. The resource matching model may include a scoring card model. The scoring card model refers to an evaluation model constructed based on logistic regression according to various attributes and behavioral data of the evaluation object, and scores the evaluation object. Since the scoring card model has the characteristics of a simple model and strong interpretability, the matching efficiency can be improved in the process of matching the first object resource with the second object resource using the scoring card model, and an explainable reason can be provided.

[0070] According to an embodiment of the present disclosure, during the training process of the resource matching model, by using multiple training object attribute data sets to train multiple machine learning models, the training diversity of the machine learning model can be achieved, thereby improving the accuracy of the model output results. In addition, the training method of the resource matching model provided by the embodiment of the present disclosure also performs a model performance evaluation on the intermediate machine learning model after at least one round of training to obtain a model evaluation information set, and determines from the model evaluation information set that the machine learning model with better performance enters the next round of training process, eliminates the machine learning model with poor model performance, and realizes a dynamic optimization process of model training, thereby improving the efficiency of model training while improving the accuracy of the model output results.

[0071] According to embodiments of the present disclosure, the training object attribute dataset and the test object attribute dataset can be obtained by splitting the object attribute data. The object attribute data can be obtained through a preset data interface. The training object attribute dataset and the test object attribute dataset can include the same information dimensions, with the main difference being that the training object attribute dataset is used to train the machine learning model set, while the test object attribute dataset is used to verify the overall stability and assess the accuracy and discrimination of the intermediate machine learning model.

[0072] According to an embodiment of the present disclosure, the training object attribute dataset and the test object attribute dataset can be obtained by randomly splitting the object attribute data, or by splitting the object attribute dataset according to the ratio between the data volume of the training object attribute dataset and the data volume of the test object attribute data. For example, the ratio between the data volume of the training object attribute dataset and the data volume of the test object attribute data is 7:3, and the data volume of the test object attribute dataset accounts for 30% of the total data volume of the object attribute data. Then, according to the ratio of 30%, 30% of the attribute data can be randomly obtained from the object attribute data, and this 30% of the attribute data can be used as the test object attribute dataset. The intermediate machine learning model set is evaluated using the test object attribute dataset to obtain a model evaluation information set corresponding to the intermediate machine learning model.

[0073] According to an embodiment of the present disclosure, a training object attribute dataset for training is obtained by referring to the aforementioned method for obtaining a test object attribute dataset. For example, if the ratio between the data volume of the training object attribute dataset and the data volume of the test object attribute data is 7:3, and the data volume of the training object attribute dataset accounts for 70% of the total data volume of the object attribute data, then according to the 70% ratio, 70% of the attribute data can be randomly obtained from the object attribute data and used as the initial training object attribute dataset for training. The training object attribute dataset can be obtained by processing this initial training object attribute dataset.

[0074] According to an embodiment of the present disclosure, after splitting an object attribute dataset to obtain an initial training object attribute dataset and a test object attribute dataset, data processing can be performed on each of the initial training object attribute dataset and the test object attribute dataset. The resulting training object attribute dataset and the test object attribute dataset can then be used for training and testing, respectively. In this embodiment of the present disclosure, the data processing process is described using the processing of the initial training object attribute dataset to obtain the training object attribute dataset as an example. The processing process for the test object attribute dataset can refer to the processing process for the initial training attribute dataset and will not be repeated here.

[0075] According to an embodiment of the present disclosure, processing an initial training object attribute dataset to obtain a training object attribute dataset may include the following operations: discretizing the initial training object attribute dataset to obtain a first intermediate training object attribute dataset; continuating the first intermediate training object attribute dataset to obtain a second intermediate training object attribute dataset; and obtaining a training object attribute dataset based on the second intermediate training object attribute dataset.

[0076] According to an embodiment of the present disclosure, before discretizing the initial training object attribute data, a data cleaning operation can be performed on the initial training object attribute data set. The missing rate, the proportion of different values, and the proportion of unique values ​​of each object attribute data are calculated; and attribute data with a missing rate greater than 95% and a unique value proportion greater than 95% are deleted.

[0077] The missing rate of object attribute data can be calculated as the ratio of the number of missing data to the total number of all data. The total number of all data refers to the total number of attribute data that an object should have, while the number of missing data is the number of missing data among all attribute data. For example, if an object should have 100 attribute data, but only 5 data are actually obtained, the missing rate of object attribute data is 95%. The attribute data of this object cannot accurately train the machine learning model, so this object's attribute data can be deleted.

