Method and apparatus for training and providing identification result of financial system user identification model

Abstract

The application relates to a financial system user identification model training method, device, medium and product, and relates to the fields of artificial intelligence technology and financial technology. The method comprises the following steps: acquiring user behavior data samples and user label sample values of a financial system; constructing a user continuous feature matrix and a user category feature matrix, and inputting the two matrices into a feature cross model based on a deep neural network and a feature cross model based on a domain matrix factor decomposition machine respectively to obtain first cross features and second cross features; inputting the two cross features into a feature cross model based on a multi-layer self-attention mechanism to obtain third cross features; inputting the third cross features into a user identification model to be trained to obtain user label prediction values; determining a model loss according to the user label sample values and the user label prediction values, adjusting model parameters of the user identification model to be trained until the model loss meets a preset condition. The method can obtain an identification model with higher precision and accuracy.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence technology and financial technology, and in particular to a method for training a user identification model for a financial system, a method for providing user identification results, computer equipment, storage media, and computer program products. Background Technology

[0002] With the development of information technology, more and more financial institutions are beginning to provide services to users through interactive financial systems. In order to improve the user experience, they usually identify users based on the information they provide with their authorization, obtain corresponding user tags, and then provide personalized services based on these tags.

[0003] However, the current financial system typically uses simple linear prediction models built on machine learning algorithms to identify users by analyzing user authorization information. This ignores the complex interactions between different types of user information and fails to process user information of different natures in a targeted manner, resulting in low accuracy of prediction models in identifying users and making it difficult to effectively improve users' financial service experience. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for training a user identification model for a financial system, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a method for training a user identification model for a financial system. The method includes:

[0006] Obtain user behavior data samples and corresponding user tag sample values ​​from the financial system.

[0007] Based on user behavior data samples, construct a continuous feature matrix for users and a feature matrix for user categories;

[0008] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0009] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0010] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0011] Input the third cross feature into the user identification model to be trained, and obtain the user label prediction value output by the user identification model to be trained.

[0012] The model loss is determined based on the user tag sample values ​​and the user tag predicted values;

[0013] Based on the model loss, adjust the model parameters of the user recognition model to be trained until the model loss meets the preset conditions to obtain the trained user recognition model.

[0014] In one embodiment, the deep neural network includes multiple processing layers; a feature cross-model based on the deep neural network is used to convert the input user continuous feature matrix into a corresponding vector and transmit it to the first processing layer of the deep neural network. The first processing layer generates an intermediate processing vector from the input vector and transmits it to the next processing layer. Each non-first processing layer generates a new intermediate processing vector based on the received intermediate processing vector and passes it to the next processing layer, until the last processing layer generates an intermediate processing vector as the first cross feature.

[0015] In one embodiment, the user category feature matrix includes multiple user category feature vectors; a feature cross model based on a domain matrix factorization machine is used to obtain the user category feature domain corresponding to each user category feature vector according to the user category feature matrix, obtain the domain interaction matrix corresponding to every two user category feature domains according to the user category feature domains, obtain the cross representation of every two user category feature vectors according to the user category feature matrix and the domain interaction matrix, and obtain the second cross feature based on multiple cross representations.

[0016] In one embodiment, the feature cross model based on a multi-layer self-attention mechanism includes a multi-layer self-attention module. The self-attention module generates an interaction relationship vector and a value weight vector based on the input vector, and obtains an attention cross feature vector based on the interaction relationship vector and the value weight vector. The feature cross model based on the multi-layer self-attention mechanism is used to convert the first cross feature and the second cross feature into corresponding vectors and input them into the first-layer self-attention module. The first-layer self-attention module obtains the attention cross feature vector and transmits it to the next-layer self-attention module. Subsequent self-attention modules continue to generate new attention cross feature vectors based on the received attention cross feature vectors, until the attention cross feature vector generated by the last self-attention module is used as the third cross feature.

[0017] In one embodiment, constructing a continuous user feature matrix and a category user feature matrix based on user behavior data samples includes: extracting multiple sub-behavior data from the user behavior data samples; obtaining derived sub-behavior data from the multiple sub-behavior data; identifying continuous behavior data and category behavior data in the sub-behavior data and derived sub-behavior data to form a continuous behavior dataset and a category behavior dataset; constructing a continuous user feature matrix based on the continuous behavior dataset; and constructing a category user feature matrix based on the category behavior dataset.

[0018] In one embodiment, constructing a user category feature matrix based on the category behavior dataset includes: performing word embedding processing on the category behavior data in the category behavior dataset to form a user category feature vector; and constructing a user category feature matrix based on the category behavior feature vector corresponding to the category behavior dataset.

[0019] Secondly, this application also provides a method for providing user identification results in a financial system. The method includes:

[0020] After obtaining user behavior data authorized by users for analysis, construct a continuous user feature matrix and a user category feature matrix based on the user behavior data;

[0021] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0022] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0023] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0024] Input the third cross feature into the trained user identification model to obtain the user label prediction value output by the trained user identification model;

[0025] Based on the predicted user tag values, the user identification results are obtained;

[0026] Display the user identification results on the page of the financial system that the user enters;

[0027] The trained user identification model is obtained by training any of the above-mentioned financial system user identification model training methods.

[0028] Thirdly, this application also provides a training device for a user identification model in a financial system. The device includes:

[0029] The acquisition module is used to acquire user behavior data samples and corresponding user tag sample values ​​from the financial system.

[0030] The building module is used to construct user continuous feature matrices and user category feature matrices based on user behavior data samples;

[0031] The first cross module is used to input the user's continuous feature matrix into the feature cross model based on a deep neural network to obtain the first cross feature;

[0032] The second cross module is used to input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross features;

[0033] The third cross module is used to input the first cross features and the second cross features into the feature cross model based on the multi-layer self-attention mechanism to obtain the third cross features.

[0034] The prediction module is used to input the third cross feature into the user identification model to be trained and obtain the user label prediction value output by the user identification model to be trained.

