Transfer learning method of credit evaluation model and credit evaluation method

Abstract

The invention provides a transfer learning method of a credit evaluation model and a credit evaluation method. The transfer learning method of the credit evaluation model comprises the steps of obtaining sample data; mapping the source domain data and the target domain data with the common features to the same feature dimension, determining a mapping relationship between the source domain data and the feature dimension and a mapping relationship between the target domain data and the feature dimension, and constructing a feature space for the newly added industry based on the determined mapping relationship; inputting any sample data into a feature extraction network of a credit investigation evaluation model obtained by training the source domain data, so as to determine credit investigation features of the training sample in a feature space through the feature extraction network; and determining a prediction domain classification corresponding to the sample data according to the credit investigation features, and adjusting network parameters of the feature extraction network according to a deviation between the prediction domain classification and an actual domain classification of the training sample, so as to determine a migrated credit investigation evaluation model based on the adjusted feature extraction network.

Description

Technical Field

[0001] This specification relates to one or more embodiments in the field of computer technology, and in particular to a transfer learning method for a credit assessment model and a credit assessment method. Background Technology

[0002] When a new industry emerges, it is often necessary to rebuild the credit assessment model for that industry. Since new industries generally lack a mature modeling foundation and lack clear credit labels as in traditional lending industries, it is difficult to achieve effective model training. Furthermore, the repetitive modeling and manual processing of relevant data features further result in low overall assessment efficiency and make it difficult to adapt to the needs of new industries. Summary of the Invention

[0003] In view of the above, one or more embodiments of this specification provide the following technical solutions: According to a first aspect of one or more embodiments of this specification, a transfer learning method for a credit assessment model is proposed, comprising: Acquire sample data; wherein, the sample data is credit data, including source domain data corresponding to existing industries and target domain data corresponding to newly added industries; Source domain data and target domain data with common features are mapped to the same feature dimension, and the mapping relationship between the source domain data and target domain data and the feature dimension is determined respectively. Based on the determined mapping relationship, a feature space for the new industry is constructed. Any sample data is input into the feature extraction network of the credit assessment model trained from the source domain data, so as to determine the credit features of the training sample in the feature space through the feature extraction network; Based on the credit characteristics, the predicted domain classification corresponding to the sample data is determined, and the network parameters of the feature extraction network are adjusted according to the deviation between the predicted domain classification and the actual domain classification of the training sample, so as to determine the transferred credit assessment model based on the adjusted feature extraction network; the transferred credit assessment model is applied to the credit assessment of the newly added industry.

[0004] According to a second aspect of one or more embodiments of this specification, a credit assessment method is proposed, comprising: Obtain credit data for the target industry; The credit data is input into a pre-trained credit assessment model to obtain the credit assessment result output by the credit assessment model. The credit assessment model is trained using the transfer learning method of the aforementioned credit assessment model.

[0005] According to a third aspect of one or more embodiments of this specification, a training apparatus for a credit assessment model is provided, comprising: The acquisition module is used to acquire sample data; wherein, the sample data is credit data, including source domain data corresponding to existing industries and target domain data corresponding to newly added industries; The construction module is used to map source domain data and target domain data with common features to the same feature dimension, determine the mapping relationship between the source domain data and target domain data and the feature dimension respectively, and construct a feature space for the newly added industry based on the determined mapping relationship. The input module is used to input any sample data into the feature extraction network of the credit assessment model trained by the source domain data, so as to determine the credit features of the training sample in the feature space through the feature extraction network; The transfer module is used to determine the predicted domain classification corresponding to the sample data based on the credit features, and to adjust the network parameters of the feature extraction network based on the deviation between the predicted domain classification and the actual domain classification of the training sample, so as to determine the transferred credit assessment model based on the adjusted feature extraction network; the transferred credit assessment model is applied to the credit assessment of the newly added industry.

[0006] According to a fourth aspect of one or more embodiments of this specification, a credit assessment apparatus is provided, comprising: The acquisition module is used to acquire credit data for the target industry; The evaluation module is used to input the credit data into a pre-trained credit evaluation model and obtain the credit evaluation result output by the credit evaluation model.

[0007] According to a fifth aspect of one or more embodiments of this specification, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor implements the steps of the method described above by executing the executable instructions.

[0008] According to a sixth aspect of one or more embodiments of this specification, a computer-readable storage medium is provided that stores computer instructions thereon, which, when executed by a processor, implement the steps of the method described above.

[0009] According to a seventh aspect of one or more embodiments of this specification, a computer program product is provided, comprising a computer program / instructions that, when executed by a processor, implement the steps of the method described above.

