Model construction method and device, storage medium and electronic device

Abstract

The application provides a model construction method and device, a storage medium and an electronic device, and the method comprises the following steps: obtaining N first historical sample sets and a second historical sample set, a first historical sample in the same first historical sample set corresponds to the same interest rate information, one second historical sample corresponds to randomly generated interest rate information, N is a positive integer; each first historical sample set in the N first historical sample sets is used to determine N single interest rate models, each single interest rate model in the N single interest rate models corresponds to each first historical sample set in a one-to-one manner; and the second historical sample set and each single interest rate model are used to construct an interest rate sensitive model, and the interest rate sensitive model is used to predict the interest rate sensitivity of a target object. The interest rate sensitivity of the target object can be accurately obtained, and the interest rate information suitable for the target object can be determined based on the interest rate sensitivity of the target object.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a model construction method and device, a storage medium and an electronic device. BACKGROUND

[0002] At present, the interest rate (also referred to as interest rate information) pricing method in the credit scenario has important value. If a more high-quality object (i.e., a user) group is given a low price (i.e., a lower interest rate), the profitability will be reduced. If a high-quality sensitive object (i.e., an object with a higher interest rate sensitivity) is given a high price, the object may be lost. As can be seen, the interest rate sensitivity can play an important role in the interest rate pricing method. In related technologies, the interest rate is usually priced based on risk, or the object's willingness to perform, social value, and repayment ability (i.e., the interest rate information is given by combining an object's willingness to perform model, a social value model, and a repayment ability model), or the object's marketing response degree is classified by an unsupervised machine learning method, and then the object's sensitivity to the interest rate (i.e., the object's interest rate sensitivity) is evaluated. The so-called risk pricing refers to the price determination of a risky asset, which reflects a functional relationship between the future income and risk of a capital asset. The risk pricing system needs to consider factors such as operating costs, target profit margin, fund supply and demand relationship, market interest rate level, and customer risk. That is, different interest rates can be given to different object groups based on risk, and an artificial strategy or a logistic regression model is usually used. However, whether the interest rate is priced based on risk or the object's willingness to perform, social value, and repayment ability, it is impossible to differentiate the pricing of a subdivided object group, i.e., only a corresponding interest rate can be given to an object group. The sample of unsupervised learning (i.e., solving various problems in pattern recognition according to a training sample with unknown categories (not labeled)) is limited and the accuracy is difficult to evaluate, so that the related technologies are difficult to accurately analyze the target object when pricing the interest rate. Therefore, how to accurately obtain the interest rate sensitivity of the target object and determine the appropriate interest rate information for the target object based on the interest rate sensitivity of the target object has become a research hotspot. SUMMARY

[0003] Therefore, the embodiments of the present application provide a model construction method and device, a storage medium, and an electronic device to solve the problems that it is difficult to accurately analyze the target object when pricing the interest rate and it is difficult to determine the appropriate interest rate information for the target object. That is, the embodiments of the present application can accurately obtain the interest rate sensitivity of the target object and determine the appropriate interest rate information for the target object based on the interest rate sensitivity of the target object.

[0004] According to an aspect of the present application, a model construction method is provided, which includes:

[0005] N first historical sample sets are obtained, and a second historical sample set is obtained, first historical samples in a same first historical sample set correspond to same interest rate information, and interest rate information corresponding to a second historical sample is randomly generated interest rate information, N is a positive integer;

[0006] Each of the N first historical sample sets is used to determine a single interest rate model, each of the N single interest rate models corresponds to a same first historical sample set;

[0007] Based on the second historical sample set and the single interest rate models, an interest rate sensitive model is constructed, and the interest rate sensitive model is used to predict interest rate sensitivity of a target object.

[0008] According to another aspect of the present application, a model construction device is provided, and the device comprises:

[0009] An obtaining unit is configured to obtain N first historical sample sets, and obtain a second historical sample set, first historical samples in a same first historical sample set correspond to same interest rate information, and interest rate information corresponding to a second historical sample is randomly generated interest rate information, N is a positive integer;

[0010] A processing unit is configured to use each of the N first historical sample sets to determine a single interest rate model, each of the N single interest rate models corresponds to a same first historical sample set;

[0011] The processing unit is further configured to construct an interest rate sensitive model based on the second historical sample set and the single interest rate models, and the interest rate sensitive model is used to predict interest rate sensitivity of a target object.

[0012] According to another aspect of the present application, an electronic device is provided, and the electronic device comprises a processor and a memory storing a program, wherein the program comprises instructions which, when executed by the processor, cause the processor to perform the above-mentioned method.

[0013] According to another aspect of the present application, a non-transitory computer readable storage medium storing computer instructions is provided, and the computer instructions are used to cause a computer to perform the above-mentioned method.

[0014] The embodiment of the present application can determine N single interest rate models by using each first historical sample set in the N first historical sample sets after obtaining the N first historical sample sets and obtaining the second historical sample set, the first historical samples in the same first historical sample set correspond to the same interest rate information, the interest rate information corresponding to one second historical sample is randomly generated interest rate information, each single interest rate model in the N single interest rate models corresponds to each first historical sample set in one-to-one, and N is a positive integer; then, the interest rate sensitive model is constructed based on the second historical sample set and each single interest rate model, and the interest rate sensitive model is used to predict the interest rate sensitivity of the target object. It can be seen that the embodiment of the present application can construct the interest rate sensitive model for the interest rate pricing, so as to obtain the interest rate sensitivity of the target object through the interest rate sensitive model, and to use the interest rate sensitivity of the target object to make more differentiated interest rate pricing for the target object, that is, the embodiment of the present application can determine the appropriate interest rate information for the target object based on the interest rate sensitivity of the target object. At the same time, the embodiment of the present application can learn important information on the N first historical sample sets based on the feature construction method of the transfer learning, and fine-tune on the second historical sample set, which has great significance for the construction of the interest rate sensitive model feature information, can effectively improve the model performance of the interest rate sensitive model, so as to accurately obtain the interest rate sensitivity of the target object through the interest rate sensitive model, that is, the embodiment of the present application can improve the accuracy of the interest rate sensitivity. BRIEF DESCRIPTION OF DRAWINGS

[0015] In the following description of exemplary embodiments in conjunction with the drawings, more details, features and advantages of the present application are disclosed, in which:

[0016] Figure 1 A flowchart of a model construction method according to an exemplary embodiment of the present application is shown;

[0017] Figure 2 A schematic diagram of a construction process of an interest rate sensitive model according to an exemplary embodiment of the present application is shown;

[0018] Figure 3 A flowchart of another model construction method according to an exemplary embodiment of the present application is shown;

[0019] Figure 4 A schematic diagram of transfer learning according to an exemplary embodiment of the present application is shown;

[0020] Figure 5 A schematic diagram of an application process of an interest rate sensitive model according to an exemplary embodiment of the present application is shown;

[0021] Figure 6 A schematic block diagram of a model construction device according to an exemplary embodiment of the present application is shown;

[0022] Figure 7 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present application is shown. DETAILED DESCRIPTION

[0023] Embodiments of the present application will be described in more detail with reference to the drawings. While the present application is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the drawings are not intended to limit the present application to the particular embodiments disclosed, but on the contrary, the intention is to cover all modifications, equivalents and alternatives falling within the spirit and scope of the present application as defined by the appended claims. It should be understood that the present application is not limited in scope to the specific embodiments disclosed herein, which are intended as examples only and which will be implemented or applied in a variety of ways. Those skilled in the art will appreciate the scope of the present application and recognize applications and implications thereof based on the following detailed disclosure together with the attached figures.