[0078] The proportion of different values ​​of object data can refer to the ratio between the number of non-repeated data between data and the total number of all data under the dimension in a certain dimension; similarly, the ratio between the number of repeated data between data and the total number of all data under the dimension in a certain dimension can be the ratio of unique values. For example, the attribute data of an object includes evaluation attribute dimension data. Among the evaluation attribute dimension data of 100 objects, 95 are the same data, all with the same value, and only 5 data are different from each other. In this case, the unique value ratio is 95% and the different value ratio is 5%. In this case, the same evaluation attribute dimension data is not good for the training of the machine learning model, so the 100 evaluation attribute dimension data can be deleted. According to an embodiment of the present disclosure, the discretization process can include data binning processing (Splitting processing). The initial training object attribute data set is binned to obtain a first intermediate training object attribute data set. The first intermediate training object attribute data set may include multiple first intermediate training object attribute dimension data subsets corresponding to each of the multiple candidate dimensions, and the first intermediate training object attribute dimension data subsets may include multiple first intermediate training object attribute dimension data.

[0079] According to an embodiment of the present disclosure, the binning process of the embodiment of the present disclosure may be supervised binning, such as Chi Square Box Splitting. Chi Square Box Splitting may be that if two adjacent intervals have very similar class distributions, the two intervals may be merged; otherwise, they may be separated. Optionally, the number of bins for binning may be set to 4, 5, or 6. When binning, the boundary value of the object attribute dimension data may be determined first, and the first intermediate training object attribute data set may be determined based on the boundary value and the amount of data in each box. Candidate dimensions may include one or more of basic attribute dimensions, capability attribute dimensions, and evaluation attribute dimensions.

[0080] According to the embodiments of the present disclosure, after binning, a series of discrete variables are obtained. These variables need to be encoded to convert them into continuous variables. During encoding, supervised encoding can be used: WOE (Weight of Evidence) encoding, which encodes discrete variables to obtain continuous variables.

[0081] According to an embodiment of the present disclosure, a first intermediate training object attribute dataset is serialized to obtain a second intermediate training object attribute dataset, including: determining, for each candidate dimension of at least one candidate dimension and for each first intermediate training object attribute dimension data subset in a plurality of first intermediate training object attribute dimension data subsets corresponding to the candidate dimension, a first number and a second number corresponding to the first intermediate training object attribute dimension data subset, wherein the first number represents the number of first-category data included in the first intermediate training object attribute dimension data subset, and the second number represents the number of second-category data included in the first intermediate training object attribute dimension data subset, the first-category data representing data whose label corresponding to the first intermediate training object attribute dimension data is a first label, and the second-category data representing data whose label corresponding to the first intermediate training object attribute dimension data is a second label; determining, based on a third number, a fourth number, the first number, and the second number, second intermediate training object attribute data corresponding to the first intermediate training object attribute dimension data subset, wherein the third number represents the number of first-category data included in the first intermediate training object attribute dimension data set, and the fourth number represents the number of second-category data included in the first intermediate training object attribute dimension data set; and obtaining the second intermediate training object attribute dataset based on the plurality of second intermediate training object attribute data corresponding to each of the plurality of candidate dimensions.

[0082] As a supervised encoding method, WOE can convert nonlinear data into linear data. The encoding process using WOE can be shown as formula (1).

[0083] Among them, Bad i Can represent the first number, the first type of data can refer to the negative label; Good i Can represent the second number, the second type of data can refer to the positive label; Bad T Can represent the third number; Good T Can represent the fourth number. i is the WOE code corresponding to the i-th bin of a certain feature, i is the i-th bin of a certain feature, Bad i is the number of negative labels for the i-th bin, Bad T is the total number of negative labels. T represents all bins. In one embodiment, Bad1=20, Bad T =50, Good i =80, Good T =450, then WOE1=0.8109.

[0084] The above formula shows the relationship between the proportion of negative labels and the Word of Equality (WOE) value: the greater the proportion of negative labels, the greater the WOE. Therefore, the WOE encoding value not only represents a classification but also represents the weight of that classification. Therefore, WOE encoding is particularly effective for training logistic regression models.

[0085] By using binning and WOE encoding to discretize and continuousize the initial training sample data set, continuous variables can be converted into categorical variables through binning, and categorical variables can be converted into continuous variables through WOE. This conversion between nonlinear data and linear data is helpful for the training of generalized linear models (GLM).

[0086] According to the embodiments of the present disclosure, WOE encoding can convert nonlinear features into linear ones. WOE encoding can also perform smoothly when encountering abnormal data. For example, the basic attribute dimension data of an object resource is 20, but it is mistakenly input as 200. Because WOE is insensitive to fluctuations in the input data, the final result will not fluctuate by 10 times. In addition, WOE encoding is easy to operate and has a wide range of applications. Compared with one-hot vector encoding, WOE encoding can ensure the integrity of variables while avoiding sparse matrix problems and dimensionality issues.