[0035] The loss determination module is used to determine the model loss based on the user label sample values ​​and the predicted user label values;

[0036] The adjustment module is used to adjust the model parameters of the user recognition model to be trained based on the model loss, until the model loss meets the preset conditions to obtain the trained user recognition model.

[0037] Fourthly, this application also provides an apparatus for providing user identification results in a financial system. The apparatus includes:

[0038] The module is used to construct a continuous feature matrix and a category feature matrix of users based on the user behavior data after obtaining user authorization analysis.

[0039] The first cross module is used to input the user's continuous feature matrix into the feature cross model based on a deep neural network to obtain the first cross feature;

[0040] The second cross module is used to input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross features;

[0041] The third cross module is used to input the first cross features and the second cross features into the feature cross model based on the multi-layer self-attention mechanism to obtain the third cross features.

[0042] The prediction module is used to input the third cross feature into the trained user identification model and obtain the user label prediction value output by the trained user identification model.

[0043] The recognition module is used to obtain the user recognition result based on the predicted value of the user tag;

[0044] The display module is used to show the user identification results on the pages of the financial system that the user enters;

[0045] The trained user identification model is obtained by training any of the above-mentioned financial system user identification model training methods.

[0046] Fifthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0047] Obtain user behavior data samples and corresponding user tag sample values ​​from the financial system.

[0048] Based on user behavior data samples, construct a continuous feature matrix for users and a feature matrix for user categories;

[0049] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0050] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0051] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0052] Input the third cross feature into the user identification model to be trained, and obtain the user label prediction value output by the user identification model to be trained.

[0053] The model loss is determined based on the user tag sample values ​​and the user tag predicted values;

[0054] Based on the model loss, adjust the model parameters of the user recognition model to be trained until the model loss meets the preset conditions to obtain the trained user recognition model.

[0055] Sixthly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0056] After obtaining user behavior data authorized by users for analysis, construct a continuous user feature matrix and a user category feature matrix based on the user behavior data;

[0057] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0058] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0059] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0060] Input the third cross feature into the trained user identification model to obtain the user label prediction value output by the trained user identification model;

[0061] Based on the predicted user tag values, the user identification results are obtained;

[0062] Display the user identification results on the page of the financial system that the user enters;

[0063] The trained user identification model is obtained by training any of the above-mentioned financial system user identification model training methods.

[0064] Seventhly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0065] Obtain user behavior data samples and corresponding user tag sample values ​​from the financial system.

[0066] Based on user behavior data samples, construct a continuous feature matrix for users and a feature matrix for user categories;

[0067] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0068] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0069] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0070] Input the third cross feature into the user identification model to be trained, and obtain the user label prediction value output by the user identification model to be trained.

[0071] The model loss is determined based on the user tag sample values ​​and the user tag predicted values;

[0072] Based on the model loss, adjust the model parameters of the user recognition model to be trained until the model loss meets the preset conditions to obtain the trained user recognition model.

[0073] Eighthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0074] After obtaining user behavior data authorized by users for analysis, construct a continuous user feature matrix and a user category feature matrix based on the user behavior data;

[0075] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0076] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0077] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0078] Input the third cross feature into the trained user identification model to obtain the user label prediction value output by the trained user identification model;

[0079] Based on the predicted user tag values, the user identification results are obtained;

[0080] Display the user identification results on the page of the financial system that the user enters;

[0081] The trained user identification model is obtained by training any of the above-mentioned financial system user identification model training methods.

[0082] Ninthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0083] Obtain user behavior data samples and corresponding user tag sample values ​​from the financial system.

[0084] Based on user behavior data samples, construct a continuous feature matrix for users and a feature matrix for user categories;

[0085] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0086] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0087] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0088] Input the third cross feature into the user identification model to be trained, and obtain the user label prediction value output by the user identification model to be trained.

[0089] The model loss is determined based on the user tag sample values ​​and the user tag predicted values;

[0090] Based on the model loss, adjust the model parameters of the user recognition model to be trained until the model loss meets the preset conditions to obtain the trained user recognition model.

[0091] Tenthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0092] After obtaining user behavior data authorized by users for analysis, construct a continuous user feature matrix and a user category feature matrix based on the user behavior data;

[0093] Input the user's continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0094] Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0095] The first and second cross features are input into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0096] Input the third cross feature into the trained user identification model to obtain the user label prediction value output by the trained user identification model;

[0097] Based on the predicted user tag values, the user identification results are obtained;

[0098] Display the user identification results on the page of the financial system that the user enters;

[0099] The trained user identification model is obtained by training any of the above-mentioned financial system user identification model training methods.

[0100] The aforementioned training method, apparatus, computer equipment, storage medium, and computer program product for the user identification model in the financial system can construct a continuous user feature matrix and a user category feature matrix based on data from user behavior data samples. It then processes the continuous user feature matrix using a feature cross-model based on a deep neural network to obtain a first cross-feature, processes the user category feature matrix using a feature cross-model based on a domain matrix factorization machine to obtain a second cross-feature, and further processes the first and second cross-features using a feature cross-model based on a multi-layer self-attention mechanism to obtain a third cross-feature. This third cross-feature is then input into the user identification model to be trained. The model loss is determined based on the predicted user label values ​​obtained by the model and the user labels corresponding to the user behavior data samples, and the model is adjusted. The above process involves two main steps. First, cross-referencing different types of behavioral data within the user behavior data samples allows for a better understanding of their complex interactions. Second, for data of different natures within the user behavior data samples, continuous feature matrices and categorical feature matrices are constructed separately. Then, based on the characteristics of these two matrices, different cross-referencing methods are used to perform feature cross-referencing on each matrix. The results are then further fused and learned, which allows for better processing of the two types of data. The resulting third cross-feature more accurately reflects the characteristics of the user behavior data. Using this third cross-feature as input to the user identification model allows for the training of a model with higher user identification accuracy. Subsequently, using this model to predict user tags based on user-authorized information can provide users with more accurate identification results, ultimately enabling the provision of more personalized financial services tailored to user needs. Attached Figure Description

[0101] Figure 1 This is an application environment diagram of a method for training a user identification model in a financial system and a method for providing user identification results in a financial system, as shown in one embodiment.