[0024] It should be understood that the various steps of the method embodiments of the present application can be performed in different orders and / or in parallel. In addition, the method embodiments can include additional steps and / or omit performing the steps shown. The scope of the present application is not limited in this respect.

[0025] The term "comprises" and variations thereof used in the present document such as "comprising" and "comprises" are open-ended, that is, "including, but not limited to." The term "based on" is intended to mean "based, at least in part, on" The term "one embodiment" means "at least one embodiment." The term "another embodiment" means "at least one additional embodiment." The term "some embodiments" means "at least some embodiments." Related terms are defined as follows. It should be noted that the concepts mentioned in the present application are merely illustrative and not restrictive, and those skilled in the art should understand that "one", "multiple" modification is illustrative and not restrictive, and unless otherwise explicitly stated in the context, it should be understood as "one or more".

[0026] It should be noted that the "one", "multiple" modification mentioned in the present application is illustrative and not restrictive, and those skilled in the art should understand that unless otherwise explicitly stated in the context, it should be understood as "one or more".

[0027] The names of the messages or information exchanged between the devices in the embodiments of the present application are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0028] It should be noted that the execution subject of the model construction method provided by the embodiments of the present application can be one or more electronic devices, and the present application does not limit this; wherein the electronic device can be a terminal (i.e. client) or a server, so when the execution subject includes multiple electronic devices, and the multiple electronic devices include at least one terminal and at least one server, the model construction method provided by the embodiments of the present application can be executed by the terminal and the server together. Correspondingly, the terminal mentioned here can include but not limited to: smart phone, tablet computer, notebook computer, desktop computer, smart watch, smart voice interaction device, smart home appliance, vehicle-mounted terminal, aircraft, etc. The server mentioned here can be a standalone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platform, etc.

[0029] Based on the above description, the embodiments of the present application provide a model construction method, which can be executed by the above-mentioned electronic device (terminal or server); or the model construction method can be executed by the terminal and the server together. In order to facilitate description, the model construction method is executed by the electronic device as an example in the following; for example, as shown in the figure, the model construction method can include the following steps S101-S103: Figure 1

[0030] S101, obtain N first historical sample sets, and obtain a second historical sample set, the first historical samples in the same first historical sample set correspond to the same interest rate information, and the interest rate information corresponding to one second historical sample is randomly generated interest rate information, N is a positive integer.

[0031] ​The sample set composed of the N first historical sample sets can also be referred to as a full-amount historical sample. Based on this, when the N first historical sample sets are acquired, the electronic device can acquire the full-amount historical sample, and divide the sample space according to the N interest rates on the full-amount historical sample, that is, divide the sample space according to the interest rate information on the full-amount historical sample, so as to obtain the N first historical sample sets, and each first historical sample set includes the first historical samples in a sample space. In this case, the electronic device can divide the first historical samples with the same interest rate information in the same sample space, so that the interest rate information corresponding to each first historical sample in the same sample space is the same, that is, the interest rate information corresponding to each first historical sample in the same first historical sample set is the same. Optionally, the electronic device can also directly acquire the N first historical sample sets, that is, the full-amount historical sample has been divided into the N first historical sample sets according to the interest rate information, that is, the full-amount historical sample has been divided into the N first historical sample sets, at this time, the electronic device can no longer divide the sample space of the full-amount historical sample after acquiring the full-amount historical sample, and the like; the present application does not limit this.

[0032] It should be noted that one sample (such as a first historical sample or a second historical sample) can be used to represent the object characteristics of an object (i.e., a user); optionally, a sample can include but is not limited to: historical loan amount, whether there is a mortgage, and income information, and the like; the present application does not limit this.

[0033] In the embodiments of the present application, the acquisition method of the N first historical sample sets and the second historical sample set is taken as an example for description, and the acquisition method of the N first historical sample sets and the second historical sample set can include but is not limited to the following several methods:

[0034] The first acquisition method: the electronic device can acquire a first download link of the N first historical sample sets, and download the sample set according to the first download link, so as to download the sample set based on the first download link as the N first historical sample sets; correspondingly, the electronic device can acquire a second download link of the second historical sample set, and download the sample set according to the second download link, so as to download the sample set based on the second download link as the second historical sample set.

[0035] The second acquisition method: the electronic device stores a plurality of first historical sample sets and a plurality of second historical sample sets, so that the electronic device can select N first historical sample sets from the plurality of first historical sample sets to acquire the N first historical sample sets; and the electronic device can select a second historical sample set from the plurality of second historical sample sets to acquire the second historical sample set, and the like.

[0036] S102, respectively adopting each of the N first historical sample sets, determine N single interest rate models, each of the N single interest rate models corresponds to each of the first historical sample sets.

[0037] In a specific implementation, since one first historical sample set includes a first historical sample in one sample space, respectively adopting each of the first historical sample sets to determine N single interest rate models can refer to: in each of the N sample spaces, independently training N single interest rate models (also referred to as sub-models), as shown in FIG. 2. Based on this, it can be ensured that the response under the single interest rate (i.e., whether to use the credit) is fitted in each sample space. Figure 2

[0038] For example, assuming that the value of N is 2, the N first historical sample sets include a first historical sample set 1 and a first historical sample set 2, and the interest rate information corresponding to each first historical sample in the first historical sample set 1 is interest rate information 1, and the interest rate information corresponding to each first historical sample in the first historical sample set 2 is interest rate information 2, then the electronic device can perform model training in the sample space corresponding to the first historical sample set 1 to obtain the single interest rate model 1 corresponding to the first historical sample set 1, and the single interest rate model 1 fits the response under the interest rate information 1; and, the model training can be performed in the sample space corresponding to the first historical sample set 2 to obtain the single interest rate model 2 corresponding to the first historical sample set 2, and the single interest rate model 2 fits the response under the interest rate information 2, and so on.

[0039] S103, based on the second historical sample set and each single interest rate model, constructing an interest rate sensitive model, the interest rate sensitive model is used to predict the interest rate sensitivity of the target object.

[0040] Wherein, the target object can be any object that needs to handle the target business, and the target business can refer to a housing loan business, or a car loan business, etc., which is not limited by the present application. Optionally, the above interest rate sensitivity can also be referred to as interest rate sensitivity difference or sensitive model classification.

[0041] ​It should be understood that, since the random pricing (i.e., the randomly generated interest rate information) has an uncertain impact on the indicators of the target business (such as the situation that the randomly generated interest rate information is too large to cause the object to default, etc.), the number of the second historical samples in the second historical sample set is small to avoid fluctuations in the business data. Based on this, when constructing the interest rate sensitive model, if only the full amount of historical samples (i.e., the N first historical sample sets) are used for training, the learned interest rate sensitivity will be more affected by the historical pricing strategy; if only the historical random pricing samples (i.e., the second historical samples) are used for training, the sample is too small to cause the model to learn insufficiently or unstably; and although the full amount of historical samples is used to train separately in different sample spaces, the historical pricing strategy is learned instead of the real sensitivity characteristics of the object, but the small amount of historical random pricing samples can be fine-tuned and a large amount of feature information under the data is learned. It can be seen that, based on the feature construction method of transfer learning, the important information is learned on a large amount of data (i.e., the N first historical sample sets), and fine-tuning is performed on a small amount of historical random pricing samples, which can improve the model performance of the interest rate sensitive model; wherein, the so-called transfer learning is a method of machine learning, which refers to a pre-trained model being used in another task.