[0087] According to an embodiment of the present disclosure, after the initial training attribute dataset has been cleaned, binned, and WOE encoded, a correlation analysis can be performed. Specifically, obtaining the training object attribute dataset based on the second intermediate training object attribute dataset can include the following operations: determining the importance corresponding to each of the multiple candidate dimensions based on multiple second intermediate training object attribute dimension data corresponding to each of the multiple candidate dimensions; determining multiple target dimensions from the multiple candidate dimensions based on the importance corresponding to each of the multiple candidate dimensions; and obtaining the training object attribute dataset based on the second intermediate training object attribute dimension data corresponding to each of the multiple target dimensions of the multiple training objects. The second intermediate training object attribute dataset includes multiple second intermediate training object attribute dimension data subsets corresponding to each of the multiple candidate dimensions, and the second intermediate training object attribute dimension data subsets include multiple second intermediate training object attribute dimension data.

[0088] According to an embodiment of the present disclosure, the correlation analysis includes calculating the correlation between object attribute data and object attribute data and the matching degree between object attribute data and the first object resource and the second object resource, that is, calculating the correlation between independent variables and the correlation between independent variables and dependent variables. Among the independent variables with a correlation value greater than 0.7, the independent variable with a high correlation with the dependent variable can be retained, and the other independent variable can be deleted. In order to improve the training efficiency of the model, for multiple independent variables with high correlation values, only one of them can be retained. The importance can be the correlation value. Independent variables with a correlation value less than or equal to 0.7, and independent variables related to the dependent variable retained among the independent variables with a correlation value greater than 0.7, can be used as variable data of the target dimension. Based on the variable data of these target dimensions, a training object attribute data set can be generated.

[0089] According to an embodiment of the present disclosure, after the initial training object attribute data set undergoes the above-mentioned data cleaning, binning, WOE encoding, and correlation analysis to obtain the training object attribute data set, it can be used to train the corresponding machine learning models.

[0090] FIG3 is a schematic diagram of a training process of a resource matching model according to another embodiment of the present disclosure.

[0091] As shown in FIG3 , before using a training object attribute dataset to train a machine learning model, the training object attribute dataset 301 can be divided into multiple second training object attribute data subsets corresponding to multiple machine learning models, such as second training object attribute data subset A0 302_0, second training object attribute data subset A1 302_1, ..., and second training object attribute data subset A(n-1) 302_n-1, through operation S310. A first training object attribute data subset corresponding to a machine learning model can then be determined from each second training object attribute data subset.

[0092] According to an embodiment of the present disclosure, operation S310 may include dividing the training object attribute dataset into multiple second training object attribute data subsets using a duplicate partitioning strategy and a non-duplicate partitioning strategy. Specifically, the partitioning method for dividing the training object attribute dataset into the second training object attribute data subsets corresponding to each of the multiple machine learning models may include one of the following: dividing the training object attribute dataset into multiple second training object attribute data subsets with no duplicate object attribute data between the subsets; or dividing the training object attribute dataset into multiple second training object attribute data subsets with duplicate object attribute data between the subsets.

[0093] According to an embodiment of the present disclosure, determining a first training object attribute data subset corresponding to a machine learning model from each second training object attribute data subset may include the following operations: randomly dividing the second training object attribute data subset into K second training object attribute data groups, where K is an integer greater than 1; randomly determining K-1 first training object attribute data groups from multiple second training object attribute data groups; and obtaining a first training object attribute data subset based on the K-1 first training object attribute data groups.

[0094] Continuing with FIG3 , operation S320 may be performed on each second training object attribute data subset to divide each second training object attribute data subset into a second training object attribute data group B0 303_0, a second training object attribute data group B1 303_1, ..., and a second training object attribute data group B(k-1) 303_k-1. The first training object attribute data subset may be composed of k-1 data groups from the k data groups. For example, second training object attribute data group B0 303_0, second training object attribute data group B1 303_1, ..., and a second training object attribute data group B(k-2) 303_k-2 may be selected from the k second training object attribute data groups. These k-1 second training object attribute data groups serve as the first training object attribute data group, constituting the first training object attribute data subset. The remaining second training object attribute data group B(k-1) 303_k-1 may serve as test data for the machine learning model being trained. Operation 320 may include randomly dividing the second training object attribute data subset into a plurality of second training object attribute data groups. The process of determining the first training object attribute data subset corresponding to the machine learning model from the second training object attribute data subset may include a process of performing random K-fold sampling on the second training object attribute data subset.

