User data processing method and apparatus, computer device, and storage medium

Abstract

The application relates to the technical field of artificial intelligence, in particular to a user data processing method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring to-be-processed user data, wherein the to-be-processed user data comprises initial user data at different time points; performing graph construction on the initial user data at each time point to obtain a plurality of static knowledge graphs; performing feature extraction on the plurality of static knowledge graphs respectively to obtain static features corresponding to the plurality of static knowledge graphs, wherein the static features are used for representing feature data of fusing features of each node in the static knowledge graphs; concatenating the static features corresponding to each time point to obtain dynamic features; and obtaining target feature data corresponding to the to-be-processed user data according to the dynamic features. The method can accurately evaluate the to-be-processed user data.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a user data processing method, apparatus, computer equipment, storage medium, and computer program product. Background Technology

[0002] Loans are credit activities that involve lending monetary funds at a fixed interest rate and for a predetermined term; they are the most important item in commercial banks' asset operations. The profitability of loans is directly related to the loan price. Specifically, higher loan prices result in higher profits, but reduce loan demand. Conversely, lower loan prices result in lower profits, but increase loan demand. Therefore, it is necessary to set appropriate loan interest rates for different borrowers.

[0003] In related technologies, for example, in the 1940s, some banks in the United States began to study credit scoring methods to quickly process a large number of credit applications; in 1956, engineer Bill Fair and mathematician Earl Isaac jointly invented the famous FICO scoring method, which uses logistic regression as its core technology and is currently the most mature credit risk scoring model in the industry; in the 1960s to 1980s, with the advancement of information technology and the rapid development of business, credit scoring models were widely used in credit cards, consumer credit, mortgage loans and small business loans.

[0004] However, these methods have many limitations in the era of big data with its data explosion, resulting in low efficiency and inaccurate evaluation results. Summary of the Invention

[0005] Therefore, it is necessary to provide a user data processing method, apparatus, computer equipment, and storage medium that can accurately evaluate the user data to be processed, in order to address the aforementioned technical problems.

[0006] Firstly, this application provides a user data processing method, which includes:

[0007] Acquire user data to be processed, which includes initial user data at different points in time.

[0008] Multiple static knowledge graphs were obtained by constructing graphs from the initial user data at various time points.

[0009] Feature extraction is performed on multiple static knowledge graphs to obtain static features corresponding to multiple static knowledge graphs. Static features are used to characterize feature data that integrates the features of each node in the static knowledge graph.

[0010] The dynamic features are obtained by concatenating the static features corresponding to each time point;

[0011] The target feature data corresponding to the user data to be processed is obtained based on the dynamic features.

[0012] In one embodiment, the above-mentioned graph construction of initial user data at various time points yields multiple static knowledge graphs, including:

[0013] Initial user data at each time point is extracted to obtain triplet data;

[0014] Multiple static knowledge graphs were constructed based on triplet data.

[0015] In one embodiment, after constructing multiple static knowledge graphs based on triplet data, the process includes:

[0016] When a node in a static knowledge graph has a missing label, the triple data of the node with the missing label is read, and the node with the missing label is supplemented according to the label of the triple data.

[0017] In one embodiment, the above-mentioned feature extraction of multiple static knowledge graphs to obtain static features corresponding to multiple static knowledge graphs includes:

[0018] Based on multiple static knowledge graphs, the feature matrices corresponding to the multiple static knowledge graphs are obtained;

[0019] By fusing the features of each feature matrix, we can obtain static features corresponding to multiple static knowledge graphs.

[0020] In one embodiment, obtaining the target feature data corresponding to the user data to be processed based on dynamic features includes:

[0021] Feature extraction is performed on dynamic features to obtain retention features;

[0022] The retention features are updated to obtain the updated features;

[0023] The target feature data is obtained by calculating the updated features.

[0024] In one embodiment, the above-mentioned extraction of features from multiple static knowledge graphs to obtain the static features corresponding to multiple static knowledge graphs is achieved by a pre-trained first model;

[0025] Obtaining the target feature data corresponding to the user data to be processed based on dynamic features is achieved through a pre-trained second model.

[0026] In one embodiment, the training process of the first model and the second model includes:

[0027] Acquire sample data, which includes labeled data and feature labels;

[0028] The sample data is input into the first model so that the sample data can be extracted by the first model to obtain sample features;

[0029] The first objective loss function is calculated based on the sample features and the feature labels in the sample data. The first loss function is used to optimize the first model until the first model completes training.