[0042] After the N first historical sample sets and the second historical sample set are obtained, the embodiments of the present application can determine N single interest rate models by using each of the first historical sample sets in the N first historical sample sets, respectively, the first historical samples in the same first historical sample set correspond to the same interest rate information, and the interest rate information corresponding to a second historical sample is randomly generated interest rate information, each of the single interest rate models in the N single interest rate models corresponds to each of the first historical sample sets one by one, and N is a positive integer; then, based on the second historical sample set and each of the single interest rate models, an interest rate sensitive model is constructed, and the interest rate sensitive model is used to predict the interest rate sensitivity of the target object. It can be seen that, the embodiments of the present application can construct an interest rate sensitive model for interest rate pricing, so as to obtain the interest rate sensitivity of the target object through the interest rate sensitive model, and to use the interest rate sensitivity of the target object to price the target object with more differentiated interest rate, that is, the embodiments of the present application can determine appropriate interest rate information for the target object based on the interest rate sensitivity of the target object. At the same time, based on the feature construction method of transfer learning, the embodiments of the present application can learn important information on the N first historical sample sets, fine-tune on the second historical sample set, which has great significance for the construction of the feature information of the interest rate sensitive model, and can effectively improve the model performance of the interest rate sensitive model, so as to accurately obtain the interest rate sensitivity of the target object through the interest rate sensitive model, that is, the embodiments of the present application can improve the accuracy of the interest rate sensitivity.

[0043] Based on the above description, the embodiment of the present application further proposes a more specific model construction method. Correspondingly, the model construction method can be executed by the electronic device (terminal or server) mentioned above; or the model construction method can be executed by the terminal and the server together. In order to facilitate the description, the model construction method executed by the electronic device is taken as an example for description hereinafter; please refer to Figure 3 The model construction method can include the following steps S301-S307:

[0044] S301, obtaining N first historical sample sets, and obtaining a second historical sample set, the first historical samples in the same first historical sample set correspond to the same interest rate information, and the interest rate information corresponding to one second historical sample is randomly generated interest rate information, N is a positive integer.

[0045] In an embodiment, the number of second historical samples included in the second historical sample set can be determined based on a preset sample ratio and a target data amount, and the target data amount can include any one of the following: the number of first historical samples in the N first historical sample sets and the business data amount of the target business handled within a target time range, and the preset sample ratio is less than a sample ratio threshold. Optionally, the target data amount can also be a preset data amount threshold (such as one thousand or ten thousand, etc.), and the present application does not limit the specific value of the target data amount. It should be understood that the above-mentioned target time range can be one year in the past, or one month in the past, etc., and the present application does not limit this; correspondingly, the above-mentioned preset sample ratio can be 0.1, or 0.2, etc., and the present application does not limit this.

[0046] It should be noted that in order to avoid the fluctuation of the business data of the target business caused by random pricing (i.e. randomly generating interest rate information), the number of second historical samples in the second historical sample set must be controlled in a small proportion, that is, the above-mentioned preset sample ratio needs to be less than the sample ratio threshold, so that the number of second historical samples is small. Optionally, the sample ratio threshold can be 0.3, or 0.4, etc., and the present application does not limit this.

[0047] In another embodiment, the number of second historical samples included in the second historical sample set can be determined based on a preset generation quantity threshold, in which case the number of second historical samples is less than the preset generation quantity threshold, etc. Optionally, the preset generation quantity threshold can be set according to experience, or set according to actual needs, and the present application does not limit this.

[0048] Further, the second historical sample set can include the first random interest rate sample set and the second random interest rate sample set, and the interest rate information corresponding to any first random interest rate sample in the first random interest rate sample set is greater than or equal to the interest rate information corresponding to each second random interest rate sample in the second random interest rate sample set; in this case, the first random interest rate sample can also be referred to as a high-price random sample, and the second random interest rate sample can also be referred to as a low-price random sample. It should be understood that, since the number of second historical samples in the second historical sample set is small, the number of first random interest rate samples in the first random interest rate sample set and the number of second random interest rate samples in the second random interest rate sample set are also small.

[0049] Optionally, the first random interest rate samples and the second random interest rate samples in the second historical sample set can be arranged in disorder, that is, the first random interest rate samples in the first random interest rate sample set and the second random interest rate samples in the second random interest rate sample set are fused into one set (i.e., the second historical sample set); in this case, the electronic device can divide the second historical sample set into the first random interest rate sample set and the second random interest rate sample set according to the interest rate information.

[0050] In a specific implementation, the electronic device can determine a first interest rate range corresponding to the first random interest rate sample set and a second interest rate range corresponding to the second random interest rate sample set, and then the electronic device can divide, according to the interest rate information corresponding to each second historical sample in the second historical sample set, the second historical sample whose interest rate information is located in the first interest rate range into the first random interest rate sample set, and divide the second historical sample whose interest rate information is located in the second interest rate range into the second random interest rate sample set. Wherein, any interest rate information in the first interest rate range is greater than or equal to each interest rate information in the second interest rate range; optionally, the first interest rate range and the second interest rate range can be set according to experience or set according to actual needs, and the present application does not limit this.

[0051] In another specific implementation, the electronic device can sort each second historical sample in descending order of the interest rate information corresponding to each second historical sample to obtain a first sorting result; then, determine a first division ratio, and determine first division position information according to the first division ratio and the number of second historical samples, and divide the second historical samples in the first sorting result located at and before the first division position information into the first random interest rate sample set, and divide the second historical samples in the first sorting result located after the first division position information into the second random interest rate sample set. The first division ratio can be set according to experience or actual demand, and the present application does not limit this. For example, when the product of the first division ratio and the number of second historical samples is taken as the first division position information, assuming that the first division ratio is 0.5 and the number of second historical samples is 100, the first division position information can be 50. In this case, the electronic device can divide the second historical samples in the first sorting result located in the first 50 (including the first second historical sample to the 50th second historical sample) into the first random interest rate sample set, and divide the second historical samples in the first sorting result located in the last 50 (including the 50th second historical sample to the 100th second historical sample) into the second random interest rate sample set.

[0052] In another specific implementation, the electronic device can sort each second historical sample in descending order of the interest rate information corresponding to each second historical sample to obtain a first sorting result; then, determine a first division ratio, and determine first division position information according to the first division ratio and the number of second historical samples, and divide the second historical samples in the first sorting result located at and before the first division position information into the first random interest rate sample set, and divide the second historical samples in the first sorting result located after the first division position information into the second random interest rate sample set. The first division ratio can be set according to experience or actual demand, and the present application does not limit this. For example, when the product of the first division ratio and the number of second historical samples is taken as the first division position information, assuming that the first division ratio is 0.5 and the number of second historical samples is 100, the first division position information can be 50. In this case, the electronic device can divide the second historical samples in the first sorting result located in the first 50 (including the first second historical sample to the 50th second historical sample) into the first random interest rate sample set, and divide the second historical samples in the first sorting result located in the last 50 (including the 50th second historical sample to the 100th second historical sample) into the second random interest rate sample set.

[0053] It should be understood that if the obtained second historical sample set is a set that has been divided into a first random interest rate sample set and a second random interest rate sample set, the electronic device does not need to divide the second historical sample set according to the interest rate information.

[0054] S302, respectively using each first historical sample set in the N first historical sample sets to determine N single interest rate models, each single interest rate model in the N single interest rate models corresponding to each first historical sample set.