[0102] Figure 2 This is a flowchart illustrating a method for training a user identification model in a financial system, as shown in one embodiment.

[0103] Figure 3 This is a schematic diagram of the structure of a user identification model training method for a financial system in one embodiment;

[0104] Figure 4 This is a flowchart illustrating the process of constructing a continuous user feature matrix and a user category feature matrix based on user behavior data samples in one embodiment.

[0105] Figure 5 This is a flowchart illustrating a method for providing user identification results in a financial system, as shown in one embodiment.

[0106] Figure 6This is a schematic diagram of the structure of a financial system user identification model training device in one embodiment;

[0107] Figure 7 This is a schematic diagram of the structure of a device that provides user identification results in a financial system, as shown in one embodiment.

[0108] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0109] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0110] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0111] The user identification model training method and the user identification result provision method in the financial system provided in this application embodiment can be applied to, for example, financial systems. Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located on the cloud or other network servers. Specifically, the data storage system can be used to store user behavior data samples and user behavior data authorized for analysis. Users and staff of the financial system can interact with the financial system through terminal 102 and initiate the training and recognition process of the recognition model. Server 104 can process the user behavior data samples and user-authorized user behavior data, and further train the model or use the model for recognition. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Server 104 can be implemented using a standalone server or a server cluster composed of multiple servers.

[0112] In one embodiment, such as Figure 2 As shown, a method for training a user identification model for a financial system is presented. Please also refer to... Figure 3This is a schematic diagram of the feature cross model used in this method. The method is applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0113] Step S201: Obtain user behavior data samples from the financial system and the corresponding user tag sample values.

[0114] Specifically, user behavior data samples can be information authorized by some users from the financial system, including users' personal information and financial behavior information. Personal information may include users' age, region, occupation, related relatives, and related companies, while financial behavior information may include transaction information generated by users when using the financial system, such as transfers, consumption, investment and wealth management, third-party payments, public utility payments, insurance, payroll, and product holdings, as well as other related financial behavior information provided by users themselves, such as credit reports, deposits, and loans. Authorized information from multiple different users can form this user behavior data sample.

[0115] Furthermore, when users provide the above information and authorize analysis, they can also submit their evaluation of the financial system's service experience and the personalized services they wish to receive. Based on the above content, the user tag sample value corresponding to the user can be determined.

[0116] Step S202: Based on user behavior data samples, construct a continuous user feature matrix and a user category feature matrix.

[0117] Specifically, the user behavior data sample contains various types of data. For example, data such as deposits and spending directly reflect users' behavior in corresponding transactions, and these values ​​can vary within a continuous range. Data such as region and occupation, on the other hand, use text to describe the categories to which users belong in different areas. Based on this, the first type of data in the user behavior data sample can be categorized as continuous behavior data, and the second type as categorical behavior data.

[0118] To facilitate subsequent feature cross-processing, each continuous or categorical behavior data point can be converted into a vector in this step, forming a user continuous feature matrix and a user category feature matrix, respectively. Specifically, for continuous behavior data, each data point can be normalized to form a corresponding vector, and the vectors formed from multiple continuous behavior data points can form the user continuous feature matrix. For categorical behavior data, its textual expression can first be converted into numerical values, then vectors can be formed, and further, a user category feature matrix can be formed based on the vectors corresponding to multiple categorical behavior data points.

[0119] Furthermore, since the user behavior data samples come from user-authorized data, the data provided by different users may have different omissions or deviate from the statistical range, making them unsuitable for direct model training. Therefore, this step can also preprocess the user behavior data samples. For example, if a user provides their current account balance but not their current deposit balance, the current account balance can be used to impute the current deposit balance; or, if a user's personal consumption is too high and statistically insignificant, they can be removed from the user behavior data samples. After the above preprocessing, user behavior data samples that are more suitable for feature crossing and model training can be obtained.

[0120] Step S203: Input the user's continuous feature matrix into the feature cross model based on a deep neural network to obtain the first cross feature.

[0121] Specifically, the feature cross model based on a deep neural network is used to input each vector in the user continuous feature matrix obtained in step S202 into the first layer of a pre-trained deep neural network. After multiple layers of learning by the deep neural network, the vector output by the last layer is the first cross feature corresponding to the user continuous feature matrix. This first cross feature can reflect the interaction relationship of different user continuous behavior data in the user behavior data sample.

[0122] Step S204: Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature.

[0123] Specifically, the feature cross model based on the domain matrix factorization machine is used to map each vector in the user category feature matrix obtained in step S202 to a multi-feature domain space, and generate a domain interaction matrix according to the interaction relationship between different feature domains. Then, the domain matrix factorization machine is used to calculate the interaction relationship between different vectors in the user category feature matrix, and obtain a second cross feature that can reflect the interaction relationship.

[0124] Step S205: Input the first cross feature and the second cross feature into the feature cross model based on the multi-layer self-attention mechanism to obtain the third cross feature.

[0125] Specifically, this feature cross-feedback model based on a multi-layer self-attention mechanism includes multiple self-attention modules. The first and second cross-feedback features obtained in steps S203 and S204 are input into the first self-attention module to learn their interaction relationships. The results are then input into the next module. After processing by multiple modules, the output of the last self-attention module is the third cross-feedback feature. This third cross-feedback feature can comprehensively reflect the interaction relationships between various types of data, including continuous user behavior data and user category behavior data.

[0126] Step S206: Input the third cross feature into the user identification model to be trained, and obtain the user label prediction value output by the user identification model to be trained.

[0127] Specifically, in this step, a user identification model to be trained can be constructed based on the specific target to be identified.

[0128] For example, a user identification model can be constructed as shown in the following formula:

[0129]

[0130] Among them, W P ,b represents the model parameters, and P is the third cross feature to be input into the model. This represents the prediction result of the recognition model, where δ is the activation function, which can be selected from functions such as ReLU and Sigmoid depending on the specific recognition purpose and the characteristics of the data.