[0010] As can be seen from the above embodiments, in the process of training the credit assessment model for a new industry, this specification mines the common features in the source domain data of the existing industry and the target domain data of the new industry, mapping them to the same feature dimension. This generates a mapping relationship between the source domain data and the target domain data under each feature dimension, and constructs a feature space for the new industry based on this mapping relationship. In this way, the credit features of the source domain data and the credit features of the target domain data can be unified into the same feature space. The deviation between the predicted domain classification determined by any sample data and the actual domain classification of the training sample is used to train and optimize the model parameters of the existing industry model. This allows for the reuse of the credit assessment capabilities of the existing industry model, enabling targeted optimization for the new industry without relying on the labeled samples of the new industry, further improving the deployment efficiency and scenario adaptability of the credit assessment model. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the architecture of a transfer learning service system provided in an exemplary embodiment; Figure 2 This is a schematic diagram of the transfer learning process for a credit assessment model provided in an exemplary embodiment; Figure 3 This is a schematic diagram of a transfer learning model architecture provided in an exemplary embodiment; Figure 4 This is an exemplary embodiment of a flowchart of a transfer learning method. Figure 5 This is a schematic diagram of a credit assessment method provided in an exemplary embodiment; Figure 6 This is a schematic diagram of the structure of a device provided in an exemplary embodiment; Figure 7 This is a block diagram of a transfer learning device for a credit assessment model provided in an exemplary embodiment; Figure 8 This is a block diagram of a credit assessment device provided in an exemplary embodiment. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0013] Existing credit assessment technologies have revealed numerous insurmountable shortcomings when dealing with diverse scenarios, especially emerging industries, and cannot meet the current development needs of credit reporting services: Because emerging industries and scenarios lack mature modeling foundations and clear credit labels found in traditional lending industries, it's impossible to directly utilize existing labeling systems to train effective credit assessment models. Existing general credit scores haven't been specifically optimized for the industry logic and risk characteristics of particular scenarios, making it difficult to meet the needs of precise scenario-based assessments. Adding new industry scenarios requires starting data collection, sample labeling, feature engineering, and model training from scratch, resulting in extremely high time and manpower costs for repetitive modeling. Even with essentially the same performance behavior across different scenarios, differences in data field names, value ranges, and frequencies mean that existing technologies rely on expert experience for manual feature mapping, easily overlooking key related features. Furthermore, mapping rules are difficult to iterate efficiently as scenarios expand, limiting the scope of credit assessment technology. To simplify, this mainly highlights that when a new industry emerges, a new credit model is often needed, but due to the limited sample size and lack of labels in new industries, effective model training is difficult.

[0014] Based on this, this specification provides a transfer learning approach for credit assessment models. By mining common credit features in source domain data of existing industries and target domain data of new industries, a mapping relationship between source domain data and target domain data under each feature dimension is generated. In this way, a feature space for new industries can be constructed based on this mapping relationship. And based on the mature credit assessment model trained on the source domain data, the transfer and adaptation of the feature space enables the credit assessment model to fully reuse the effective modeling experience and data value of existing industries, thereby improving the deployment efficiency and scenario adaptability of the credit assessment model.

[0015] The technical solutions described in the various embodiments of this application will be explained in detail below with reference to the accompanying drawings.

[0016] Figure 1 This is a schematic diagram of the architecture of a transfer learning service system provided in an exemplary embodiment. For example... Figure 1 As shown, the system may include a server 11, a network 12, and several electronic devices, such as a personal computer (PC) 13, a mobile phone 14, etc.

[0017] Server 11 can be a physical server containing an independent host, or it can be a virtual server hosted in a host cluster. During operation, server 11 can run server-side programs for a certain application to implement the relevant functions of that application. For example, when server 11 runs a transfer learning service program, it can function as a corresponding transfer learning service platform.

[0018] PC13 and mobile phone14 are just some of the types of electronic devices that users can use. In reality, users can obviously also use electronic devices such as tablets, laptops, PDAs (Personal Digital Assistants), wearable devices (such as smart glasses, smartwatches, etc.), etc., and one or more embodiments in this specification do not limit this. During operation, the electronic device can run a client-side program of an application to achieve the relevant functions of that application. For example, when the electronic device runs a transfer learning service program, it can act as a client for that transfer learning service. The aforementioned transfer learning service client application can be launched and run on the electronic device. This client-side program can be a native application installed on the electronic device, or it can be a mini-program, quick app, or other similar form. Of course, when using web technologies such as HTML5 or similar, the relevant functions can be achieved through a page displayed by a browser. This browser can be a standalone browser application or a browser module embedded in some applications.

[0019] As for the network 12 that enables interaction between electronic devices such as PC13 and mobile phone 14 and server 11, communication can be achieved using either wired or wireless networks, depending on the communication methods supported by the respective electronic devices. This specification does not impose any restrictions on this. For example, PC13 can support both wired and wireless communication, so it can use either wired or wireless networks as needed. Mobile phone 14 typically only supports wireless communication, so it can use a wireless network for communication.

[0020] Figure 2 This is a schematic diagram of the transfer learning process for a credit assessment model provided in an exemplary embodiment, including the following steps: S200: Obtain sample data; wherein, the sample data is credit data, including source domain data corresponding to existing industries and target domain data corresponding to newly added industries.

[0021] In this specification, the execution subject for performing transfer learning of the credit assessment model can be a designated device such as a server, or it can be a device such as a mobile phone or computer, or a client installed on these devices. For ease of description, the following will use a server as the execution subject to explain the transfer learning of the credit assessment model provided in this specification.

[0022] The server can obtain sample data, which is credit data. Specifically, it can include source domain data corresponding to existing industries and target domain data corresponding to newly added industries.

[0023] Existing industries and newly added industries can correspond to different industry sectors under the credit reporting service. Existing industries can refer to mature industries that have launched credit reporting service functions and deployed a trained credit assessment model, which is trained based on source domain data of existing industries.