[0055] Specifically, for the i-th first historical sample set in the N first historical sample sets, the electronic device can call the initial interest rate model corresponding to the i-th first historical sample set, perform credit use rate prediction on each first historical sample in the i-th first historical sample set, obtain the predicted credit use rate of each first historical sample, and determine the predicted category of each first historical sample based on the predicted credit use rate of each first historical sample, i ∈ [1, N]. Wherein, the credit use rate of a sample refers to: the probability of the object corresponding to the corresponding sample performing the performance under the interest rate information corresponding to the corresponding sample; optionally, the credit use rate of a sample can also refer to: the probability of the object corresponding to the corresponding sample performing the performance according to the preset validity period under the interest rate information corresponding to the corresponding sample, that is, the probability of performing the performance within the preset validity period; optionally, the preset validity period can be 30 days, 45 days, etc., which is not limited by the present application. Therefore, whether to use credit (i.e., whether to perform the performance) within the preset validity period can be used as a response for model training, that is, the model can be trained by whether to use credit category.

[0056] Specifically, whether to use credit category can include credit and non-credit; correspondingly, if the predicted credit use rate of any sample is greater than the preset credit use rate, the electronic device can determine that the category to which the sample belongs is credit (i.e., the predicted category of the sample is credit), and if the predicted credit use rate of the sample is less than or equal to the preset credit use rate, the electronic device can determine that the category to which the sample belongs is non-credit (i.e., the predicted category of the sample is non-credit). Optionally, the preset credit use rate can be set according to experience or actual demand, which is not limited by the present application.

[0057] Further, the electronic device can calculate the initial interest rate model loss value of the initial interest rate model based on the predicted category of each first historical sample and the labeled category of the corresponding first historical sample, and optimize the model parameters in the initial interest rate model in the direction of reducing the initial interest rate model loss value, to determine the single interest rate model corresponding to the i-th first historical sample set based on the initial interest rate model after model optimization. Wherein, the labeled category of any sample can be the actual whether-to-use credit category of the object corresponding to the sample, that is, if the object corresponding to the sample performs the performance in the actual situation, the labeled category of the sample can be credit, and if the object corresponding to the sample does not perform the performance in the actual situation, the labeled category of the sample can be non-credit.

[0058] In the embodiment of the present application, when the single interest rate model corresponding to the i th first historical sample set is determined based on the initial interest rate model optimized based on the model, the electronic device can continue to optimize the model of the initial interest rate model optimized based on the model until the interest rate model convergence condition is met, and the interest rate model that meets the interest rate model convergence condition is taken as the single interest rate model corresponding to the i th first historical sample set. Optionally, the interest rate model convergence condition can refer to that the number of iterations reaches the interest rate model iteration number threshold, or the interest rate model loss value is less than the interest rate model loss threshold, etc., which is not limited in the present application. Correspondingly, the interest rate model iteration number threshold and the interest rate model loss threshold can be set according to experience or actual demand, which is not limited in the present application.

[0059] Optionally, the initial interest rate model can be a Lightgbm model (Light Gradient Boosting Machine, a distributed gradient boosting framework based on decision tree algorithm), an XGBoost model (eXtreme Gradient Boosting, an optimized distributed gradient boosting library), a deep neural network model, etc. The present application is not limited in this regard.

[0060] It should be noted that the initial interest rate models corresponding to each first historical sample set can be the same (i.e., the model parameters in each initial interest rate model can be the same), or different (i.e., the model parameters in each initial interest rate model can be different), etc. The present application is not limited in this regard.

[0061] In the embodiment of the present application, the electronic device can train the sample space of each first historical sample set through the initial interest rate model corresponding to each first historical sample set, thereby obtaining the single interest rate model corresponding to each first historical sample set.

[0062] S303, each single interest rate model is called to predict the credit utilization rate of each first random interest rate sample in the first random interest rate sample set, and the predicted credit utilization rate of each first random interest rate sample under each single interest rate model is obtained.

[0063] Any single interest rate model is used to predict the credit utilization rate of any sample under the interest rate information corresponding to any single interest rate model, and the credit utilization rate of any sample refers to the probability of performance of the object corresponding to any sample under the corresponding interest rate information. In the embodiment of the present application, the predicted credit utilization rate under any single interest rate model can also be referred to as the single interest rate model score under the single interest rate model, and correspondingly, the predicted credit utilization rate of a random interest rate sample under each single interest rate model can also be referred to as the single interest rate model score of the corresponding random interest rate sample.

[0064] In a specific implementation, the electronic device can use the predicted confidence of each single interest rate model for each first random interest rate sample as a feature of the second layer model in the interest rate sensitive model to determine the first random interest rate model and the second random interest rate model, and the determination process of the first random interest rate model and the second random interest rate model can be seen from the following description, which will not be described here in detail.

[0065] In S304, the first random interest rate model is determined based on the predicted confidence of each single interest rate model for each first random interest rate sample.

[0066] In a specific implementation, the electronic device can use the predicted confidence of each single interest rate model for each first random interest rate sample to determine the first random training data corresponding to each first random interest rate sample respectively, and one first random training data includes the predicted confidence of the corresponding first random interest rate sample in each single interest rate model; then, the electronic device can use the first random training data corresponding to each first random interest rate sample to perform model training on the initial first random interest rate model to obtain the initial first random interest rate model after model training, and determine the first random interest rate model based on the initial first random interest rate model after model training. For example, assuming that the value of N is 3, the predicted confidence of each single interest rate model for one first random interest rate sample can be predicted confidence 1, predicted confidence 2 and predicted confidence 3 respectively, in which case, the first random training data corresponding to the first random interest rate sample can include predicted confidence 1, predicted confidence 2 and predicted confidence 3, and be used as a feature of the initial first random interest rate model in training the second layer model.

[0067] The initial first random interest rate model can be an LR (Logistics Regression, Logistics Regression) model (i.e. a generalized linear regression method, generally used for classification tasks such as classification tasks of determining whether an object is overdue), a deep neural network model (i.e. a technology in the field of machine learning (Machine Learning, ML), which generally refers to a method of using a multi-level network structure for prediction), a decision tree model (i.e. a simple and easy-to-use non-parametric classifier, which does not require any prior assumptions about the data, has fast computing speed, easy-to-interpret results, and strong robustness), etc. The present application does not limit this. It should be understood that the models in the embodiments of the present application are all supervised learning, and the so-called supervised learning is a machine learning task of inferring a function from a labeled training data set, that is, the model can be trained based on the labeled categories.

[0068] Specifically, when the initial first random interest rate model is trained based on the first random training data corresponding to each first random interest rate sample to obtain a trained initial first random interest rate model, the electronic device can use the trained initial first random interest rate model to predict the creditworthiness of the first random training data corresponding to each first random interest rate sample to obtain a predicted creditworthiness of each first random interest rate sample, and determine a predicted category of each first random interest rate sample based on the predicted creditworthiness of each first random interest rate sample. Then, the electronic device can calculate an initial first random interest rate model loss value of the initial first random interest rate model based on the predicted category of each first random interest rate sample and the labeled category of the corresponding first random interest rate sample, and train the model parameters in the initial first random interest rate model in a direction of reducing the initial first random interest rate model loss value to obtain a trained initial first random interest rate model. The predicted creditworthiness of a sample predicted by a random interest rate model (such as the first random interest rate model or the second random interest rate model) can also be referred to as a random interest rate model score (such as a first random interest rate model score or a second random interest rate model score).

[0069] In the embodiments of the present application, when the first random interest rate model is determined based on the trained initial first random interest rate model, the electronic device can continue to train the trained initial first random interest rate model until the first random interest rate model converges, and the random interest rate model that meets the first random interest rate model convergence condition is used as the first random interest rate model. Alternatively, the first random interest rate model convergence condition can refer to that the number of iterations reaches a first random interest rate model iteration threshold, or that the first random interest rate model loss value is less than a first random interest rate model loss threshold, and the present application does not limit this. Correspondingly, the first random interest rate model iteration threshold and the first random interest rate model loss threshold can be set according to experience or actual needs, and the present application does not limit this.