[0131] Furthermore, by inputting the third cross feature obtained in step S205 into the constructed model, the predicted user label value can be output for the corresponding user.

[0132] Step S207: Determine the model loss based on the user tag sample values ​​and the user tag predicted values.

[0133] Specifically, in this step, the model loss is determined based on the user label sample values ​​obtained in step S201 and the user label predicted values ​​obtained in step S206. For example, the user label sample values ​​and the user label predicted values ​​can be compared, and the model loss can be determined based on the difference between the two.

[0134] For example, in this embodiment, a loss function based on the cross-entropy method can be used to obtain the model loss, as shown in the following equation:

[0135]

[0136] Where R is the number of users, y r This represents the user tag sample value of the r-th user. This is used to identify the predicted user label value for the r-th user output by the model.

[0137] For example, for the identification objective of "whether a user needs to use service A in the financial system," the user tag sample value has two possible values: "need" and "don't need," which can be represented by 1 and 0. The user identification model can output a value between 0 and 1 as the predicted user tag value. After inputting both into the aforementioned loss function, the model loss can be determined. The smaller the model loss value, the higher the identification accuracy of the model.

[0138] Step S208: Adjust the model parameters of the user recognition model to be trained according to the model loss until the model loss meets the preset conditions to obtain the trained user recognition model.

[0139] Specifically, user behavior data samples can be divided into multiple batches. First, the model loss of the user recognition model is obtained by using the first batch of samples according to the above steps. Then, the model parameters are adjusted. Then, the model loss of the adjusted user recognition model is obtained by using the second batch of samples. This process is repeated multiple times until the model loss meets the preset conditions, such as being less than or equal to a set value. At this point, the user recognition model can be considered to have completed training.

[0140] In the aforementioned training method for the user identification model in the financial system, on the one hand, different types of behavioral data in the user behavior data samples are cross-processed to better uncover their complex interactions; on the other hand, continuous feature matrices and categorical feature matrices are constructed for different types of data in the user behavior data samples. Then, based on the characteristics of the two matrices, different cross-processing methods are used to perform feature cross-processing on the two matrices themselves, and the results are further fused and learned. This method can better process the two types of data, and the resulting third cross-feature can more realistically reflect the characteristics of user behavior data. Using this third cross-feature as input to the user identification model can train a model with higher user identification accuracy. Subsequently, using the model to predict user tags based on user authorization information can also provide users with more accurate identification results, and thus provide more personalized financial services that meet user needs.

[0141] In one embodiment, the deep neural network includes multiple processing layers; a feature cross model based on the deep neural network is used to convert the input user continuous feature matrix into a corresponding vector and transmit it to the first processing layer of the deep neural network, the first processing layer generates an intermediate processing vector for the input vector and transmits it to the next processing layer, each non-first processing layer generates a new intermediate processing vector based on the received intermediate processing vector and passes it to the next processing layer, until the last processing layer generates an intermediate processing vector as the first cross feature.

[0142] Specifically, this embodiment provides a specific implementation method for obtaining the first intersection feature in the above step S203.

[0143] The feature cross-processing model based on a deep neural network uses a deep neural network comprising multiple processing layers. Each processing layer performs cross-processing on the input vectors and outputs an intermediate processed vector. Except for the first and last processing layers, each processing layer takes the intermediate processed vector output by the previous processing layer as input and inputs the intermediate processed vector generated by the current layer into the next layer. Specifically, the process of each processing layer generating an intermediate processed vector can be represented by the following formula:

[0144]

[0145] in This is represented as an intermediate processing vector output by the (l-1)th processing layer of a deep neural network. and b l For model parameters, This represents the intermediate processing vector of the output of the l-th layer of the deep neural network.

[0146] The user continuous feature matrix in this embodiment can be represented by the following formula:

[0147]

[0148] Where N is the number of subsequences, i.e., the number of continuous user behavior data, T is the dimension of each continuous user behavior data, and D is the number of subsequences. n ={d n,1 ,...,d n,T} represents the continuous behavior data vector of the nth user.

[0149] After inputting the above matrix into a feature cross model based on a deep neural network, the model extracts different user continuous behavior data vectors from the user continuous feature matrix, transforms them into vectors suitable for input into the deep neural network, and uses them as input to the first processing layer of the deep neural network. This process can be expressed by the following formula:

[0150]

[0151] After converting the user's continuous feature matrix into a corresponding vector using the above method and inputting it into the first processing layer of the deep network, the intermediate processing vector output by the last processing layer after multiple processing layers is the first cross feature corresponding to the user's continuous behavior data.

[0152] This embodiment extracts continuous user behavior data vectors from the continuous user feature matrix and processes them through a multi-layer deep neural network. It can automatically learn and combine features based on the interaction relationships between the continuous user behavior data vectors, thereby uncovering deeply hidden features and combinations in the continuous user behavior data. This can effectively improve the user recognition accuracy and user recognition precision of the user recognition model trained subsequently.

[0153] In one embodiment, the user category feature matrix includes multiple user category feature vectors; the feature cross model based on the domain matrix factorization machine is used to obtain the user category feature domain corresponding to each user category feature vector according to the user category feature matrix, obtain the domain interaction matrix corresponding to every two user category feature domains according to the user category feature domains, obtain the cross representation of every two user category feature vectors according to the user category feature matrix and the domain interaction matrix, and obtain the second cross feature based on multiple cross representations.

[0154] Specifically, in this embodiment, the user category feature matrix can be represented by the following formula:

[0155]

[0156] Where M is the number of user category feature vectors, K is the dimension of each user category feature vector, and E m ={e m,1 ,...,e m,K} represents the feature vector of the m-th user category.

[0157] Each user category feature vector has its own feature domain. A feature domain categorizes user category feature vectors; a feature domain may contain only one type of user category feature vector or multiple types. For example, if occupation is used as a feature domain, it will only contain one type of user category feature vector corresponding to occupation; while if user attributes are used as feature domains, they can contain multiple user category feature vectors corresponding to region, occupation, interests, etc. Specifically, in this embodiment, analysts can pre-define different feature domains based on the characteristics of the financial industry and the financial system used, and determine the types of user category feature vectors contained in each feature domain.