[0024] The newly added industries can be those industries for which corresponding credit reporting service functions have not yet been developed and implemented, and for which corresponding credit assessment models have not yet been trained and deployed.

[0025] Taking the credit card lending industry as an example, its source domain data can include: number of credit card repayments, number of credit card delinquencies, average monthly consumption repayment ratio, etc.

[0026] The newly added industry could be the personal rental credit industry, and its target data could include: rent payment on time rate, number of rental defaults, and monthly rent payment fulfillment time, etc.

[0027] It should be noted that in practical applications, new industries can also include emerging industries such as live-streaming e-commerce, flexible employment, community services, rental housing, and instant delivery, but the training method for the corresponding credit assessment model is the same. This manual will not list them one by one.

[0028] S202: Map source domain data and target domain data with common features to the same feature dimension, and determine the mapping relationship between the source domain data and target domain data and the feature dimension respectively, and construct a feature space for the newly added industry based on the determined mapping relationship.

[0029] The server can mine common credit features between source domain data and target domain data, and perform equivalent mapping on combinations of source domain data and target domain data based on these common credit features. This maps source domain data and target domain data with common credit features to the same feature dimension. In this way, each feature dimension corresponds to a set of source domain data and target domain data with common features, thereby determining the mapping relationship between source domain data and target domain data and the corresponding feature dimension under each feature dimension.

[0030] In practical applications, the server can first determine the mapping relationship between the source domain data and the corresponding feature dimension, and then determine the mapping relationship between the target domain data and the corresponding feature dimension. In this way, there is an equivalence relationship between the source domain data and the target domain data corresponding to the same feature dimension by default.

[0031] Of course, the server can also first determine the equivalence relationship between source domain data and target domain data that have common features, and then anchor the same feature dimension that the two can map based on the equivalence relationship, and then determine the mapping relationship between each and the corresponding feature dimension.

[0032] In this specification, the server can determine the above mapping relationship based on a pre-trained equivalent mapping model (such as a Large Language Model (LLM)).

[0033] The server can construct prompt words based on sample data. These prompt words contain source domain data and target domain data carrying domain labels, as well as instructions for generating mapping relationships. The prompt words are then input into a pre-trained equivalence mapping model, which can be...

[0034] The equivalent mapping model, guided by the prompt word, can further perform feature alignment and commonality matching on the source and target domain data with domain labels, thereby outputting the accurate mapping relationship between the source and target domain data under each feature dimension, and clarifying the common credit feature type corresponding to each mapping relationship.

[0035] For example, if the source domain data is credit card repayment records and the target data is rent payment records, both of which are periodic performance behaviors and have common characteristics, then credit card repayment records and rent payment records can be mapped to the feature dimension of periodic performance behaviors, and a mapping relationship between credit card repayment records and rent payment records can be constructed.

[0036] For example, if the source domain data is personal loan repayment records on time and the target domain data is ride-hailing fare payment records on time, both belong to small-amount, high-frequency financial repayment behaviors and have significant common credit characteristics. Therefore, personal loan repayment records on time and ride-hailing fare payment records on time can be mapped to the feature dimension of small-amount, high-frequency repayment behaviors, and a mapping relationship between personal loan repayment records on time and ride-hailing fare payment records on time can be constructed.

[0037] Furthermore, during the mapping process between source domain data and target domain data, the server can identify multiple credit assessment dimensions that affect the credit assessment results. Then, for each sample data, it determines the degree of correlation between the sample data and each credit assessment dimension. Based on the degree of correlation between each sample data and each credit assessment dimension, it identifies source domain data and target domain data that have commonalities and maps the source domain data and target domain data to the same feature dimension.

[0038] Among them, the credit assessment dimension that has a similar degree of correlation with the corresponding data of the two samples can be used as the same feature dimension for mapping between the two.

[0039] For example, if the credit assessment dimensions affecting the credit assessment results include: credit assessment dimensions A, B, and C, and the correlation between the source domain data s and assessment dimensions A, B, and C is 0.3, 0.5, and 0.8 respectively, and the correlation between the target domain data t and assessment dimensions A, B, and C is 0.4, 0.5, and 0.7 respectively, then the correlation between credit assessment dimension C and source domain data s is the same as the correlation between credit assessment dimension B and target domain data t (both are 0.5). Therefore, the source domain data s and the target domain data t can be uniformly mapped to credit assessment dimension B.

[0040] Alternatively, the credit assessment dimension with the highest average correlation can be used as the same feature dimension for mapping between the two.

[0041] Continuing from the previous example, the calculated mean correlations between credit assessment dimensions A, B, and C and source domain data s and target domain data t are 0.35, 0.5, and 0.75, respectively. Among them, the mean correlation of credit assessment dimension C is the largest. Therefore, source domain data s and target domain data t can be uniformly mapped to credit assessment dimension C.

[0042] It should be noted that for source domain data s and target domain data t, the same feature dimension mapped to them can be a new higher-order feature dimension other than the original feature dimension corresponding to the source domain data s and the original feature dimension corresponding to the target domain data t. Of course, it can also be the feature dimension corresponding to the source domain data s (that is, mapping the target domain data t to the feature dimension corresponding to the source domain data s, without mapping the source domain data s), or the feature dimension corresponding to the target domain data t (that is, mapping the source domain data s to the feature dimension corresponding to the target domain data t, without mapping the target domain data t).