[0070] S305, respectively calling each single interest rate model, predicting the creditworthiness of each second random interest rate sample to obtain the predicted creditworthiness of each second random interest rate sample under each single interest rate model.

[0071] S306, determining the second random interest rate model based on the predicted creditworthiness of each second random interest rate sample under each single interest rate model.

[0072] In a specific implementation, the electronic device can determine a second random training data corresponding to each second random interest rate sample based on the predicted confidence of each second random interest rate sample under each single interest rate model, where one second random training data includes the predicted confidence of the corresponding second random interest rate sample under each single interest rate model; then, the electronic device can perform model training on the initial second random interest rate model based on the second random training data corresponding to each second random interest rate sample, to obtain a model-trained initial second random interest rate model, and determine the second random interest rate model based on the model-trained initial second random interest rate model.

[0073] The initial second random interest rate model can be an LR (Logistics Regression) model, a deep neural network model, a decision tree model, or the like, which is not limited in the present application. It should be noted that the initial second random interest rate model can be the same as or different from the initial first random interest rate model, which is not limited in the present application. Optionally, since the embodiments of the present application can calculate the interest rate sensitivity of the object corresponding to a sample based on the difference between the predicted confidence of the sample under the first random interest rate model and the predicted confidence under the second random interest rate model, the embodiments of the present application can preferably use the same initial first random interest rate model and initial second random interest rate model to avoid errors caused by model differences, thereby improving the accuracy of the interest rate sensitivity. For example, when the initial first random interest rate model and the initial second random interest rate model are both logistic regression models, the electronic device can learn the interest rate sensitivity based on a double logistic regression model; when the initial first random interest rate model and the initial second random interest rate model are both deep neural network models, the electronic device can learn the interest rate sensitivity based on a double deep neural network model; and when the initial first random interest rate model and the initial second random interest rate model are both decision tree models, the electronic device can learn the interest rate sensitivity based on a double decision tree model.

[0074] Optionally, the first random interest rate sample set can be used to determine the first random interest rate model in the interest rate sensitivity model, and the second random interest rate sample set can be used to determine the second random interest rate model in the interest rate sensitivity model. Since the number of random interest rate samples included in the first random interest rate sample set and the second random interest rate sample set is small, the embodiments of the present application can preferably use logistic regression models for the initial first random interest rate model and the initial second random interest rate model, i.e., the first random interest rate model and the second random interest rate model are both logistic regression models. That is, the embodiments of the present application can preferably use a model with low complexity to construct a random interest rate model. In this case, the embodiments of the present application can learn the interest rate sensitivity based on a double logistic regression model, which solves the problem that the interest rate sensitivity cannot be learned using a supervised learning method, and solves the problem of a small number of modeling samples through a transfer learning method.

[0075] Specifically, when the initial second random interest rate model is trained based on the second random training data corresponding to each second random interest rate sample to obtain a model-trained initial second random interest rate model, the electronic device can call the initial second random interest rate model to perform credit rating prediction on the second random training data corresponding to each second random interest rate sample to obtain a predicted credit rating of each second random interest rate sample, and determine a predicted category of each second random interest rate sample based on the predicted credit rating of each second random interest rate sample. Then, the electronic device can calculate an initial second random interest rate model loss value of the initial second random interest rate model based on the predicted category of each second random interest rate sample and the labeled category of the corresponding second random interest rate sample, and perform model training on the model parameters in the initial second random interest rate model in a direction of reducing the initial second random interest rate model loss value to obtain a model-trained initial second random interest rate model.

[0076] It should be noted that when the second random interest rate model is determined based on the model-trained initial second random interest rate model, the electronic device can continue to perform model training on the model-trained initial second random interest rate model until the second random interest rate model convergence condition is met, and the random interest rate model that meets the second random interest rate model convergence condition is taken as the second random interest rate model. Optionally, the second random interest rate model convergence condition can refer to that the number of iterations reaches a second random interest rate model iteration threshold, or that the second random interest rate model loss value is less than a second random interest rate model loss threshold, etc., which is not limited by the present application. Correspondingly, the second random interest rate model iteration threshold and the second random interest rate model loss threshold can be set according to experience or actual needs, which is not limited by the present application.

[0077] In summary, the electronic device can train N single interest rate models through the N first historical sample sets and the labeled categories of each first historical sample; then, call each single interest rate model to perform trust rate prediction on each first random interest rate sample to obtain the predicted trust rate of each first random interest rate sample under each single interest rate model (i.e., single interest rate model score), and call each single interest rate model to perform trust rate prediction on each second random interest rate sample to obtain the predicted trust rate of each second random interest rate sample under each single interest rate model. Then, determine the first random training data corresponding to each first random interest rate sample by using the predicted trust rate of each first random interest rate sample under each single interest rate model, and determine the second random training data corresponding to each second random interest rate sample by using the predicted trust rate of each second random interest rate sample under each single interest rate model, to realize the migration of the predicted trust rate under each single interest rate model to the corresponding random interest rate sample. Further, the initial first random interest rate model can be trained based on the first random training data corresponding to each first random interest rate sample (i.e., the random interest rate sample migrated by the single interest rate model score) and the labeled category of each first random interest rate sample, and the initial second random interest rate model can be trained based on the second random training data corresponding to each second random interest rate sample and the labeled category of each second random interest rate sample, to realize fine-tuning on a small amount of historical random pricing samples (i.e., random interest rate samples), thereby realizing the model training of the second layer model in the interest rate sensitive model, that is, the N single interest rate model scores corresponding to the random interest rate sample can be used as features to fit the response under the random interest rate sample, thereby completing the feature-level transfer learning, such as Figure 4

[0078] It should be understood that Figure 4 The transfer learning is only exemplarily represented, and the present application is not limited thereto; for example, the N single interest rate models can be further included between the single interest rate model score and the trust category, that is, the single interest rate model score is obtained by each single interest rate model; for another example, the random interest rate model can be further included after the random interest rate sample set and the trust category, and the like.

[0079] S307, adopt each single interest rate model, the first random interest rate model and the second random interest rate model to construct the interest rate sensitive model.

[0080] ​In the embodiments of the present application, when the interest rate sensitive model comprises the respective single interest rate model, the first random interest rate model and the second random interest rate model, the electronic device can acquire the object prediction sample corresponding to the target object, and respectively invoke the respective single interest rate model to perform the credit rating prediction on the object prediction sample, to obtain the predicted credit rating of the object prediction sample under the respective single interest rate model, and determine the target prediction sample corresponding to the object prediction sample by using the predicted credit rating of the object prediction sample under the respective single interest rate model. Then, the electronic device can invoke the first random interest rate model to perform the credit rating prediction on the target prediction sample to obtain the first predicted credit rating of the target prediction sample, and invoke the second random interest rate model to perform the credit rating prediction on the target prediction sample to obtain the second predicted credit rating of the target prediction sample, so as to calculate the interest rate sensitivity of the target object by using the difference between the first predicted credit rating and the second predicted credit rating, such as Figure 5

[0081] For example, assuming that the N single interest rate models comprise the single interest rate model 1, the single interest rate model 2 and the single interest rate model 3, the electronic device can obtain the predicted credit rating 1 of the object prediction sample under the single interest rate model 1, the predicted credit rating 2 of the object prediction sample under the single interest rate model 2 and the predicted credit rating 3 of the object prediction sample under the single interest rate model 3, to obtain the target prediction sample, and the target prediction sample comprises the predicted credit rating 1, the predicted credit rating 2 and the predicted credit rating 3; and assuming that the first predicted credit rating is 0.3 and the second predicted credit rating is 0.8, the electronic device can calculate the interest rate sensitivity of the target object as 0.5, and so on.