[0030] The sample features are input into the second model so that the second model can predict the sample features and obtain the sample target features of the sample features;

[0031] The second target loss function is calculated based on the target features of the sample and the labeled data. The second loss function is used to optimize the second model until the second model is trained.

[0032] Secondly, this application also provides a risk management method, which includes:

[0033] Obtain the user data to be processed corresponding to the user to be predicted;

[0034] The target feature data corresponding to the user data to be processed is obtained according to the method in any of the above embodiments;

[0035] The risk level of the user to be predicted is determined based on the target feature data.

[0036] In one embodiment, determining the risk level of the user to be predicted based on target feature data includes:

[0037] Obtain the pre-set evaluation level;

[0038] The risk level of the user to be predicted is obtained based on the target feature data and the evaluation level.

[0039] Thirdly, this application also provides a risk assessment device, which includes:

[0040] The data acquisition module is used to acquire user data to be processed, which includes initial user data at different points in time.

[0041] The graph construction module is used to construct multiple static knowledge graphs from the initial user data at various time points.

[0042] The feature extraction module is used to extract features from multiple static knowledge graphs to obtain static features corresponding to the multiple static knowledge graphs. The static features are used to characterize the feature data of the fused static knowledge graphs.

[0043] The feature processing module is used to concatenate the static features corresponding to each time point to obtain dynamic features;

[0044] The target feature calculation module is used to obtain the target feature data corresponding to the user data to be processed based on dynamic features.

[0045] Fourthly, this application also provides a risk management apparatus, which includes:

[0046] The pending user data acquisition module is used to acquire pending user data corresponding to the users to be predicted.

[0047] The risk prediction module is used to obtain target feature data corresponding to the user data to be processed based on the device in any of the above embodiments.

[0048] The risk level assessment module is used to determine the risk level of the user to be predicted based on the target feature data.

[0049] 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 implement the steps of the methods in any of the above embodiments.

[0050] Sixthly, 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, implements the steps of the methods in any of the above embodiments.

[0051] In a seventh aspect, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the methods in any of the above embodiments.

[0052] The aforementioned user data processing method, apparatus, computer equipment, and storage medium first acquire user data to be processed, which includes initial user data at different time points. Then, multiple static knowledge graphs are constructed from the initial user data at each time point, allowing for the storage of large amounts of data and facilitating data extraction. Secondly, by establishing corresponding static knowledge graphs for the initial user data at each time point, a dynamic knowledge graph incorporating temporal information is obtained. This allows for the analysis of changes and trends in the graph structure over time, thereby grasping key information. Then, feature extraction is performed on the multiple static knowledge graphs to obtain corresponding static features. Even if there are missing features in the graphs, they can be filled in using the extracted static features. Finally, the static features corresponding to each time point are concatenated to obtain dynamic features, and prediction is performed based on these dynamic features to obtain target feature data. This allows for accurate prediction of the target feature data for the next time point based on the initial user data at each time point. Attached Figure Description

[0053] Figure 1 This is an application environment diagram of a user data processing method in one embodiment;

[0054] Figure 2 This is a flowchart illustrating a user data processing method in one embodiment;

[0055] Figure 3 This is a schematic diagram illustrating feature fusion using a first model in one embodiment;

[0056] Figure 4 This is a schematic diagram of the forget door operation in another embodiment;

[0057] Figure 5 This is a schematic diagram of an update gate operation in one embodiment;

[0058] Figure 6 This is a flowchart illustrating a risk handling method in one embodiment;

[0059] Figure 7 This is a schematic diagram illustrating risk level assessment in one embodiment;

[0060] Figure 8 This is a schematic diagram illustrating the training and prediction of a personal credit risk rating model in one embodiment;

[0061] Figure 9 This is a schematic diagram of the structure of a user data processing device in one embodiment;

[0062] Figure 10 This is a schematic diagram of the risk handling device in one embodiment;

[0063] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0064] 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.