[0043] In practical applications, the above mapping relationship can be stored in the form of a mapping table. After the mapping relationship is determined, the server can further utilize the mapping relationship to construct the feature space of the new industry through feature engineering.

[0044] The aforementioned feature space is used to uniformly represent the higher-order credit features under each of the aforementioned feature dimensions. The server uses the aforementioned mapping relationship to map source domain data and target domain data with common credit features to feature values ​​under the same feature dimension in the feature space. In this way, the industry specificity of data from different industries can be separated, and common features that reflect the essence of user credit can be retained and aggregated.

[0045] For example, if the source domain data is credit card repayment records and the target domain data is rent payment records, the two can be mapped to the feature dimension of periodic performance behavior through a mapping relationship. The on-time repayment rate reflected by the credit card repayment records and the on-time payment rate reflected by the rent payment records will then correspond to the specific feature value of performance stability under this feature dimension.

[0046] For example, if the source domain data is shopping frequency records on e-commerce platforms and the target domain data is ride-hailing frequency records, the two can be mapped to the feature dimension of regular consumption and travel behavior through a mapping relationship. The frequency of consumption operations reflected by the e-commerce shopping frequency records and the usage patterns of travel devices reflected by the ride-hailing frequency records are uniformly matched with the specific feature value of behavioral stability under this feature dimension.

[0047] For example, if the source domain data is the record of changes in emergency contacts for credit cards and the target domain data is the record of interaction frequency with friends on social platforms, the two can be mapped to the feature dimension of continuous social connection behavior through a mapping relationship. The stability of contacts reflected by the record of changes in emergency contacts for credit cards and the intensity of social interaction reflected by the record of interaction frequency with friends on social platforms will then correspond to the specific feature value of social relationship stability under this feature dimension.

[0048] It should be noted that the feature space can be a cross-domain unified credit representation space obtained by reconstructing the feature representation system based on the mapping relationship on the credit assessment model trained by the source domain data and reconfiguring the original feature dimensions and feature weights through feature engineering.

[0049] In this specification, the aforementioned feature space can serve as a predefined feature representation constraint domain. Based on this space, the basic rules for feature equivalence mapping are defined, ensuring that the original features of the source and target domains, after mapping, belong to this feature space. After configuring the original credit assessment model through this feature space, the parameters and feature representations of the equivalence mapping rules can be iteratively optimized through model training. This allows the model to fully learn the ability to accurately represent and refine credit features under cross-domain sample-driven conditions, thereby achieving both the accuracy of feature mapping and the effectiveness of credit representation within the feature space, ultimately completing a unified cross-domain credit feature representation.

[0050] S204: Input any sample data into the feature extraction network of the credit assessment model trained from the source domain data, so as to determine the credit features of the training sample in the feature space through the feature extraction network; S206: Based on the credit characteristics, determine the predicted domain classification corresponding to the sample data, and adjust the network parameters of the feature extraction network according to the deviation between the predicted domain classification and the actual domain classification of the training sample, so as to determine the transferred credit assessment model based on the adjusted feature extraction network; the transferred credit assessment model is applied to the credit assessment of the newly added industry.

[0051] In this specification, the credit assessment model trained from source domain data may include: a feature extraction network and a credit assessment network. The feature extraction network is used to extract features from the input credit data or sample data, and the credit assessment network is used to determine the feature assessment result (such as credit score, credit rating, risk classification, etc.) based on the credit features extracted by the feature extraction network.

[0052] In addition, to ensure the consistency of credit features between source and target domain data, a domain discrimination network can be temporarily set up during the transfer learning process. This domain discrimination network can be built into the credit assessment model or externally connected to the credit assessment model through interface calls, etc., to classify the industry domain of the input data based on the credit features extracted by the feature extraction network (i.e., to determine whether it is source domain data or target domain data).

[0053] Figure 3 This is a schematic diagram of a transfer learning model architecture provided in an exemplary embodiment.

[0054] like Figure 3 As shown, the server can input any sample data carrying a domain label into the feature extraction network of the credit assessment model trained from the source domain data, so as to determine the credit features of the training sample in the feature space through the feature extraction network. Then, the credit features are input into the domain discrimination network to obtain the predicted domain classification output by the domain discrimination network.

[0055] For feature extraction networks, it is necessary to make the domain discrimination model as indistinguishable as possible from the domain type (source domain / target domain) of the input data, so that the feature extraction network can learn the common features between the source domain data and the target domain data. Therefore, the server can determine a first loss value based on the deviation between the predicted domain classification and the actual domain classification of the training sample. This first loss value is negatively correlated with the deviation between the two. Then, the model parameters of the feature extraction network are adjusted with the goal of minimizing the first loss value to enhance the consistency of credit features across different domains.

[0056] Furthermore, for the domain discrimination network, it is necessary to enable it to accurately distinguish the domain type of the input data as much as possible, thereby improving its domain classification accuracy. This allows for adversarial training against the feature extraction network, gradually optimizing the feature learning performance of the feature extraction network during the adversarial training process. Therefore, the network parameters of the domain discrimination network can be adjusted with the goal of maximizing the first loss value.

[0057] For example, for source domain data s, after inputting it into the feature extraction network of the credit assessment model, credit feature x can be obtained. After inputting credit feature x into the domain discrimination network, the domain discrimination network outputs its corresponding domain classification c (c=s / t).