[0095] According to an embodiment of the present disclosure, multiple machine learning models may have their own corresponding hyperparameter sets, and there may be at least two hyperparameter sets among the multiple hyperparameter sets that are different from each other. It may also include that each machine learning model and the hyperparameter set corresponding to each machine learning model are different, so as to reflect the diversity of the model hyperparameter sets.

[0096] Continuing with reference to FIG3 , operation S330 may include combining multiple machine learning models and multiple sets of hyperparameters to obtain multiple machine learning models for training, such as machine learning model B1A0 304_0, machine learning model B1A1 304_1, ... machine learning model B1A(n-1)304_n-1.

[0097] Operation S340 may include training k×n models (k-fold and N-class base models). That is, within the k data sets, each time the first training object attribute data subset used for training is changed, k training models can be obtained, as shown in first machine learning model set 305_0, first machine learning model set 305_1, ..., first machine learning model set 305_n-1. Exemplarily, the second training object attribute data subset is divided into three second training object attribute data groups, namely, second training object attribute data group B0, second training object attribute data group B1, and second training object attribute data group B2. The first round of training can use the second training object attribute data group B0 and the second training object attribute data group B1 to form the first training object attribute data subset for training, and the second training object attribute data group B2 is used to test this training; the second round of training can use the second training object attribute data group B0 and the second training object attribute data group B2 to form the first training object attribute data subset for training, and the second training object attribute data group B1 is used to test this training; the third round of training can use the second training object attribute data group B1 and the second training object attribute data group B2 to form the first training object attribute data subset for training, and the second training object attribute data group B0 is used to test this training. This is for example only and does not limit the solutions of the embodiments of the present disclosure.

[0098] According to an embodiment of the present disclosure, operation S350 may include outputting the machine learning model ultimately obtained from the training. The output may be based on the structural type of the machine learning model, for example, outputting machine learning model A0-306_0, machine learning model A1-306_1, ... machine learning model A(n-1)-306_n-1. The output may also be based on the order of model performance results from highest to lowest, or only the machine learning model with the best performance results may be output.

[0099] According to an embodiment of the present disclosure, operation S360 may include integrating the output models to obtain a final integrated resource matching model 307 .

[0100] According to the embodiments of the present disclosure, during the model training process, by combining the diversity of the model structure, the diversity of the first training object attribute data subset, and the diversity of the model hyperparameters, overfitting can be reduced, the performance of the resource matching model can be improved, and the accuracy of the resource matching model can be improved.

[0101] According to an embodiment of the present disclosure, the model evaluation information set includes a model evaluation value corresponding to each of at least one intermediate machine learning models; wherein, based on the model evaluation information set, multiple target intermediate machine learning models are determined from the intermediate machine learning model set to obtain a machine learning model set for the next round of training, which may include the following operations: sorting the intermediate machine learning model set according to the model evaluation value corresponding to each of the at least one intermediate machine learning models to obtain an intermediate machine learning model sorting result; selecting a preset number of intermediate machine learning models from the intermediate machine learning model sorting results to obtain a target intermediate machine learning model; and obtaining a machine learning model set for the next round of training based on the target intermediate machine learning model.

[0102] FIG4 is a flowchart of dynamic optimization during resource matching model training according to an embodiment of the present disclosure.

[0103] As shown in FIG4 , for example, the machine learning model set 401 initially participating in training may include 10 machine learning models. After operation S410 , the first intermediate machine learning model set 402 obtained may include 10 intermediate machine learning model sets corresponding to the machine learning models. Operation S410 may include performing at least one round of training on the machine learning model set.

[0104] By evaluating each intermediate machine learning model of the first intermediate machine learning model set 402 using the test object attribute data set, 10 evaluation values ​​can be obtained. A higher evaluation value represents a better performance of the intermediate machine learning model. These 10 evaluation values ​​are sorted from high to low, and the top 8 intermediate machine learning models are selected from the sorting results as target intermediate machine learning models to enter the next round of training, as shown in the second intermediate machine learning model set 403. Operation S420 may include screening out the intermediate machine learning model for the next round of training based on the model evaluation information set.

[0105] Operation S430 may include continuing to train the intermediate machine learning model in the second intermediate machine learning model set 403, and continuing to screen out the machine learning model set for the next round of training from the model evaluation information set obtained based on the intermediate machine learning model set after continued training, as shown in the third intermediate machine learning model set 404.