[0158] Furthermore, after setting the feature domains, a domain interaction matrix can be set based on the interaction relationships between different feature domains. This domain interaction matrix includes multiple sub-matrices, and the interaction relationship between any two feature domains can be reflected by one of these sub-matrices. Specifically, after setting the feature domains, analysts can analyze the correlation between different feature domains based on the business characteristics of the financial industry or the financial system used, and further set the domain interaction matrix accordingly.

[0159] Based on this, in step S204 above, after inputting the user category feature matrix into the feature cross model based on the domain matrix factorization machine, the model first obtains the feature domain corresponding to each user category feature vector according to the pre-divided feature domain. Then, based on the domain interaction matrix, it performs feature cross on any two user category feature vectors in the user category feature matrix through the domain matrix factorization machine, and obtains the cross features between the two. Further, by combining the multiple cross representations obtained in this process, the second cross feature corresponding to the user category behavior data can be obtained. Specifically, this process can be expressed by the following formula:

[0160]

[0161] Among them, E i Let E be the feature vector of the i-th user category. j Let I be the feature vector of the j-th user category. i,j E represents the parameters of the domain interaction matrix between category feature i and category feature j, that is, the submatrix corresponding to the feature domains to which they belong in the domain interaction matrix. i,j This represents the cross-representation between user category feature vector i and user category feature vector j, where M is the number of user category feature vectors and C represents the second cross-feature corresponding to the user category behavior data.

[0162] This embodiment addresses the sparse, multi-feature-domain characteristics of user category behavior data by introducing a domain interaction matrix during feature crossing. This matrix maps user category feature vectors into a multi-feature-domain space, and then calculates the cross-interactions between user category feature vectors from different feature domains. This approach better captures the relationships between user category behavior data, effectively improving the user recognition accuracy and precision of the subsequently trained user recognition model. Furthermore, the domain interaction matrix allows user category feature vectors with different dimensions to cross-interact, ensuring effective cross-interaction while reducing the dimensionality of some user category feature vectors, resulting in lower time complexity and less memory usage.

[0163] In one embodiment, the feature cross model based on a multi-layer self-attention mechanism includes a multi-layer self-attention module; wherein, the self-attention module generates an interaction relationship vector and a value weight vector based on the input vector, and obtains an attention cross feature vector based on the interaction relationship vector and the value weight vector; the feature cross model based on a multi-layer self-attention mechanism is used to convert the first cross feature and the second cross feature into corresponding vectors and input them into the first-layer self-attention module, the first-layer self-attention module obtains the attention cross feature vector and transmits the attention cross feature vector to the next-layer self-attention module, and each subsequent layer self-attention module continues to generate a new attention cross feature vector based on the received attention cross feature vector, until the attention cross feature vector generated by the last self-attention module is used as the third cross feature.

[0164] Specifically, the feature cross model based on the multi-layer self-attention mechanism used in this embodiment includes a multi-layer self-attention module based on Transformer.

[0165] Each self-attention module performs operations based on the input vector to generate an interaction vector and a value weight vector, and further generates an attention cross-feature vector based on the interaction vector and the value weight vector. For example, this process can be represented by the following formula:

[0166]

[0167] Where A represents the processing procedure of the self-attention module, Q, K, and V are the query weight vector, key weight vector, and value weight vector of the self-attention module, respectively, and d k This represents the dimensions of the query weight vector Q and the key weight vector K. In this process, the attention module calculates three vectors Q, K, and V based on the input vector. Then, it performs a dot product between the query weight vector Q and the key weight vector K, and normalizes the result using a softmax function to ensure gradient stability, thus obtaining the interaction vector. Finally, it performs a dot product between the interaction vector and the value weight vector V and accumulates the results to obtain the attention cross-feature vector corresponding to this layer's self-attention module.

[0168] Based on this, in step S205 above, the feature cross model based on the multi-layer self-attention mechanism obtains the first cross feature D according to steps S203 and S204. dnn The second cross feature C and the first cross feature C are input together into the first layer self-attention module. Following the process described above, the corresponding attention cross feature vector for that layer is generated and then input into the second layer self-attention module. This process continues, with each layer self-attention module generating a new attention cross feature vector based on the received attention cross feature vector, until the attention cross feature vector generated by the last self-attention module is obtained, which is the third cross feature P.

[0169] In this embodiment, a multi-layer self-attention module is used to further fuse and learn the first cross-feature reflecting the interaction relationship of continuous user behavior data and the second cross-feature reflecting the interaction relationship of user category behavior data. This allows for a thorough learning of the interaction relationships between user behavior data of different natures, thereby obtaining a third cross-feature that more comprehensively reflects the user situation. Using this third cross-feature as input to the user recognition model to be trained helps to train a model with higher user recognition accuracy and precision.

[0170] In one embodiment, such as Figure 4 As shown, based on user behavior data samples, the following steps are taken to construct a continuous user feature matrix and a user category feature matrix:

[0171] Step S401: Extract multiple sub-behavior data based on the user behavior data sample.

[0172] Specifically, as described in step S202 of the above embodiment, the information in the user behavior data sample contains various different data such as region, occupation, savings, and consumption. The information provided by the user also contains some data that is not related to the training of the user identification model. Therefore, in step S401, the different data are first distinguished and identified in the data provided by the user, and the data to be used for model training is extracted from them. Each type of data is a sub-behavioral data.

[0173] Step S402: Obtain derived sub-behavior data based on multiple sub-behavior data.

[0174] To further improve the accuracy of the user recognition model in subsequent training, this step performs feature derivation based on the sub-behavior data obtained in step S401 to obtain derived sub-behavior data, thereby increasing the amount of data available for model training.