[0058] For target domain data t, after inputting it into the feature extraction network of the credit assessment model, credit features y can be obtained. After inputting credit features y into the domain discrimination network, the domain discrimination network outputs its corresponding domain classification c.

[0059] For feature extraction networks, it is necessary to ensure that the source and target domain credit features extracted tend to be consistent (i.e., This prevents the domain discrimination network from identifying whether the domain classification c determined based on credit features is source domain data s or target domain data t.

[0060] For domain discrimination networks, it is necessary to ensure that they can accurately determine the domain classification based on credit features x / y, thereby enabling adversarial training with the feature extraction network and forcing the feature extraction network to continuously optimize its feature extraction capabilities.

[0061] This allows for adversarial training between the feature extraction network and the domain discrimination network, ensuring the consistency of cross-domain credit features.

[0062] After adjusting the network parameters of the feature extraction network during adversarial training, the transferred credit assessment model can be determined based on the adjusted feature extraction network.

[0063] To further improve the adaptability of the credit assessment model to new industries, the server can also adjust the credit assessment network of the credit assessment model.

[0064] Specifically, when there are a small number of labeled samples in the target domain, the server can determine the sample label corresponding to the target domain data. This sample label includes the actual credit assessment result for the target domain data. Then, the target domain data is input into the credit assessment network to obtain the predicted credit assessment result output by the credit assessment network.

[0065] Then, based on the deviation between the actual credit assessment result and the predicted credit assessment result, a second loss value is determined. The network parameters of the credit assessment network are then adjusted with the goal of minimizing this second loss value. This second loss value is positively correlated with the deviation.

[0066] When there are no labeled samples in the target domain, the server can input the data of each target domain into the credit assessment model to obtain the credit assessment results and their corresponding confidence levels output by the credit assessment network. Then, the server can identify the target credit assessment results with confidence levels higher than the preset value from the credit assessment results and construct subsamples based on the target credit assessment results and their corresponding target domain data.

[0067] The subsamples are then input into the credit assessment network to obtain the predicted credit assessment results output by the network. Based on the deviation between the target credit assessment result and the predicted credit assessment results of the corresponding target domain data, a third loss value is determined. The network parameters of the credit assessment network are then adjusted with the goal of minimizing this third loss value. This third loss value is positively correlated with the deviation.

[0068] Furthermore, in order to quickly adapt to the addition of new industries, after the credit assessment model is adjusted, the model data and feature space configuration of the adjusted credit assessment model can be used as the industry configuration data for the new industries and stored in the industry configuration library. This will provide a configuration reference for the model building of the new industries and reduce the adaptation cost.

[0069] When a new industry needs to be adapted, the industry configuration data of an existing industry that is closest to the industry scenario information of the new industry can be determined from the preset industry configuration library based on the industry scenario information of the new industry.

[0070] In the process of constructing a new industry feature space, the feature space configuration information of the existing credit assessment model corresponding to the existing industry can be adjusted based on the cross-industry feature dimension mapping relationship in order to construct a feature space for the new industry.

[0071] This allows for the rapid construction of a new industry-specific feature space, improving the efficiency of model deployment.

[0072] For example, if the newly added industry is "new energy vehicle financial leasing", the industry configuration database already stores the configuration data and credit assessment model of the "traditional auto loan" industry. Since the industry scenario information of the two is quite similar, both belonging to auto credit, the configuration data and credit assessment model of the "traditional auto loan" industry can be directly extracted from it and applied to the credit assessment of the "new energy vehicle financial leasing" industry after adaptive adjustments.

[0073] To facilitate understanding, this manual also provides an overall flowchart of the transfer learning method, such as... Figure 4 As shown.

[0074] The process begins by semantically aligning source and target domain data with common features in the sample data. This mapping then maps the two to a unified high-order feature dimension, determining the mapping relationship between source and target domain features and that feature dimension. Based on this mapping, the feature space of the credit assessment model is reconstructed. Subsequently, the credit assessment model undergoes adversarial training and fine-tuning with a small amount of target domain data to obtain a credit assessment model adapted to the newly added industry.

[0075] Furthermore, this specification also provides a credit assessment method applied to a credit assessment model, which is obtained from the transfer learning methods described in S200 to S206 above.

[0076] Figure 5 This is a schematic diagram of a credit assessment method provided in an exemplary embodiment, including the following steps: S500: Obtain credit data for the target industry; S502: Input the credit data into a pre-trained credit assessment model to obtain the credit assessment result output by the credit assessment model.

[0077] For example, when a user engages in a housing rental transaction, the server can input the user's historical rent payment records and historical rental contracts into the credit assessment model, thereby determining the user's rental credit score through the credit assessment model.

[0078] Furthermore, after obtaining the credit assessment results, specific credit-related business can be carried out based on these results. Continuing from the previous example, after determining the user's rental credit score, a rental strategy that matches the credit score can be further determined, such as the types of houses that can be rented, whether a deposit is required and the amount of the deposit, and the rent payment method (such as quarterly, monthly, or annual payment). Then, the rental of houses to the user can be carried out based on the rental strategy.