[0065] The user data processing method provided in this application embodiment can be applied to, for example, Figure 1In 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 placed on a cloud or other network server. First, user data to be processed is acquired, including initial user data at different time points. Then, multiple static knowledge graphs are constructed from the initial user data at each time point, i.e., one static knowledge graph corresponds to one time point. Feature extraction is performed on the multiple static knowledge graphs to obtain static features corresponding to each static knowledge graph. Static features are used to represent the feature data of each node in the fused static knowledge graphs. The static features corresponding to each time point are concatenated to obtain dynamic features. Finally, the target feature data corresponding to the user data to be processed is obtained based on the dynamic features to achieve accurate evaluation of the user data to be processed. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, etc. Portable wearable devices can be smartwatches, smart bracelets, head-mounted devices, etc. Server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0066] In one embodiment, such as Figure 2 As shown, a user data processing method is provided, which is applied to... Figure 1 Taking server 104 as an example, the following steps are included:

[0067] S202, Obtain user data to be processed, which includes initial user data at different time points.

[0068] The user data to be processed refers to data that needs to be evaluated. This can be data from users requiring loan assessments, or other data requiring trust or risk assessments. For example, if the user data to be processed is from a user requiring a loan assessment, then the user data to be processed would include the user's gender, age, assets, industry, etc. The user data to be processed may include multiple initial user data records, all arranged in chronological order.

[0069] Initial data refers to raw data that has not undergone any processing, and it can be data directly exported from a database.

[0070] Specifically, the server first acquires initial user data at different time points, where one time point can correspond to one initial data point. In other embodiments, one time point can correspond to multiple initial data points, which is not specifically limited here. Then, the server summarizes and aggregates the acquired initial data from different time points to obtain user data to be processed. Optionally, the initial user data from different time points can be structured to obtain user data to be processed. In other embodiments, the initial user data from different times can also be filtered to remove invalid information to obtain user data to be processed.

[0071] S204: Graph construction is performed on the initial user data at each time point to obtain multiple static knowledge graphs.

[0072] Among them, static knowledge graphs are knowledge graphs constructed based on initial user data at various points in time. For example, one knowledge graph corresponds to one point in time. A knowledge graph is a graph constructed based on user data, where a point in the graph represents an entity and an edge represents the relationship between entities.

[0073] Specifically, the server processes the initial user data at various time points and constructs a knowledge graph. Optionally, the initial user data is distinguished into entities and relationships, and then a knowledge graph is constructed based on entities, relationships, and tags. In other embodiments, the initial user data can first be segmented into words, and then multiple static knowledge graphs can be constructed according to preset rules, i.e., one static knowledge graph corresponds to one time point. By analyzing the static knowledge graphs corresponding to different time points, the structural changes of the knowledge graph can be observed.

[0074] In one embodiment, when a node in the static knowledge graph has a missing label, the triple data containing the node with the missing label can be read, and then the label can be supplemented for that node based on the triple data containing the node with the missing label. In another embodiment, the server groups the initial user data according to access frequency and / or information importance, and places the initial user data with an access frequency greater than or equal to a first threshold in the static knowledge graph, while placing the data with an access frequency less than the first threshold, or whose information importance does not meet the requirements, in a traditional relational database. Specifically, frequently used information can be stored in the static knowledge graph, while information with low access frequency and irrelevant to relational analysis can be placed in a traditional relational database. In addition, strongly connected graphs can be found from the static knowledge graph and marked. A strongly connected graph means that every node can reach other nodes through some path, which indicates that there are strong relationships between these nodes, so further processing can be performed based on the strongly connected graph.

[0075] In one embodiment, multiple static knowledge graphs can be built from different dimensions at a single point in time. Specifically, if building a static knowledge graph related to finance, the server needs to acquire initial user data from customers, such as identity information, financial status, card information, transaction information, loan information, repayment information, loan application information, loan records, credit card records, quasi-credit card records, special transaction records, and query records. Besides financial information, customer social network information plays a crucial role. This is because modern transactions are increasingly completed online rather than face-to-face, thus understanding customer interests and emotions increasingly requires analyzing customer behavior data. By parsing the semantic meaning of customer textual data, such as customer service feedback, customer reviews on social media, and customer survey feedback, and tagging customers, a corresponding customer relationship graph can be built to gain better customer insights.

[0076] S206, feature extraction is performed on multiple static knowledge graphs to obtain static features corresponding to multiple static knowledge graphs. Static features are used to characterize feature data that integrates the features of each node in the static knowledge graph.

[0077] Static features refer to the features extracted from the static knowledge graph at a given time point. These features are used to characterize the feature data that integrates the features of each node in the static knowledge graph.