[0175] Specifically, in this embodiment, the third-party Python libraries Pandas and Numpy can be used to process the sub-behavioral data obtained in the previous step to obtain derivative sub-behavioral data, including the proportion of active accounts, the proportion of demand deposits to total assets, the proportion of time deposits to total assets, the proportion of wealth management to total assets, the proportion of investment to total assets, and the proportion of online transactions.

[0176] Step S403: Identify continuous behavior data and category behavior data in the sub-behavior data and derived sub-behavior data to form continuous behavior datasets and category behavior datasets.

[0177] Specifically, this step involves identifying and distinguishing continuous and categorical behavioral data within the identified sub-behavioral data and derived sub-behavioral data, and forming continuous behavioral datasets and categorical behavioral datasets based on the distinction results. Computers can be used to distinguish between continuous and categorical behavioral data. Considering the potential difficulties computers may face in directly recognizing textually represented data, the textual data can first be converted into numerical values; for example, for regional data, 1 can represent domestic and 2 can represent international. After this conversion, the computer can analyze sub-behavioral data and derived sub-behavioral data from multiple users, distinguishing between continuous and categorical behavioral data based on whether each data point is countable, and placing the corresponding data into either the continuous behavioral dataset or the categorical behavioral dataset.

[0178] Step S404: Construct a continuous user feature matrix based on the continuous behavior dataset, and construct a user category feature matrix based on the category behavior dataset.

[0179] After obtaining the continuous behavior dataset and the category behavior dataset according to the above steps, the user continuous feature matrix and the user category feature matrix can be constructed respectively based on the data in the dataset.

[0180] For continuous behavior datasets, each data point can be normalized to form a corresponding vector. The vectors from multiple continuous behavior data points can then form a user continuous feature matrix. For time-related continuous behavior data, the data can be divided according to time points to construct corresponding vectors. Based on the vectors corresponding to all continuous behavior data in the continuous behavior dataset, the user continuous feature matrix can be constructed.

[0181] For categorical behavior datasets, each categorical behavior data point can be first converted into a numerical value, and then a corresponding vector can be formed based on that value; alternatively, one-hot vectors can be used to encode the categorical behavior data and convert it into a corresponding vector. Based on the vectors corresponding to all categorical behavior data in the dataset, a user category feature matrix can be constructed.

[0182] This embodiment preprocesses and derives user behavior data samples provided by user authorization, effectively obtaining richer user behavior data. This increases the types of data available for subsequent feature cross-referencing, helping to obtain cross-features that better reflect the user's true situation in subsequent processing. This allows for the training of a user recognition model with higher user identification accuracy. Furthermore, this embodiment performs numerical conversion processing on user behavior data represented in text, enabling the computer to automatically identify and distinguish between continuous and categorical behavior data, thus improving the overall speed of the training process.

[0183] In one embodiment, constructing a user category feature matrix based on a category behavior dataset includes: performing word embedding processing on the category behavior data in the category behavior dataset to form a user category feature vector; and constructing a user category feature matrix based on the category behavior feature vectors corresponding to the category behavior dataset.

[0184] Specifically, in this embodiment, for the category behavior dataset, word embedding methods such as Word2Vec or GloVa can be used to process each category behavior data to generate a category behavior feature vector corresponding to that category behavior data. Then, a user category feature matrix is ​​constructed based on the category behavior feature vectors corresponding to all category behavior data in the category behavior dataset.

[0185] This embodiment uses word embedding to process categorical behavior data, which can better represent different words in the data and improve generalization ability. The categorical behavior feature vectors obtained by this method can better reflect the interaction relationship between different categorical behavior data in the subsequent feature crossing process, thereby obtaining more representative cross features and improving the user recognition accuracy and precision of the user recognition model trained subsequently.

[0186] In one embodiment, such as Figure 5 As shown, this application also provides a method for providing user identification results in a financial system, which can be applied to, for example... Figure 1 Terminal 102 in the application environment shown includes the following steps:

[0187] Step S501: After obtaining user behavior data for user authorization analysis, construct a continuous user feature matrix and a user category feature matrix based on the user behavior data.

[0188] Step S502: Input the user's continuous feature matrix into the feature cross model based on a deep neural network to obtain the first cross feature.

[0189] Step S503: Input the user category feature matrix into the feature cross model based on the domain matrix factorization machine to obtain the second cross feature;

[0190] Step S504: Input the first cross feature and the second cross feature into the feature cross model based on the multi-layer self-attention mechanism to obtain the third cross feature.

[0191] Step S505: Input the third cross feature into the trained user identification model and obtain the user label prediction value output by the trained user identification model.

[0192] Step S506: Obtain the user identification result based on the predicted user tag value.

[0193] Step S507: Display the user identification results on the page of the financial system that the user enters.

[0194] The trained user identification model is obtained according to the method in the above embodiments.

[0195] Specifically, in this embodiment, the user can interact with the financial system through terminal 102, authorize information analysis through the financial system's page, and enter other information they wish to provide besides the information already stored in the system. After obtaining user authorization, the entered information and the information within the authorized scope stored in the system can be combined to form user behavior data, and the method in this embodiment can be further applied to provide the user with user identification results.

[0196] The specific processing steps in S501 to S505 can be implemented using the methods in steps S202 to S206 of the above embodiments, or they can be implemented using related methods in other embodiments, which will not be elaborated here.

[0197] In step S506, after obtaining the predicted user tag value according to the previous steps, the user identification result can be obtained. For example, for the identification purpose of "whether the user needs to use service A in the financial system", after obtaining the predicted user tag value through the user identification model in step S505, the user identification result can be converted according to the size of the predicted value. For example, when the predicted user tag value is greater than 0.5, the identification result is "need".

[0198] After obtaining the user identification result through the above process, in step S507, the user identification result can be displayed on the user's interaction page with the financial system through text, images, or other means. The financial system can further provide personalized services to the user based on the user's feedback on the user identification result, thereby improving the user's service experience.