[0079] Figure 6 This is a schematic structural diagram of a device provided in an exemplary embodiment. For example... Figure 6 As shown, device 600 mainly consists of a communication interface 602, a user interface 604, a processor 606, and a data storage 608. These components are interconnected and communicate with each other via a system bus, network, or other connection mechanism 610. The communication interface 602 enables device 600 to communicate with other devices, access networks, and transmission networks via analog or digital modulation. For example, the communication interface 602 may include a chipset and antenna for wireless communication with a radio access network or access point. Furthermore, the communication interface 602 can be a wired interface such as Ethernet, Token Ring, or a USB port, or a wireless interface such as Wi-Fi, Bluetooth, Global Positioning System (GPS), or a wide-area wireless interface (e.g., WiMAX or LTE). Of course, the communication interface 602 can also support other forms of physical layer interfaces and standard or proprietary communication protocols. The communication interface 602 may also include multiple physical communication interfaces, such as Wi-Fi, Bluetooth, and wide-area wireless interfaces.

[0080] User interface 604 includes receiving user input and providing output to the user. Therefore, user interface 604 may include input components such as a keypad, keyboard, touch-sensitive or presence-sensitive panel, computer mouse, trackball, joystick, microphone, still camera, and video camera, and output components such as a display screen (which may be combined with a touch-sensitive panel), CRT, LCD, LED, display using DLP technology, printer, and other similar devices known or developed in the future. User interface 604 may also generate auditory output via speakers, speaker jacks, audio output ports, audio output devices, headphones, and other similar devices known or developed in the future. In some embodiments, user interface 604 may include software, circuitry, or other forms of logic capable of transmitting and receiving data from external user input / output devices. Additionally or alternatively, device 600 may support remote access from other devices via communication interface 602 or another physical interface (not shown). User interface 604 may be configured to receive user input, the position and movement of which may be indicated by indicators or cursors described herein. User interface 604 may also be configured as a display device for rendering or displaying text fragments.

[0081] Processor 606 may contain one or more general-purpose processors and / or special-purpose processors.

[0082] Data storage 608 may include one or more volatile and / or non-volatile storage components and may be integrated wholly or partially with processor 606. Data storage 608 may include removable and non-removable components.

[0083] Processor 606 is capable of executing program instructions 618 (e.g., compiled or uncompiled program logic and / or machine code) stored in data storage 608 to perform the various functions described herein. Data storage 608 may contain a non-transitory computer-readable medium on which program instructions are stored, which, when executed by device 600, enable device 600 to perform any methods, processes, or functions disclosed in this specification and / or the accompanying drawings. Execution of program instructions 618 by processor 606 may result in processor 606 using data 612.

[0084] For example, program instructions 618 may include an operating system 622 (e.g., an operating system kernel, device drivers, and / or other modules) installed on device 600 and one or more applications 620 (e.g., a browser, social application, or game application). Similarly, data 612 may include operating system data 616 and application data 614. Operating system data 616 is primarily accessible to the operating system 622, while application data 614 is primarily accessible to one or more applications 620. Application data 614 may reside in a file system visible or hidden from the user of device 600.

[0085] Application 620 can communicate with operating system 622 through one or more application programming interfaces (APIs). These APIs help application 620 read and / or write application data 614, transmit or receive information via communication interface 602, receive or display information on user interface 604, etc.

[0086] In some terminology, application 620 may be simply referred to as "app". Furthermore, application 620 can be downloaded to device 600 through one or more online app stores or app markets. However, applications can also be installed on device 600 in other ways, such as through a web browser or a physical interface on device 600 (e.g., a USB port).

[0087] Please refer to Figure 7 The transfer learning mechanism of the credit assessment model can be applied to, for example... Figure 6 The device shown is used to implement the technical solution of this specification. The transfer learning device for the credit assessment model may include: The acquisition module 700 is used to acquire sample data; wherein, the sample data is credit data, including source domain data corresponding to existing industries and target domain data corresponding to newly added industries; The construction module 702 is used to map source domain data and target domain data with common features to the same feature dimension, and to determine the mapping relationship between the source domain data and target domain data and the feature dimension respectively, and to construct a feature space for the newly added industry based on the determined mapping relationship. Input module 704 is used to input any sample data into the feature extraction network of the credit assessment model trained by the source domain data, so as to determine the credit features of the training sample in the feature space through the feature extraction network. The transfer module 706 is used to determine the predicted domain classification corresponding to the sample data based on the credit features, and to adjust the network parameters of the feature extraction network based on the deviation between the predicted domain classification and the actual domain classification of the training sample, so as to determine the transferred credit assessment model based on the adjusted feature extraction network; the transferred credit assessment model is applied to the credit assessment of the newly added industry.

[0088] Optionally, the transfer module 706 is specifically configured to: input the target credit features into a temporarily set domain discrimination network to obtain the predicted domain classification output by the domain discrimination network; determine a first loss value based on the deviation between the predicted domain classification and the actual domain classification of the training sample; wherein the first loss value is negatively correlated with the deviation; and adjust the model parameters of the feature extraction network with the goal of minimizing the first loss value.

[0089] Optionally, the migration module 706 is specifically configured to adjust the network parameters of the feature extraction network with the optimization objective of minimizing the first loss value, and to adjust the network parameters of the domain discrimination network with the optimization objective of maximizing the first loss value.