[0078] Specifically, the server extracts features from multiple static knowledge graphs. Optionally, a pre-trained machine learning model can be used to extract features from the static knowledge graphs, obtaining fused static features from each node in the static knowledge graphs. In other embodiments, a corresponding feature matrix can first be obtained from the static knowledge graphs, and then the feature matrix can be processed, for example, by inputting the feature matrix into a pre-trained machine learning model to obtain fused static features. In one embodiment, the customer feature stored in the static graph is X, and after feature extraction, the customer's feature becomes Z.

[0079] S208, the static features corresponding to each time point are concatenated to obtain the dynamic features.

[0080] Among them, dynamic features refer to features that include time-series information, which are static features linked together according to time sequence.

[0081] Specifically, the server obtains static features corresponding to multiple static knowledge graphs and concatenates them in chronological order to obtain dynamic features.

[0082] In one embodiment, if multiple static knowledge graphs are included at the same time point, for example, if the user-loan product graph and the user-user relationship graph are included at the first time point, after extracting features from these two graphs respectively, the static features corresponding to the two graphs are obtained. Then, these two static features are concatenated, and at the next time point, the static features are concatenated in the same way. Then, the static features are concatenated according to the time point order to obtain dynamic features.

[0083] In one embodiment, if multiple static knowledge graphs are included at the same time, the multiple static knowledge graphs can be spliced ​​together according to the way of constructing the graph, that is, the "entity-relationship-entity" method, that is, spliced ​​into a single static knowledge graph. Then, feature extraction is performed to obtain static features, and then the static features are concatenated according to the chronological order to obtain dynamic features.

[0084] In one embodiment, if multiple static knowledge graphs are included at the same time point, features can be extracted from the multiple static knowledge graphs at the same time point to obtain multiple static features at that time point. Then, the multiple static features corresponding to each time point are concatenated in the order of their appearance to obtain dynamic features.

[0085] S210, obtain the target feature data corresponding to the user data to be processed based on the dynamic features.

[0086] Among them, target feature data refers to the prediction results used to characterize the user data to be processed after processing.

[0087] Specifically, the server makes predictions based on dynamic features of the user data to be processed, thereby obtaining the target feature data corresponding to the user data to be processed. Optionally, the dynamic features can be input into a pre-trained machine learning model, which can predict the features at the next point in time based on the dynamic features, i.e., the target feature data.

[0088] In one embodiment, if the goal is to predict an individual's credit risk level, a knowledge graph can be constructed based on user data to obtain multiple static knowledge graphs. Features are then extracted from these static knowledge graphs to obtain static features. These static features are then concatenated in chronological order to obtain dynamic features. Finally, the target feature data, i.e., the predicted risk score for the user, is predicted based on the dynamic features. Combined with the characteristics of the loan product, a corresponding loan interest rate can be formulated for the customer. Simultaneously, follow-up investigations and feedback are conducted on the customer, and the knowledge graph information is updated in real time to optimize the prediction of the individual's credit risk level.

[0089] The aforementioned user data processing method first acquires the user data to be processed, which includes initial user data at different time points. Then, it constructs multiple static knowledge graphs from the initial user data at each time point, thus storing a large amount of data and facilitating data extraction. Secondly, by establishing corresponding static knowledge graphs for the initial user data at each time point, a dynamic knowledge graph incorporating temporal information is obtained. This allows for analysis of the graph structure's changes and trends over time, thereby grasping key information. Then, feature extraction is performed on the multiple static knowledge graphs to obtain corresponding static features. Even if there are missing features in the graph, they can be filled in using the extracted static features. Finally, the static features corresponding to each time point are concatenated to obtain dynamic features, and prediction is performed based on these dynamic features to obtain target feature data. This allows for accurate prediction of the target feature data for the next time point based on the initial user data at each time point.

[0090] In one embodiment, the above-mentioned graph construction of initial user data at each time point to obtain multiple static knowledge graphs includes: extracting initial user data at each time point to obtain triple data; and constructing graphs based on triple data to obtain multiple static knowledge graphs.

[0091] Among them, triplet data refers to the data set used to construct a static knowledge graph, which includes at least two entities and the relationship information between the entities. Optionally, the entities can be obtained by segmenting user data.

[0092] Specifically, initial user data at various times is extracted to obtain multiple triple data. Optionally, the initial user data can be segmented to obtain individual segmented data. Based on a preset entity information table, relation information table, and tag system information table, it is determined whether each segmented word is an entity, a relation, or an entity's tag. The entity-relation-entity pairs are then connected in a certain direction to obtain triple data.