[0082] It should be noted that the electronic device can also determine the target interest rate information for the target object according to the interest rate sensitivity of the target object; wherein the target interest rate information can decrease as the interest rate sensitivity of the target object increases, that is, when the interest rate sensitivity of the target object is large, the electronic device can determine a smaller target interest rate information for the target object, and when the interest rate sensitivity of the target object is small, the electronic device can determine a larger target interest rate information for the target object. Optionally, the electronic device can construct an interest rate function by using the plurality of interest rate sensitivities and the interest rate information corresponding to each interest rate sensitivity in the plurality of interest rate sensitivities, and calculate the target interest rate information by using the interest rate function and the interest rate sensitivity of the target object; or the electronic device can divide the interest rate sensitivity range into a plurality of interest rate sensitivity sub-ranges, and one interest rate information corresponds to one interest rate sensitivity sub-range, based on which the electronic device can take the interest rate information corresponding to the interest rate sensitivity sub-range to which the interest rate sensitivity of the target object belongs as the target interest rate information, and so on; the present application does not make any limitation in this regard.

[0083] ​The embodiment of the present application can determine N single interest rate models by using each of the N first historical sample sets after obtaining the N first historical sample sets and obtaining the second historical sample set, each of the N single interest rate models corresponds to each of the first historical sample sets, and the second historical sample set includes the first random interest rate sample set and the second random interest rate sample set. Then, each of the single interest rate models is called to perform interest rate prediction on each of the first random interest rate samples in the first random interest rate sample set to obtain the predicted interest rate of each of the first random interest rate samples under each of the single interest rate models, and the first random interest rate model is determined based on the predicted interest rate of each of the first random interest rate samples under each of the single interest rate models; and each of the single interest rate models is called to perform interest rate prediction on each of the second random interest rate samples to obtain the predicted interest rate of each of the second random interest rate samples under each of the single interest rate models, and the second random interest rate model is determined based on the predicted interest rate of each of the second random interest rate samples under each of the single interest rate models; based on this, the interest rate sensitive model can be constructed by using each of the single interest rate models, the first random interest rate model and the second random interest rate model, and the interest rate sensitive model is used to predict the interest rate sensitivity of the target object. It can be seen that the embodiment of the present application can reuse each of the single interest rate models in the construction process of the first random interest rate model and the second random interest rate model, so that the model performance of the interest rate sensitive model can be effectively improved by correcting and adjusting a small amount of historical random pricing samples, and the characteristic information under a large amount of data can also be learned, so that the interest rate sensitivity of the target object can be accurately obtained through the interest rate sensitive model, that is, the embodiment of the present application can improve the accuracy of the interest rate sensitivity.

[0084] Based on the description of the related embodiments of the above model construction method, the embodiment of the present application further proposes a model construction device, which can be a computer program (including program code) running in an electronic device; as shown in the Figure 6 , the model construction device can include an acquisition unit 601 and a processing unit 602. The model construction device can perform the model construction method shown in the Figure 1 or Figure 3 , that is, the model construction device can run the above units:

[0085] The acquisition unit 601 is used to acquire N first historical sample sets and acquire a second historical sample set, the first historical samples in the same first historical sample set correspond to the same interest rate information, and the interest rate information corresponding to one second historical sample is randomly generated interest rate information, N is a positive integer;

[0086] determine, by using each of the N first historical sample sets respectively, N single interest rate models, each of the N single interest rate models corresponding to one of the first historical sample sets respectively;

[0087] The processing unit 602 is further configured to construct an interest rate sensitivity model based on the second historical sample set and the single interest rate models, the interest rate sensitivity model being used to predict the interest rate sensitivity of the target object.

[0088] In an embodiment, the second historical sample set includes a first random interest rate sample set and a second random interest rate sample set, and the interest rate information corresponding to any first random interest rate sample in the first random interest rate sample set is greater than or equal to the interest rate information corresponding to each second random interest rate sample in the second random interest rate sample set; when the processing unit 602 constructs the interest rate sensitivity model based on the second historical sample set and the single interest rate models, it can be specifically configured to:

[0089] invoke each of the single interest rate models respectively to perform credit usage rate prediction on each of the first random interest rate samples in the first random interest rate sample set, to obtain the predicted credit usage rate of each of the first random interest rate samples under each of the single interest rate models, any single interest rate model being used to predict the credit usage rate of any sample under the interest rate information corresponding to the single interest rate model, and the credit usage rate of the sample being the probability of the object corresponding to the sample performing the performance under the corresponding interest rate information;

[0090] determine the first random interest rate model based on the predicted credit usage rate of each of the first random interest rate samples under each of the single interest rate models;

[0091] invoke each of the single interest rate models respectively to perform credit usage rate prediction on each of the second random interest rate samples, to obtain the predicted credit usage rate of each of the second random interest rate samples under each of the single interest rate models;

[0092] determine the second random interest rate model based on the predicted credit usage rate of each of the second random interest rate samples under each of the single interest rate models;

[0093] construct the interest rate sensitivity model by using each of the single interest rate models, the first random interest rate model, and the second random interest rate model.

[0094] In another embodiment, when the processing unit 602 determines the first random interest rate model based on the predicted credit usage rate of each of the first random interest rate samples under each of the single interest rate models, it can be specifically configured to:

[0095] Using the predicted credit rates of the respective first random interest rate samples under the respective single interest rate models, first random training data corresponding to the respective first random interest rate samples are determined, each first random training data including the predicted credit rates of the corresponding first random interest rate samples under the respective single interest rate models;

[0096] Performing model training on an initial first random interest rate model using the first random training data corresponding to each of the first random interest rate samples to obtain a trained initial first random interest rate model, and determining a first random interest rate model based on the trained initial first random interest rate model;

[0097] When determining the second random interest rate model based on the predicted credit rates of the second random interest rate samples under the single interest rate models, the processing unit 602 may be specifically configured to:

[0098] Using the predicted credit rates of the respective second random interest rate samples under the respective single interest rate models, respectively determine second random training data corresponding to the respective second random interest rate samples, where one piece of second random training data includes the predicted credit rates of the corresponding second random interest rate samples under the respective single interest rate models;

[0099] The initial second random interest rate model is trained using the second random training data corresponding to each of the second random interest rate samples to obtain an initial second random interest rate model after model training, and the second random interest rate model is determined based on the initial second random interest rate model after model training.

[0100] In another embodiment, when the processing unit 602 uses each of the N first historical sample sets to determine the N single interest rate models, it can be specifically configured to:

[0101] For the i-th first historical sample set among the N first historical sample sets, calling the initial interest rate model corresponding to the i-th first historical sample set, performing credit rate prediction on each first historical sample in the i-th first historical sample set, obtaining a predicted credit rate for each first historical sample, and determining a prediction category for each first historical sample based on the predicted credit rate for each first historical sample, i∈[1,N];

[0102] Calculating an initial interest rate model loss value of the initial interest rate model based on the predicted category of each first historical sample and the labeled category of the corresponding first historical sample;

[0103] Optimize the model parameters in the initial interest rate model in the direction of reducing the loss value of the initial interest rate model, so as to determine the single interest rate model corresponding to the i-th first historical sample set based on the initial interest rate model after model optimization.