[0090] Optionally, the credit assessment model includes: a credit assessment network; Optionally, the migration module 706 is further configured to: determine sample labels corresponding to the target domain data; the sample labels include actual credit assessment results for the target domain data; input the target domain data into the credit assessment network to obtain a predicted credit assessment result output by the credit assessment network; determine a second loss value based on the deviation between the sample labels and the predicted credit assessment result; wherein the second loss value is positively correlated with the deviation; and adjust the network parameters of the credit assessment network with minimizing the second loss value as the optimization objective.

[0091] Optionally, the credit assessment model includes: a credit assessment network. Optionally, the migration module 706 is specifically configured to: input data from each target domain into the credit assessment model to obtain credit assessment results and their corresponding confidence levels output by the credit assessment network; identify target credit assessment results with confidence levels higher than a preset value from among the credit assessment results, and construct subsamples based on the target credit assessment results and their corresponding target domain data; input the subsamples into the credit assessment network to obtain predicted credit assessment results output by the credit assessment network; determine a third loss value based on the deviation between the target credit assessment results and the predicted credit assessment results of their corresponding target domain data; wherein the third loss value is positively correlated with the deviation; and adjust the network parameters of the credit assessment network with minimizing the third loss value as the optimization objective.

[0092] Optionally, the acquisition module 700 is further configured to, based on the industry scenario information of the newly added industry, determine the industry configuration data of an existing industry that is closest to the industry scenario information of the newly added industry in a preset industry configuration library; wherein, the industry configuration information includes the feature space configuration and credit assessment model data of the existing industry; The construction module 702 is specifically used to adjust the feature space configuration information of the credit assessment model corresponding to the existing industry based on the mapping relationship, so as to construct a feature space for the newly added industry. The device further includes: The storage module 708 is used to store the model data and feature space configuration of the adjusted credit assessment model as the industry configuration data of the newly added industry in the industry configuration library.

[0093] Optionally, the construction module 702 is specifically used to: determine multiple credit assessment dimensions that affect the credit assessment results; for each sample data, determine the degree of correlation between the sample data and each credit assessment dimension; based on the degree of correlation between each sample data and each credit assessment dimension, determine source domain data and target domain data that have commonalities, and project the source domain data and target domain data onto the same feature dimension.

[0094] Please refer to Figure 8 Credit assessment devices can be applied to, for example... Figure 6 The device shown is used to implement the technical solution described in this specification. The credit assessment device may include: The acquisition module is used to acquire credit data for the target industry; The evaluation module is used to input the credit data into a pre-trained credit evaluation model and obtain the credit evaluation result output by the credit evaluation model.

[0095] For ease of description, the above devices are described by dividing them into various modules or units based on their functions. Of course, when implementing one or more of these specifications, the functions of each module or unit can be implemented in the same or different software and / or hardware, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0096] Based on the same concept as the methods described above, this specification also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor performs the steps of the method as described in any of the above embodiments by executing the executable instructions.

[0097] Based on the same concept as the methods described above, this specification also provides a computer-readable storage medium having computer instructions stored thereon that, when executed by a processor, implement the steps of the methods as described in any of the above embodiments.

[0098] Based on the same concept as the methods described above, this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the methods as described in any of the above embodiments.

[0099] What those skilled in the art will understand is: In this specification, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitation, the presence of additional identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded.

[0100] In this specification, “a,” “an,” and “the” do not specifically refer to the singular, but may also include the plural.

[0101] In this specification, ordinal numbers such as "first," "second," etc., do not necessarily indicate order; they are often used to distinguish between objects. For example, "first server" and "second server" usually refer to two servers. To differentiate between these two servers, they are described as "first server" and "second server." Of course, sometimes these two servers may be the same server.

[0102] In this specification, unless explicitly stated otherwise, "receiving and sending data" does not necessarily mean direct receiving and sending; it can also mean indirect receiving and sending. For example, A receiving data sent by B can be understood as A directly receiving the data sent by B, or it can be understood as A indirectly receiving the data sent by B through other entities such as C. Similarly, B sending data to A can be understood as B sending the data directly to A, or it can be understood as B indirectly sending the data to A through other entities such as C. Here, C can be one entity, or it can be two or more entities.

[0103] In this specification, unless explicitly stated otherwise, the relationships between structures can be direct or indirect. For example, when describing "A is connected to B," unless it is explicitly stated that A and B are directly connected, it should be understood that A can be directly connected to B or indirectly connected to B. Similarly, when describing "A is on top of B," unless it is explicitly stated that A is directly above B (AB is adjacent and A is above B), it should be understood that A can be directly above B or indirectly above B (AB is separated by other elements, and A is above B). And so on.

[0104] This specification uses specific terms to describe embodiments thereof. Terms such as "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described herein, as well as the features of those different embodiments or examples, without contradiction.

[0105] Although one or more embodiments of this specification provide method steps as described in the embodiments or flowcharts, it is understood that the order of steps listed in the embodiments or flowcharts is only one of many possible execution orders and does not represent the only execution order. Therefore, when the claims involve method steps, any changes or adjustments to the order of such steps, or the parallelism between steps, are also within the scope of protection of the claims.