[0106] After multiple rounds of training and screening, the final intermediate machine learning model set can be shown as the fourth intermediate machine learning model set 405. These intermediate machine learning models in the fourth intermediate machine learning model set 405 can be used as machine learning models with better performance to build a resource matching model.

[0107] According to the embodiments of the present disclosure, by implementing the dynamic optimization process shown in Figure 4 during the model training process, the problems of large computing resource consumption and low model training efficiency caused by continuing the machine learning model with poor performance can be avoided, thereby improving resource utilization, reducing resource consumption, and improving model training efficiency.

[0108] In the logistic regression mathematical model, there are three main functions: linear regression function, Sigmoid function and quantization function.

[0109] Assume that the output of the binary classification problem is y∈{0,1}, and the predicted value z=w generated by the linear regression model T x+b is a real value, using the monotonically differentiable Sigmoid function As a step function. Since the Sigmoid function takes values ​​in [0, 1], it can be regarded as the posterior probability estimate p(y=1|x) of category 1. That is, if there is a test sample, the result calculated by the Sigmoid function can be regarded as the probability that the sample belongs to category 1. Therefore, a quantization function is used to classify Sigmoid function results greater than or equal to 0.5 as category 1, and less than 0.5 as category 0. The formula of the quantization function is shown in formula (2).

[0110] The above parameter w can be determined by the objective function of the logistic regression function as follows. The objective function of the logistic regression function is shown in formula (3).

[0111] Among them, in logistic regression, although the least squares method can be used, the sum of squared errors is used as the cost function of logistic regression. That is, z (i) =w T x (i) +b; i represents the i-th sample point, y (i) represents the true value of the i-th sample, Represents the predicted value of the i-th sample. (i) =w T x (i) +b substitute into the Sigmoid function After that, a non-convex function will appear, which means that the cost function has many local minima, which is not conducive to model training.

[0112] Therefore, the maximum likelihood estimation can be used to solve the problem of non-convex function caused by the least squares method. As mentioned above, It can be regarded as the posterior estimate of category 1, so the process shown in formula (4) is used.

[0113] Where p(y=1|x;w) represents the probability that sample x belongs to category 1 given w. Formula (4) can also be written in the general form

[0114] Next, we can use the maximum likelihood estimation method to estimate the parameter w based on the given training set. The process is shown in formula (5).

[0115] In order to simplify the computational resources, a logarithm is taken on both sides of formula (4), and the result is shown in formula (6).

[0116] After the conversion, if we want to find the w that maximizes l(w), we can take the negative value to find the minimum value. In this case, the objective function required can be shown in formula (7).

[0117] For example, n=1, the objective function can be shown as formula (8) and formula (9).

[0118] Right now:

[0119] If the sample value is 1, the estimated value The closer it is to 1, the smaller the cost, and vice versa. Similarly, if the sample value is 0, the estimated value The closer it is to 0, the smaller the cost, and vice versa. Such a cost function is consistent with the ideal effect.

[0120] After determining the objective function, we can use the gradient descent algorithm to find the convergence value. This involves utilizing the properties of the sigmoid function to facilitate computation, as well as derivatives and Taylor expansions. When the sample size is very large, updating the weights each time can be very time-consuming. In this case, stochastic gradient descent can be used. This method reshuffles the sample at each iteration and continuously updates the weights to determine the convergence value of the objective function.

[0121] According to an embodiment of the present disclosure, the machine learning model may be a logistic regression model, and the resource matching model may be a scorecard model. Therefore, the logistic regression model needs to be converted into a scorecard model.

[0122] The actual output of the scorecard model is a score, so the logistic regression model needs to be converted into a score output. That is, the conversion is performed using formulas (10) to (11). Score = offset + factor * (WX + b) (10) → Score = (offset + factor * b) + factor * (WX) (11)

[0123] Where offset and factor are parameters to be determined. offset + factor * b is called the base score B, and factor * (WX) is called the characteristic score F.

[0124] As can be seen from formulas (10) and (11), the scorecard does not use probability conversion to convert scores, but instead converts the linear part into scores. This has the effect of not only obtaining a total score, but also the scores of each feature. The total score is the base score plus the scores of all features. This makes the scorecard score easier to interpret and allows analysis of which features contribute to a low total score. In contrast, if the probability conversion is used, only the total score is obtained, not the scores of individual features.

[0125] The parameter settings of the scorecard model can include the following operations. When converting the probability model of logistic regression to the scoring model of the scorecard, the first thing to do is to set four parameters: the basic score B, the ratio of the probability of a bad customer to the probability of a good customer The Point of Double Odds (PDO) indicates how many minutes the odds are doubled, with a multiplier of k (usually 2). The PDO and multiplier k indicate that every time the odds decrease by k times, the score increases by D.