[0104] In another implementation, the second historical sample set includes a number of second historical samples determined based on a preset sample ratio and a target data volume, the target data volume including any one of a number of first historical samples in the N first historical sample sets and a target business data volume of the target business within a target time range, and the preset sample ratio being less than a sample ratio threshold.

[0105] In another implementation, the second historical sample set includes a first random interest rate sample set and a second random interest rate sample set, the first random interest rate sample set being used to determine a first random interest rate model in the interest rate sensitive model, the second random interest rate sample set being used to determine a second random interest rate model in the interest rate sensitive model, and the first random interest rate model and the second random interest rate model both being logistic regression models.

[0106] In another implementation, the interest rate sensitive model includes the respective single interest rate models, the first random interest rate model, and the second random interest rate model, and the obtaining unit 601 is further configured to:

[0107] obtain an object prediction sample corresponding to the target object;

[0108] The processing unit 602 is further configured to:

[0109] invoke the respective single interest rate models respectively to perform interest rate prediction on the object prediction sample to obtain a predicted interest rate of the object prediction sample under the respective single interest rate models;

[0110] determine a target prediction sample corresponding to the object prediction sample by using the predicted interest rates of the object prediction sample under the respective single interest rate models;

[0111] invoke the first random interest rate model to perform interest rate prediction on the target prediction sample to obtain a first predicted interest rate of the target prediction sample, and invoke the second random interest rate model to perform interest rate prediction on the target prediction sample to obtain a second predicted interest rate of the target prediction sample;

[0112] calculate the interest rate sensitivity of the target object by using a difference between the first predicted interest rate and the second predicted interest rate.

[0113] According to one embodiment of the present application, Figure 1 or Figure 3 the steps involved in the method shown in the figure can be performed by the respective units in the model construction apparatus shown in the figure. For example, Figure 6 the step S101 shown in the figure can be performed by the obtaining unit 601 shown in the figure. Figure 1 Figure 6 ​The acquisition unit 601 shown in the above embodiment performs, the steps S102 and S103 can be performed by Figure 6 The processing unit 602 shown in the above embodiment performs. Also, the step S301 shown in the above embodiment can be performed by Figure 3 The processing unit 602 shown in the above embodiment performs. Also, the step S301 shown in the above embodiment can be performed by Figure 6 The acquisition unit 601 shown in the above embodiment performs, the steps S302-S307 can be performed by Figure 6 The processing unit 602 shown in the above embodiment performs, etc.

[0114] According to another embodiment of the present application, Figure 6 The units in the model construction apparatus shown in the above embodiment can be respectively or all combined into one or several other units to constitute, or some of the units can be further split into a plurality of units with smaller functions to constitute, which can achieve the same operation without affecting the implementation of the technical effects of the embodiments of the present application. The above units are divided based on logical functions, in actual application, the function of one unit can also be implemented by multiple units, or the functions of multiple units are implemented by one unit. In other embodiments of the present application, any model construction apparatus can also include other units, in actual application, these functions can also be implemented by other units, and can be implemented by multiple units.

[0115] According to another embodiment of the present application, the model construction apparatus as shown in the above embodiment, and the model construction method of the embodiments of the present application can be constructed by running a computer program (including program codes) related to the steps of the corresponding method shown in the above embodiment on a general electronic device such as a computer including processing elements and storage elements such as a central processing unit (CPU), a random access memory (RAM), a read-only memory (ROM), etc. Figure 1 Or Figure 3 The model construction apparatus as shown in the above embodiment, and the model construction method of the embodiments of the present application can be constructed by running a computer program (including program codes) related to the steps of the corresponding method shown in the above embodiment on a general electronic device such as a computer including processing elements and storage elements such as a central processing unit (CPU), a random access memory (RAM), a read-only memory (ROM), etc. Figure 6 The model construction apparatus as shown in the above embodiment, and the model construction method of the embodiments of the present application can be constructed by running a computer program (including program codes) related to the steps of the corresponding method shown in the above embodiment on a general electronic device such as a computer including processing elements and storage elements such as a central processing unit (CPU), a random access memory (RAM), a read-only memory (ROM), etc.

[0116] The embodiment of the present application can determine N single interest rate models by using each first historical sample set in the N first historical sample sets respectively after obtaining the N first historical sample sets and obtaining the second historical sample set, the first historical samples in the same first historical sample set correspond to the same interest rate information, the interest rate information corresponding to one second historical sample is randomly generated interest rate information, each single interest rate model in the N single interest rate models corresponds to each first historical sample set one by one, N is a positive integer; then, the interest rate sensitive model is constructed based on the second historical sample set and each single interest rate model, and the interest rate sensitive model is used to predict the interest rate sensitivity of the target object. It can be seen that the embodiment of the present application can construct the interest rate sensitive model for the interest rate pricing, so as to obtain the interest rate sensitivity of the target object through the interest rate sensitive model, and to use the interest rate sensitivity of the target object to price the target object more differently, that is, the embodiment of the present application can determine the appropriate interest rate information for the target object based on the interest rate sensitivity of the target object. At the same time, the embodiment of the present application can learn important information on the N first historical sample sets based on the feature construction method of the transfer learning, and fine-tune on the second historical sample set, which has great significance for the construction of the interest rate sensitive model feature information, can effectively improve the model performance of the interest rate sensitive model, so as to accurately obtain the interest rate sensitivity of the target object through the interest rate sensitive model, that is, the embodiment of the present application can improve the accuracy of the interest rate sensitivity.

[0117] Based on the description of the method embodiment and the device embodiment, the exemplary embodiment of the present application further provides an electronic device, comprising: at least one processor; and a memory connected with the at least one processor in communication. The memory stores a computer program capable of being executed by the at least one processor, and the computer program is used to make the electronic device execute the method according to the embodiment of the present application when executed by the at least one processor.

[0118] The exemplary embodiment of the present application further provides a non-transitory computer readable storage medium storing a computer program, wherein the computer program is used to make the computer execute the method according to the embodiment of the present application when executed by the processor of the computer.

[0119] The exemplary embodiment of the present application further provides a computer program product comprising a computer program, wherein the computer program is used to make the computer execute the method according to the embodiment of the present application when executed by the processor of the computer.

[0120] Reference Figure 7, a block diagram of an electronic device 700 that can serve as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, 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 invention described and / or required herein.

[0121] like Figure 7 As shown, electronic device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. Various programs and data required for the operation of device 700 can also be stored in RAM 703. Computing unit 701, ROM 702, and RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to bus 704.

[0122] Multiple components within electronic device 700 are connected to I / O interface 705, including an input unit 706, an output unit 707, a storage unit 708, and a communication unit 709. Input unit 706 can be any type of device capable of inputting information into electronic device 700. Input unit 706 can receive input numeric or character information and generate key input signals related to user settings and / or function control of the electronic device. Output unit 707 can be any type of device capable of presenting information and may include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. Storage unit 708 may include, but is not limited to, a magnetic disk or an optical disk. Communication unit 709 allows electronic device 700 to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunication networks and may include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver and / or a chipset, such as a Bluetooth™ device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0123] The computing unit 701 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs various methods and processes described above. For example, in some embodiments, the model construction method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, portions or all of the computer program can be loaded and / or installed onto the electronic device 700 via the ROM 702 and / or the communication unit 709. In some embodiments, the computing unit 701 can be configured to perform the model construction method by any other suitable means, such as by means of firmware.