Claims

1. A transfer learning method for a credit assessment model, comprising: Acquire sample data; wherein, the sample data is credit data, including source domain data corresponding to existing industries and target domain data corresponding to newly added industries; Source domain data and target domain data with common features are mapped to the same feature dimension, and the mapping relationship between the source domain data and target domain data and the feature dimension is determined respectively. Based on the determined mapping relationship, a feature space for the new industry is constructed. Any sample data is input into the feature extraction network of the credit assessment model trained from the source domain data, so as to determine the credit features of the training sample in the feature space through the feature extraction network; Based on the credit characteristics, the predicted domain classification corresponding to the sample data is determined, and the network parameters of the feature extraction network are adjusted according to the deviation between the predicted domain classification and the actual domain classification of the training sample, so as to determine the transferred credit assessment model based on the adjusted feature extraction network; the transferred credit assessment model is applied to the credit assessment of the newly added industry.

2. The method as described in claim 1, wherein adjusting the model parameters of the feature extraction network based on the deviation between the predicted domain classification and the actual domain classification of the training sample, specifically includes: The target credit features are input into a temporarily set domain discrimination network to obtain the predicted domain classification output by the domain discrimination network; A first loss value is determined based on the deviation between the predicted domain classification and the actual domain classification of the training sample; wherein the first loss value is negatively correlated with the deviation. The model parameters of the feature extraction network are adjusted with the goal of minimizing the first loss value.

3. The method as described in claim 2, wherein minimizing the first loss value is the optimization objective, and the network parameters of the feature extraction network are adjusted, specifically including: The network parameters of the feature extraction network are adjusted with the goal of minimizing the first loss value, and the network parameters of the domain discrimination network are adjusted with the goal of maximizing the first loss value.

4. The method as described in claim 1, wherein the credit assessment model comprises: Credit assessment network; The method further includes: Determine the sample labels corresponding to the target domain data; the sample labels include the actual credit assessment results for the target domain data; The target domain data is input into the credit assessment network to obtain the predicted credit assessment result output by the credit assessment network; A second loss value is determined based on the deviation between the sample label and the predicted credit assessment result; wherein the second loss value is positively correlated with the deviation. The network parameters of the credit assessment network are adjusted with the goal of minimizing the second loss value.

5. The method as described in claim 1, wherein the credit assessment model comprises: Credit rating network: The method further includes: Input the data from each target domain into the credit assessment model to obtain the credit assessment results and their corresponding confidence levels output by the credit assessment network. Among the various credit assessment results, a target credit assessment result with a confidence level higher than a preset value is identified, and a subsample is constructed based on the target credit assessment result and its corresponding target domain data; The subsample is input into the credit assessment network to obtain the predicted credit assessment result output by the credit assessment network; A third loss value is determined based on the deviation between the target credit assessment result and the predicted credit assessment result of the corresponding target domain data; wherein the third loss value is positively correlated with the deviation. The network parameters of the credit assessment network are adjusted with the goal of minimizing the third loss value.

6. The method of claim 1, wherein before constructing the feature space for the newly added industry based on the mapping relationship, the method further comprises: Based on the industry scenario information of the newly added industry, the industry configuration data of the existing industry that is closest to the industry scenario information of the newly added industry is determined from the preset industry configuration library; wherein, the industry configuration information includes the feature space configuration and credit assessment model data of the existing industry. Based on the mapping relationship, a feature space is constructed for the newly added industry, specifically including: Based on the mapping relationship, the feature space configuration information of the existing credit assessment model corresponding to the industry is adjusted to construct a feature space for the new industry. After adjusting the model parameters of the credit assessment model, the method further includes: The model data and feature space configuration of the adjusted credit assessment model are stored in the industry configuration library as the industry configuration data of the newly added industry.

7. The method as described in claim 1, mapping source domain data and target domain data with common features to the same feature dimension, specifically includes: Identify the various dimensions of credit assessment that influence the results; For each sample of data, determine the degree of correlation between that sample of data and each dimension of credit assessment; Based on the degree of correlation between each sample data and each credit assessment dimension, source domain data and target domain data with commonalities are identified, and these source domain data and target domain data are mapped to the same feature dimension.

8. A credit assessment method, comprising: Obtain credit data for the target industry; The credit data is input into a pre-trained credit assessment model to obtain the credit assessment result output by the credit assessment model. The credit assessment model is trained by the method described in any one of claims 1 to 7.

9. A training device for a credit assessment model, comprising: The acquisition module is used to acquire sample data; wherein, the sample data is credit data, including source domain data corresponding to existing industries and target domain data corresponding to newly added industries; The construction module is used to map source domain data and target domain data with common features to the same feature dimension, determine the mapping relationship between the source domain data and target domain data and the feature dimension respectively, and construct a feature space for the newly added industry based on the determined mapping relationship. The input module is used to input any sample data into the feature extraction network of the credit assessment model trained by the source domain data, so as to determine the credit features of the training sample in the feature space through the feature extraction network; The transfer module is used to determine the predicted domain classification corresponding to the sample data based on the credit features, and to adjust the network parameters of the feature extraction network based on the deviation between the predicted domain classification and the actual domain classification of the training sample, so as to determine the transferred credit assessment model based on the adjusted feature extraction network; the transferred credit assessment model is applied to the credit assessment of the newly added industry.

10. An electronic device, comprising: processor; A memory for storing processor-executable instructions; wherein the processor implements the steps of the method as described in any one of claims 1-8 by executing the executable instructions.

11. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1-8.

12. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method as claimed in any one of claims 1-8.

Images