[0126] The parameter derivation of the scorecard model can include the following process. The mapping from odds to scores needs to be a simple linear equation, and it has been proven that the intercept in logistic regression is approximately equal to the logarithm of the ratio of negative labels to positive labels. Therefore, the scoring process can be shown as formula (12). Score = A + B × (WX + b) (12)

[0127] Since the logistic regression expression is It can be deduced Therefore, the results shown in formulas (13) and (14) can be obtained. Score = A + B × (WX + b) = A + B × ln (0dds) (13)

[0128] According to the definition of PDO, Score + PDO = A + B × ln(2Odds) (15)

[0129] According to formulas (13) to (15), the result shown in formula (16) can be obtained. PDO = B × ln (2 Odds) - B × ln (Odds) = B × ln (2) (16)

[0130] From formula (16), we can get the result shown in formula (17).

[0131] According to formulas (13), (14), and (17), we can get the result shown in formula (18).

[0132] From formula (18), we can get the result shown in formula (19).

[0133] In formula (19), the base score is used for Score, and the average level of Odds used when establishing the logistic regression model is used for Odds, thereby obtaining the value A. At this point, both the A and B values ​​have been calculated, resulting in the scorecard model: Score = AB × ln(Odds). Constant A is often called an offset, and constant B is called a scaling factor.

[0134] For example, assuming the base score is 600, the base odds is 50, and the PDO is set to 20, then B = 20 / ln(2) = 28.9, and A = 600-B×ln(50) = 487.1. The scorecard model is Score = 487.1 + 28.9×ln(Odds).

[0135] In the case of binning and WOE coding, the score corresponding to the bin to which the variable belongs is Among them, woe i is the WOE code value of the bin of the variable, β i is the coefficient of the bin in the logistic regression coefficient matrix, and n is the number of variables.

[0136] FIG5 is a schematic diagram of output results of a resource matching model according to an embodiment of the present disclosure.

[0137] As shown in Figure 5, the output of the resource matching model may include attribute data, indicator data included in the attribute data, matching scores, matching results, etc. Indicator data can be a more fine-grained division of attribute data. For example, if attribute data 1 represents capability attribute data, indicator 11 may refer to data on project participation, and indicator 12 may refer to data on activity participation, etc.

[0138] Figure 6A is the evaluation result of the resource matching model according to the embodiment of the present disclosure; Figure 6B is the accuracy level standard of the resource matching model according to the embodiment of the present disclosure; Figure 6C is the discrimination level standard of the resource matching model according to the embodiment of the present disclosure.

[0139] As shown in Figures 6A to 6C, Figure 6A schematically shows the results obtained by evaluating the model through training sets and test sets of different second resource objects. Comparing the results obtained in Figure 6A with the grade standard comparison table shown in Figures 6B and 6C, it can be seen that the accuracy value of the model is in the interval [0.85, 1). The resource matching model corresponding to this interval can more accurately output the matching results between the first object resource and the second object resource, so the accuracy of the resource matching model is a very good level. The discrimination value of the model is between the interval (0.6, 0.9). The resource matching model corresponding to this interval can more accurately distinguish the first object resource from the second object resource, so the discrimination of the resource matching model is a very strong level.

[0140] FIG7 is a flowchart of a resource matching method according to an embodiment of the present disclosure.

[0141] As shown in FIG7 , the resource matching method includes operations S710 to S720 .

[0142] In operation S710 , target object attribute data is acquired.

[0143] In operation S720, the target object attribute data is input into the resource matching model to obtain target matching information. The resource matching model is trained using the resource matching model training method described above.

[0144] According to an embodiment of the present disclosure, the target object may refer to a first object or a second object, such as human resources and organizational resources.

[0145] According to an embodiment of the present disclosure, the target matching information may refer to the degree of matching between the first object and the second object, and may also indicate a match or a mismatch.

[0146] According to the embodiments of the present disclosure, through the resource matching model of the embodiments of the present disclosure, since the model combines the diversity of model structure, the diversity of training data sets, and the diversity of model hyperparameters during the training process, it can improve the accuracy of predicting the degree of matching between the first object and the second object, and thus accurately reflect the actual degree of matching between the first object resource and the second control resource.

[0147] According to an embodiment of the present disclosure, the resource matching model includes at least one target machine learning model; operation S720 may include the following operations: inputting target object attribute data into at least one target machine learning model to obtain candidate matching information corresponding to each of the at least one target machine learning models; and obtaining target matching information based on at least one candidate matching information.