[0199] This embodiment employs different feature cross-methods for different types of behavioral data in user authorization analysis to obtain their respective cross-features. Further fusion learning is then performed on the cross-features corresponding to the two types of data to obtain cross-features that more realistically reflect the user's situation. Furthermore, inputting these cross-features into the user identification model trained using the above methods yields user identification results with higher accuracy, thereby enabling the provision of more personalized financial services tailored to user needs.

[0200] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0201] Based on the same inventive concept, this application also provides a financial system user identification model training device for implementing the above-mentioned financial system user identification model training method. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations in one or more embodiments of the financial system user identification model training device provided below can be found in the limitations of the financial system user identification model training method described above, and will not be repeated here.

[0202] In one embodiment, such as Figure 6 As shown, a user identification model training device 600 for a financial system is provided, comprising:

[0203] The acquisition module 601 is used to acquire user behavior data samples from the financial system and user tag sample values ​​corresponding to the user behavior data samples.

[0204] Construction module 602 is used to construct a continuous user feature matrix and a user category feature matrix based on the user behavior data samples;

[0205] The first cross module 603 is used to input the user continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0206] The second cross module 604 is used to input the user category feature matrix into a feature cross model based on a domain matrix factorization machine to obtain the second cross feature.

[0207] The third cross module 605 is used to input the first cross feature and the second cross feature into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0208] Prediction module 606 is used to input the third cross feature into the user identification model to be trained, and obtain the user label prediction value output by the user identification model to be trained;

[0209] The loss determination module 607 is used to determine the model loss based on the user tag sample values ​​and the user tag predicted values;

[0210] The adjustment module 608 is used to adjust the model parameters of the user recognition model to be trained according to the model loss, until the model loss meets the preset conditions to obtain the trained user recognition model.

[0211] In one embodiment, the deep neural network includes multiple processing layers; the feature cross model based on the deep neural network is used to convert the input user continuous feature matrix into a corresponding vector and transmit it to the first processing layer of the deep neural network, the first processing layer generates an intermediate processing vector for the input vector and transmits it to the next processing layer, each non-first processing layer generates a new intermediate processing vector based on the received intermediate processing vector and passes it to the next processing layer, until the last processing layer generates an intermediate processing vector as the first cross feature.

[0212] In one embodiment, the user category feature matrix includes multiple user category feature vectors; the feature cross model based on the domain matrix factorization machine is used to obtain the user category feature domain corresponding to each user category feature vector according to the user category feature matrix, obtain the domain interaction matrix corresponding to every two user category feature domains according to the user category feature domains, obtain the cross representation of every two user category feature vectors according to the user category feature matrix and the domain interaction matrix, and obtain the second cross feature according to multiple cross representations.

[0213] In one embodiment, the feature cross model based on a multi-layer self-attention mechanism includes a multi-layer self-attention module; wherein, the self-attention module generates an interaction relationship vector and a value weight vector based on the input vector, and obtains an attention cross feature vector based on the interaction relationship vector and the value weight vector; the feature cross model based on a multi-layer self-attention mechanism is used to convert the first cross feature and the second cross feature into corresponding vectors and input them into the first-layer self-attention module, the first-layer self-attention module obtains the attention cross feature vector and transmits the attention cross feature vector to the next-layer self-attention module, and each subsequent layer self-attention module continues to generate a new attention cross feature vector based on the received attention cross feature vector, until the attention cross feature vector generated by the last self-attention module is used as the third cross feature.

[0214] In one embodiment, the construction module is further configured to extract multiple sub-behavior data based on the user behavior data sample; obtain derived sub-behavior data based on the multiple sub-behavior data; identify continuous behavior data and category behavior data in the sub-behavior data and derived sub-behavior data to form a continuous behavior dataset and a category behavior dataset; construct a user continuous feature matrix based on the continuous behavior dataset, and construct a user category feature matrix based on the category behavior dataset.

[0215] In one embodiment, the construction module is further configured to perform word embedding processing on the category behavior data in the category behavior dataset to form a user category feature vector; and construct the user category feature matrix based on the category behavior feature vector corresponding to the category behavior dataset.

[0216] In one embodiment, this application also provides an apparatus for providing user identification results in a financial system, which is used to implement the method for providing user identification results in a financial system as described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more embodiments of the apparatus for providing user identification results in a financial system provided below can be found in the limitations of the method for providing user identification results in a financial system described above, and will not be repeated here.

[0217] In one embodiment, such as Figure 7 As shown, a device 700 for providing user identification results in a financial system is provided, comprising:

[0218] The construction module 701 is used to construct a continuous user feature matrix and a user category feature matrix based on the user behavior data after obtaining the user behavior data authorized for analysis.

[0219] The first cross module 702 is used to input the user continuous feature matrix into a feature cross model based on a deep neural network to obtain the first cross feature;

[0220] The second cross module 703 is used to input the user category feature matrix into a feature cross model based on a domain matrix factorization machine to obtain the second cross feature.

[0221] The third cross module 704 is used to input the first cross feature and the second cross feature into a feature cross model based on a multi-layer self-attention mechanism to obtain the third cross feature.

[0222] Prediction module 705 is used to input the third cross feature into the trained user identification model and obtain the user label prediction value output by the trained user identification model;

[0223] The identification module 706 is used to obtain the user identification result of the user based on the predicted value of the user tag;

[0224] Display module 707 is used to display the user identification results on the page of the financial system entered by the user;

[0225] The trained user identification model is obtained by training according to the method in the above embodiments.