[0093] Specifically, the obtained triple data is still connected according to the entity-relation-entity method to obtain a static knowledge graph.

[0094] In one embodiment, the server can extract initial user data for each time period in chronological order to obtain multiple triple data, and construct multiple static knowledge graphs corresponding to each time point based on the triple data.

[0095] In one embodiment, the server has multiple threads that simultaneously extract initial user data at different times to obtain multiple triple data sets. Based on the triple data sets, a knowledge graph is constructed, resulting in multiple static knowledge graphs corresponding to each time point.

[0096] In the above embodiments, the server extracts the initial user data at each time point to obtain triple data, and constructs multiple static knowledge graphs based on the triple data. In this way, multiple static knowledge graphs can be analyzed to predict target feature data from the unprocessed user data.

[0097] In one embodiment, after constructing multiple static knowledge graphs based on triple data, the process includes: when a node in the static knowledge graph has a missing label, reading the triple data of the node with the missing label, and supplementing the node with the missing label according to the label of the triple data.

[0098] In this context, tags refer to the attributes of nodes in a static knowledge graph. Based on the entity-relationship-entity structure, we can also obtain a tag-relationship-tag structure, which facilitates deeper analysis. It should be noted that an entity can have one or more tags.

[0099] Specifically, when any node in a static knowledge graph has a missing label, the triple data containing the node with the missing label can be read, and the label for that node can be supplemented based on the triple data containing the node with the missing label. For example, if A and B are a triple data pair, and the label of A is missing, then the label of B can be used as the label of A.

[0100] In one embodiment, when there are multiple triples associated with an entity whose label is missing, the label can be supplemented based on the labels of the multiple triples, for example, by performing feature fusion or key feature extraction on the labels of the multiple triples to obtain the label of the node with missing label.

[0101] In the above embodiments, by utilizing the relationship between labeled nodes and nodes with missing labels to infer and supplement the labels of nodes with missing labels, the data can be made more complete, and the prediction of user data to be processed can be more accurate.

[0102] In one embodiment, the above-mentioned extraction of features from multiple static knowledge graphs to obtain static features corresponding to multiple static knowledge graphs includes: obtaining feature matrices corresponding to multiple static knowledge graphs based on multiple static knowledge graphs; and fusing the feature matrices to obtain static features corresponding to multiple static knowledge graphs.

[0103] The feature matrix refers to the matrix obtained from a static knowledge graph. For example, in a static knowledge graph with N nodes, each node has its own features, and these features can be combined to form a feature matrix. A feature matrix X of dimension D, where D represents the dimension value of the feature.

[0104] Specifically, the server first extracts features from multiple static knowledge graphs to obtain feature matrices corresponding to the multiple static knowledge graphs. Then, it fuses the features of each feature matrix to obtain static features corresponding to the multiple static knowledge graphs. Optionally, the feature matrices can be fused using a pre-trained machine learning model to obtain static features corresponding to the multiple static knowledge graphs.

[0105] In one embodiment, a graph convolutional neural network (a type of neural network model, GCN) can be used to train a first model capable of feature fusion of the feature matrix. Specifically, in a static knowledge graph with N nodes, the features of these nodes can form a feature fusion matrix. The feature matrix X and the relationships between nodes form an N×N adjacency matrix A. The feature matrix X and adjacency matrix A are then input into the first model. The propagation between network layers is as follows:

[0106]

[0107] in, , It is an identity matrix. yes The degree matrix, . These are features of the l-th layer. . These are the parameters for each layer. It is a non-linear activation function.

[0108] exist Figure 3 middle, Figure 3 This is a schematic diagram illustrating feature fusion using a first model in one embodiment. The stored customer features in the diagram are X, and through several layers of a graph neural network, each customer's features are transformed into Z. In this embodiment, the first model employs a two-layer graph convolutional neural network, using ReLU and Softmax activation functions respectively. The overall forward propagation formula is:

[0109]

[0110]

[0111] Calculate the cross-entropy loss function for all labeled nodes:

[0112]

[0113] in It is a collection of labeled nodes. Network parameters. and The training employed gradient descent.

[0114] In one embodiment, a graph neural network (a neural network model, GNN) can be used to train a first model capable of feature fusion of the feature matrix.