[0124] Program code for carrying out the methods of the present application can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0125] In the context of the present application, a machine-readable medium can be a tangible medium that can contain or store program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable storage media can include, without limitation, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0126] As used in the present disclosure, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus and / or device (e.g., magnetic discs, optical disks, memory, Programmable Logic Devices (PLDs)) used to provide machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal that can be used to provide machine instructions and / or data to a programmable processor.

[0127] To provide for interaction with a user, the systems and techniques described here 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 a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, 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, speech, or tactile input.

[0128] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, 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.

[0129] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

[0130] And, it should be understood that all the disclosed and / or claimed combinations can include additional elements, and that not all permutations or combinations of features are recited or claimed.

Claims

1. A model construction method characterized by comprising: The method comprises the following steps: obtaining N first historical sample sets and a second historical sample set, the first historical samples in the same first historical sample set correspond to the same interest rate information, and the interest rate information corresponding to each second historical sample is randomly generated interest rate information, N is a positive integer; the second historical sample set comprises a first random interest rate sample set and a second random interest rate sample set, and the interest rate information corresponding to any first random interest rate sample in the first random interest rate sample set is greater than or equal to the interest rate information corresponding to each second random interest rate sample in the second random interest rate sample set; determining N single interest rate models by using each first historical sample set in the N first historical sample sets respectively, each single interest rate model in the N single interest rate models corresponds to the corresponding first historical sample set; constructing an interest rate sensitive model based on the second historical sample set and the single interest rate models, comprising: calling each single interest rate model respectively to perform credit utilization rate prediction on each first random interest rate sample in the first random interest rate sample set, to obtain the predicted credit utilization rate of each first random interest rate sample under each single interest rate model, any single interest rate model is used to predict the credit utilization rate of any sample under the interest rate information corresponding to the single interest rate model, and the credit utilization rate of the sample refers to the probability of performance of the object corresponding to the sample under the corresponding interest rate information; determining a first random interest rate model based on the predicted credit utilization rate of each first random interest rate sample under each single interest rate model; calling each single interest rate model respectively to perform credit utilization rate prediction on each second random interest rate sample, to obtain the predicted credit utilization rate of each second random interest rate sample under each single interest rate model; determining a second random interest rate model based on the predicted credit utilization rate of each second random interest rate sample under each single interest rate model; constructing an interest rate sensitive model by using each single interest rate model, the first random interest rate model and the second random interest rate model; the interest rate sensitive model is used to predict the interest rate sensitivity of a target object.

2. The method of claim 1, wherein, The method comprises the following steps: determining first random training data corresponding to each first random interest rate sample by using the predicted credit utilization rate of each first random interest rate sample under each single interest rate model respectively, one first random training data comprises the predicted credit utilization rate of the corresponding first random interest rate sample under each single interest rate model; performing model training on an initial first random interest rate model by using the first random training data corresponding to each first random interest rate sample, to obtain a model trained initial first random interest rate model, and determining a first random interest rate model based on the model trained initial first random interest rate model; The method comprises the following steps: The prediction confidence of each second random interest rate sample under each single interest rate model is used to determine a second random training data corresponding to each second random interest rate sample, and one second random training data includes the prediction confidence of the corresponding second random interest rate sample under each single interest rate model. The initial second random interest rate model is trained using the second random training data corresponding to each second random interest rate sample to obtain a model-trained initial second random interest rate model, and a second random interest rate model is determined based on the model-trained initial second random interest rate model.

3. The method according to any of claims 1-2, characterized in that, The method further includes: For the i-th first historical sample set in the N first historical sample sets, the initial interest rate model corresponding to the i-th first historical sample set is called to perform confidence prediction on each first historical sample in the i-th first historical sample set to obtain the prediction confidence of each first historical sample, and the prediction category of each first historical sample is determined based on the prediction confidence of each first historical sample, i∈[1,N]; An initial interest rate model loss value of the initial interest rate model is calculated based on the prediction category of each first historical sample and the labeled category of the corresponding first historical sample; The model parameters in the initial interest rate model are optimized in the direction of reducing the initial interest rate model loss value to determine the single interest rate model corresponding to the i-th first historical sample set based on the model-optimized initial interest rate model.

4. The method according to any one of claims 1-2, characterized in that, The number of second historical samples included in the second historical sample set is determined based on a preset sample proportion and a target data amount, the target data amount includes any one of the number of first historical samples in the N first historical sample sets and the business data amount of the target business handled within a target time range, and the preset sample proportion is less than a sample proportion threshold.

5. The method of claim 4, wherein, The second historical sample set includes a first random interest rate sample set and a second random interest rate sample set, the first random interest rate sample set is used to determine a first random interest rate model in the interest rate sensitive model, the second random interest rate sample set is used to determine a second random interest rate model in the interest rate sensitive model, and the first random interest rate model and the second random interest rate model are both logistic regression models.

6. The method according to any one of claims 1-2, characterized in that, The interest rate sensitive model includes the each single interest rate model, the first random interest rate model, and the second random interest rate model, and the method further includes: An object prediction sample corresponding to the target object is obtained; Each single interest rate model is called to perform confidence prediction on the object prediction sample to obtain the prediction confidence of the object prediction sample under each single interest rate model; The target prediction sample corresponding to the object prediction sample is determined using the prediction confidence of the object prediction sample under each single interest rate model. The first random interest rate model is called to perform creditworthiness prediction on the target prediction sample, to obtain a first predicted creditworthiness of the target prediction sample, and the second random interest rate model is called to perform creditworthiness prediction on the target prediction sample, to obtain a second predicted creditworthiness of the target prediction sample; The difference between the first predicted creditworthiness and the second predicted creditworthiness is used to calculate the interest rate sensitivity of the target object.

7. A model building device characterized by comprising: The device comprises: An acquisition unit is configured to acquire N first historical sample sets and a second historical sample set, a first historical sample in a same first historical sample set corresponds to same interest rate information, and interest rate information corresponding to a second historical sample is randomly generated interest rate information, N is a positive integer; the second historical sample set comprises a first random interest rate sample set and a second random interest rate sample set, and interest rate information corresponding to any first random interest rate sample in the first random interest rate sample set is greater than or equal to interest rate information corresponding to each second random interest rate sample in the second random interest rate sample set; A processing unit is configured to determine N single interest rate models by using each first historical sample set in the N first historical sample sets respectively, and each single interest rate model in the N single interest rate models corresponds to the each first historical sample set one by one; The processing unit is further configured to include the following based on the second historical sample set and the each single interest rate model: Each single interest rate model is called to perform creditworthiness prediction on each first random interest rate sample in the first random interest rate sample set, to obtain predicted creditworthiness of the each first random interest rate sample under the each single interest rate model, any single interest rate model is used to predict creditworthiness of any sample under interest rate information corresponding to the any single interest rate model, and the creditworthiness of the any sample refers to a probability of an object corresponding to the any sample performing performance under corresponding interest rate information; A first random interest rate model is determined based on the predicted creditworthiness of the each first random interest rate sample under the each single interest rate model; Each second random interest rate sample is predicted by calling the each single interest rate model, to obtain predicted creditworthiness of the each second random interest rate sample under the each single interest rate model; A second random interest rate model is determined based on the predicted creditworthiness of the each second random interest rate sample under the each single interest rate model; The each single interest rate model, the first random interest rate model and the second random interest rate model are used to construct an interest rate sensitive model. An interest rate sensitive model is constructed, and the interest rate sensitive model is used to predict interest rate sensitivity of a target object.

8. An electronic device, comprising: Comprise: A processor; And A memory for storing programs, Wherein, the program includes instructions, which, when executed by the processor, cause the processor to perform the method according to any one of claims 1-6.

9. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to make the computer perform the method according to any one of claims 1-6.

Images