[0148] According to an embodiment of the present disclosure, each target machine learning model can obtain matching information, such as "match" or "mismatch," based on the input target object data. By counting the number of "matches" and "mismatches," the result with the largest number is used as the final output. In another embodiment, if the target machine learning model outputs a matching degree value, the matching degree values ​​obtained by multiple target machine models can be averaged, and the obtained average value is used as the final output matching degree value.

[0149] According to an embodiment of the present disclosure, the contents in operations S710 to S720 may refer to the relevant contents in operations S210 to S220 and will not be repeated here.

[0150] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0151] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method described above.

[0152] According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute the method described above.

[0153] According to an embodiment of the present disclosure, a computer program product includes a computer program, and when the computer program is executed by a processor, the computer program implements the method described above.

[0154] FIG8 is a block diagram of an electronic device suitable for implementing an image processing method according to an embodiment of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0155] As shown in Figure 8, electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In RAM 803, various programs and data required for the operation of electronic device 800 can also be stored. Computing unit 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to bus 804.

[0156] Multiple components in the electronic device 800 are connected to the I / O interface 805, including an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a magnetic disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the electronic device 800 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0157] The computing unit 801 can be a variety of general-purpose and / or specialized processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as the image processing method. For example, in some embodiments, the image processing method can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the image processing method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the image processing method by any other appropriate means (e.g., by means of firmware).

[0158] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0159] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data mining device, so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0160] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), 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 foregoing.

[0161] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0162] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0163] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0164] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not a limitation herein.

[0165] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A method for training a resource matching model, comprising: Perform training operations on the machine learning model set until the preset end conditions are met; as well as Determine a resource matching model based on a set of machine learning models obtained when the preset end condition is met; The performing of the training operation on the machine learning model set until a preset end condition is met includes: Using a training object attribute data set corresponding to the machine learning model set, training the machine learning model set for at least one round to obtain an intermediate machine learning model set corresponding to the machine learning model set; Determining a model evaluation information set corresponding to the intermediate machine learning model set; According to the model evaluation information set, determining a plurality of target intermediate machine learning models from the intermediate machine learning model set as the machine learning model set for the next round of training; Wherein, the training object attribute data set is a data set of a target application scenario, and the target application scenario characterizes matching a first object resource with a second object resource; Among them, the machine learning model set includes multiple machine learning models, the training object attribute data set includes multiple first training object attribute data subsets, and the machine learning model has a corresponding first training object attribute data subset.

2. The method according to claim 1, wherein: The model evaluation information set includes a model evaluation value corresponding to each of the at least one intermediate machine learning models; The step of determining a plurality of target intermediate machine learning models from an intermediate machine learning model set according to the model evaluation information set to obtain a machine learning model set for the next round of training includes: Sorting the intermediate machine learning model set according to the model evaluation value corresponding to each of the at least one intermediate machine learning models to obtain an intermediate machine learning model sorting result; Selecting a preset number of intermediate machine learning models from the intermediate machine learning model sorting results to obtain a target intermediate machine learning model; and Based on the target intermediate machine learning model, a machine learning model set for the next round of training is obtained.

3. The method according to claim 1, further comprising: Dividing the training object attribute data set into a plurality of second training object attribute data subsets corresponding to each of the plurality of machine learning models; as well as From each of the second training object attribute data subsets, determine the first training object attribute data subset corresponding to the machine learning model.

4. The method according to claim 3, wherein: Determining the first training object attribute data subset corresponding to the machine learning model from each of the second training object attribute data subsets comprises: Randomly dividing the second training object attribute data subset into K second training object attribute data groups, where K is an integer greater than 1; Randomly determine K-1 first training object attribute data groups from the plurality of second training object attribute data groups; and The first training object attribute data subset is obtained according to the K-1 first training object attribute data groups.

5. The method according to claim 3, wherein: The step of dividing the training object attribute data set into second training object attribute data subsets corresponding to each of the plurality of machine learning models comprises one of the following: Dividing the training object attribute data set into a plurality of second training object attribute data subsets where object attribute data is not repeated between subsets; The training object data set is divided into a plurality of second training object attribute data subsets in which object attribute data are repeated between subsets.

6. The method according to claim 4, further comprising: determining an initial training object attribute data set from the object attribute data according to a ratio between the amount of the training object attribute data and the total amount of the object attribute data; as well as The initial training object attribute data set is processed to obtain the training object attribute data set.