[0226] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for training a user identification model in a financial system and a method for providing user identification results in a financial system. The display unit is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0227] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0228] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0229] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0230] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0231] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0232] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0233] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for training a user identification model in a financial system, characterized in that, The method comprises: obtaining a user behavior data sample of a financial system and a user label sample value corresponding to the user behavior data sample; wherein the user behavior data sample comprises personal information of a user and financial behavior information of the user; based on the user behavior data sample, constructing a user continuous feature matrix and a user category feature matrix; inputting the user continuous feature matrix into a feature cross model based on a deep neural network to obtain first cross features; inputting the user category feature matrix into a feature cross model based on a domain matrix factor decomposition machine to obtain second cross features; inputting the first cross features and the second cross features into a feature cross model based on a multi-layer self-attention mechanism to obtain third cross features; inputting the third cross features into a user identification model to be trained to obtain a user label prediction value output by the user identification model to be trained; determining a model loss according to the user label sample value and the user label prediction value; adjusting the model parameters of the user identification model to be trained according to the model loss until a trained user identification model is obtained when the model loss meets a preset condition; the method comprises: extracting a plurality of sub-behavior data according to the user behavior data sample; deriving sub-behavior data according to the plurality of sub-behavior data; identifying continuous behavior data and category behavior data in the sub-behavior data and the derived sub-behavior data to form a continuous behavior data set and a category behavior data set; constructing a user continuous feature matrix according to the continuous behavior data set, and constructing a user category feature matrix according to the category behavior data set.

2. The method of claim 1, wherein, The deep neural network comprises a plurality of processing layers; the feature cross model based on the deep neural network is used to convert the input user continuous feature matrix into a corresponding vector and transmit it to the first processing layer of the deep neural network, the first processing layer generates an intermediate processing vector for the input vector and transmits it to the next processing layer, each non-first processing layer generates a new intermediate processing vector according to the received intermediate processing vector and transmits it to the next processing layer, until the last processing layer generates an intermediate processing vector as the first cross features.

3. The method of claim 1, wherein, The user category feature matrix comprises a plurality of user category feature vectors; The feature cross model based on the domain matrix factor decomposition machine is used to obtain a user category feature domain corresponding to each user category feature vector according to the user category feature matrix, obtain a domain interaction matrix corresponding to each two user category feature domains according to the user category feature domain, obtain a cross representation of each two user category feature vectors according to the user category feature matrix and the domain interaction matrix, and obtain the second cross features according to a plurality of cross representations.

4. The method of claim 1, wherein, The feature cross model based on the multi-layer self-attention mechanism comprises a plurality of self-attention modules; wherein the self-attention module generates an interaction relationship vector and a value weight vector based on the input vector, and obtains an attention cross feature vector according to the interaction relationship vector and the value weight vector; The feature cross model based on the multi-layer self-attention mechanism is used to convert the first cross feature and the second cross feature into corresponding vectors and input the first-layer self-attention module, obtain an attention cross feature vector from the first-layer self-attention module, and transmit the attention cross feature vector to a next-layer self-attention module, continue to generate a new attention cross feature vector from each subsequent self-attention module according to the received attention cross feature vector, and until the attention cross feature vector generated by the last self-attention module is used as the third cross feature.

5. The method according to any one of claims 1 to 4, characterized in that, The user category feature matrix is constructed according to the category behavior data set, including: The category behavior data in the category behavior data set is subjected to word embedding processing to form a user category feature vector; The user category feature matrix is constructed according to the category behavior feature vector corresponding to the category behavior data set.

6. A method of providing a user identification result in a financial system, characterized by, The method includes: After obtaining the user behavior data of user authorization analysis, the user continuous feature matrix and the user category feature matrix are constructed according to the user behavior data; The user continuous feature matrix is input into the feature cross model based on the deep neural network to obtain the first cross feature; The user category feature matrix is input into the feature cross model based on the domain matrix factor decomposition machine to obtain the second cross feature; The first cross feature and the second cross feature are input into the feature cross model based on the multi-layer self-attention mechanism to obtain the third cross feature; The third cross feature is input into the trained user identification model to obtain the user label prediction value output by the trained user identification model; The user identification result of the user is obtained according to the user label prediction value; The user identification result is displayed in the page of the financial system entered by the user; The trained user identification model is trained according to the method in any one of claims 1 to 5. 7.A financial system user identification model training apparatus, characterized by, It includes: The acquisition module is used to acquire the user behavior data sample of the financial system and the user label sample value corresponding to the user behavior data sample; wherein the user behavior data sample includes personal information of the user and financial behavior information of the user; The construction module is used to construct the user continuous feature matrix and the user category feature matrix based on the user behavior data sample; The first cross module is used to input the user continuous feature matrix into the feature cross model based on the deep neural network to obtain the first cross feature; The second cross module is used to input the user category feature matrix into the feature cross model based on the domain matrix factor decomposition machine to obtain the second cross feature. The third cross module is used to input the first cross feature and the second cross feature into the feature cross model based on the multi-layer self-attention mechanism to obtain the third cross characteristic. The prediction module is used to input the third cross feature into the user identification model to be trained to obtain the user label prediction value output by the user identification model to be trained. The loss determination module is used to determine the model loss according to the user label sample value and the user label prediction value. an adjusting module configured to adjust model parameters of the user identification model to be trained according to the model loss until a trained user identification model is obtained when the model loss meets a preset condition; the constructing module is further configured to, extract a plurality of sub-behavior data according to the user behavior data sample; obtain derived sub-behavior data according to the plurality of sub-behavior data; identify continuous behavior data and category behavior data in the sub-behavior data and the derived sub-behavior data to form a continuous behavior data set and a category behavior data set; construct a user continuous feature matrix according to the continuous behavior data set, and construct a user category feature matrix according to the category behavior data set.

8. A device for providing user identification results in a financial system, characterized in that, comprise: a constructing module configured to, after obtaining user behavior data for user authorization analysis, construct a user continuous feature matrix and a user category feature matrix according to the user behavior data; a first cross module configured to input the user continuous feature matrix into a feature cross model based on a deep neural network to obtain first cross features; a second cross module configured to input the user category feature matrix into a feature cross model based on a domain matrix factor decomposition machine to obtain second cross features; a third cross module configured to input the first cross features and the second cross features into a feature cross model based on a multi-layer self-attention mechanism to obtain third cross features; a prediction module configured to input the third cross features into a trained user identification model to obtain a user label prediction value output by the trained user identification model; an identification module configured to obtain a user identification result of the user according to the user label prediction value; a display module configured to display the user identification result in a page of a financial system entered by the user; wherein the trained user identification model is trained according to the method of any one of claims 1 to 5. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

11. A computer program product comprising a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

Images