7. The method according to claim 6, wherein: The processing of the initial training object attribute data set to obtain the training object attribute data set includes: Discretizing the initial training object attribute data set to obtain a first intermediate training object attribute data set; performing serial processing on the first intermediate training object attribute data set to obtain a second intermediate training object attribute data set; and The training object attribute dataset is obtained according to the second intermediate training object attribute dataset.

8. The method according to claim 7, wherein: Discretizing the initial training object attribute data set to obtain a first intermediate training object attribute data set, including: Binning the initial training object attribute data set to obtain a first intermediate training object attribute data set; The first intermediate training object attribute data set includes a plurality of first intermediate training object attribute dimension data subsets corresponding to a plurality of candidate dimensions respectively, and the first intermediate training object attribute dimension data subset includes a plurality of first intermediate training object attribute dimension data.

9. The method according to claim 8, wherein: The step of performing continuous processing on the first intermediate training object attribute data set to obtain a second intermediate training object attribute data set includes: For each candidate dimension of the at least one candidate dimension, For each first intermediate training object attribute dimension data subset in a plurality of first intermediate training object attribute dimension data subsets corresponding to the candidate dimension, Determine a first number and a second number corresponding to the first intermediate training object attribute dimension data subset, wherein the first number represents the number of first-category data included in the first intermediate training object attribute dimension data subset, the second number represents the number of second-category data included in the first intermediate training object attribute dimension data subset, the first-category data represents data whose label corresponding to the first intermediate training object attribute dimension data is a first label, and the second-category data represents data whose label corresponding to the first intermediate training object attribute dimension data is a second label; Determine, according to a third number, a fourth number, the first number, and the second number, second intermediate training object attribute data corresponding to the first intermediate training object attribute dimension data subset, wherein the third number represents the number of the first category of data included in the first intermediate training object attribute dimension data set, and the fourth number represents the number of the second category of data included in the first intermediate training object attribute dimension data set; The second intermediate training object attribute data set is obtained according to a plurality of second intermediate training object attribute data corresponding to each of the plurality of candidate dimensions.

10. The method according to claim 8, wherein: The second intermediate training object attribute data set includes a plurality of second intermediate training object attribute dimension data subsets corresponding to each of the plurality of candidate dimensions, and the second intermediate training object attribute dimension data subset includes a plurality of second intermediate training object attribute dimension data; Wherein, obtaining the training object attribute data set according to the second intermediate training object attribute data set includes: Determining the importance corresponding to each of the plurality of candidate dimensions according to the plurality of second intermediate training object attribute dimension data corresponding to each of the plurality of candidate dimensions; determining the plurality of target dimensions from the plurality of candidate dimensions according to the importance corresponding to each of the plurality of candidate dimensions; and The training object attribute data set is obtained according to the second intermediate training object attribute dimension data corresponding to each of the plurality of target dimensions of the plurality of training objects.

11. The method according to any one of claims 1 to 10, wherein: The multiple machine learning models have their own corresponding hyperparameter sets, and among the multiple hyperparameter sets, there are at least two hyperparameter sets that are different from each other.

12. The method according to any one of claims 1 to 10, wherein: The multiple machine learning models include at least two model structures that are different from each other, and the machine learning model is a logistic regression model.

13. The method according to claim 6, wherein: Determining the model evaluation information set corresponding to the intermediate machine learning model set includes: determining a test object attribute data set from the object attribute data sets according to a ratio between the amount of the test object attribute data and the total amount of the object attribute data; and The intermediate machine learning model set is evaluated for model effectiveness using the test object attribute data set to obtain a model evaluation information set corresponding to the intermediate machine learning model.

14. The method according to claim 1, wherein: The preset end condition includes one of the following: The number of training rounds reaches the preset maximum number of rounds, and the loss function constructed based on the model training results converges.

15. The method according to any one of claims 1 to 14, wherein: When there is one target intermediate machine learning model, the target intermediate machine learning model is cyclically trained using the first training object attribute data subset corresponding to the target intermediate machine learning model until the target intermediate machine learning model reaches a preset end condition.

16. A resource matching method, comprising: Get the target object attribute data; as well as Inputting the target object attribute data into a resource matching model to obtain target matching information; The resource matching model is trained by using the method according to any one of claims 1 to 15.

17. The method according to claim 16, wherein: The resource matching model includes at least one target machine learning model; The step of inputting the target object attribute data into a resource matching model to obtain target matching information includes: Inputting the target object attribute data into the at least one target machine learning model to obtain candidate matching information corresponding to each of the at least one target machine learning model; and The target matching information is obtained according to at least one of the candidate matching information.

18. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to execute the method according to any one of claims 1 to 17.

19. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method according to any one of claims 1 to 17.

20. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 17 is implemented.

Images