[0115] In the above embodiments, feature fusion can utilize information from nodes and edges to complete missing data in static knowledge graphs.

[0116] In one embodiment, obtaining the target feature data corresponding to the user data to be processed based on dynamic features includes: extracting features from the dynamic features to obtain retention features; updating the retention features to obtain updated features; and calculating the target feature data based on the updated features.

[0117] Among them, retained features refer to the features after feature extraction of dynamic features. They can determine which information can be retained by a pre-set mask, or determine which information needs to be "forgotten" by the forget function (forget gate) in the pre-trained prediction model. That is, which information needs to be retained and which information needs to be "discarded". Updated features refer to the features obtained after updating based on retained features.

[0118] Specifically, after the server obtains the dynamic features, it first extracts features from the dynamic features to obtain the retention features, and then updates the features based on the retention features to obtain the updated features. Optionally, the retention features can be updated using the update gate of a pre-trained prediction model to obtain the updated features. Finally, the target feature data is calculated based on the updated features. Optionally, the target feature data can be calculated using the state at the previous time step and the input at the current time step.

[0119] In one embodiment, target feature data can be obtained from dynamic features using a pre-trained prediction model. Specifically, dynamic features, i.e., the graph node structure and internal structure at different times, are used as input to the pre-trained prediction model. The first step of the pre-trained prediction model is to determine the information to be discarded through a forget gate layer, which reads... and Output a value between 0 and 1 to the state of each module. The numbers in the text. 1 means "keep completely", 0 means "discard completely". (Specific combination...) Figure 4 As shown, Figure 4Here is a schematic diagram of the forget door operation in one embodiment, which is as follows:

[0120]

[0121] in, This represents the state at time t-1. This represents the input at time t. This represents the forget gate at time t. This indicates that the constant term of the gate is forgotten. This indicates that the weight parameters of the forgotten gate are not specified.

[0122] The next step after training the prediction model is to determine what new information is stored in the modules. This involves two parts: the SigmaDouble layer, or "input gate layer," which determines which parameter values ​​need to be updated, and the tanh layer, which creates a new candidate vector. Add it to the state.

[0123]

[0124]

[0125] in This represents the update gate at time t. This represents the constant term of the update gate. Represents the candidate value vector The constant term, This indicates that the weight parameters of the gate are being updated. Represents the candidate value vector The weight parameters.

[0126] at this time, Updated to Then combine the old state with... Multiply, discarding information that needs to be discarded. Add... This yields new candidate values, specifically combined with... Figure 5 As shown, Figure 5 Here is a schematic diagram of an update gate operation in one embodiment, which is as follows:

[0127]

[0128] in, It represents the current state of memory, and also the state of memory cells after the update.

[0129] Finally, determine the output value. .

[0130]

[0131] in, This represents the output gate at time t. This represents the constant term of the output gate. This represents the weight parameters of the output gate.

[0132] At this point, the prediction of dynamic features is complete.

[0133] In the above embodiments, by extracting features from dynamic features to obtain retained features, updating the retained features to obtain updated features, and finally calculating based on the updated features, the target feature data can be accurately obtained.

[0134] In one embodiment, the above-mentioned extraction of features from multiple static knowledge graphs to obtain static features corresponding to multiple static knowledge graphs is achieved through a pre-trained first model; the acquisition of target feature data corresponding to the user data to be processed based on dynamic features is achieved through a pre-trained second model; wherein the training process of the first model and the second model includes: acquiring sample data, the sample data carrying labeled data and feature labels; inputting the sample data into the first model, the first model extracting from the sample data to obtain sample features; inputting the sample features into the second model, the second model predicting the sample features to obtain the sample target features of the sample features.

[0135] The first model refers to a machine learning model that extracts static features from a static knowledge graph. It can be trained based on any machine learning model that can process graphs, such as GCN or GNN. The second model refers to a machine learning model that predicts based on the static features obtained from the first model, such as RNN, LSTM, or GRU.

[0136] Here, sample data refers to the training data used to train the first and second models. It can be crawled from a pre-set database table or from publicly available data crawled from the Internet. Multiple static knowledge graphs are built based on the above data. Feature labels refer to the labels of nodes in the sample data, i.e., the labels of nodes in the multiple static knowledge graphs. They can be used to calculate the loss function of the first model, i.e., the first loss function. Labeled data refers to the real labels of the sample data that have been pre-labeled. They can be used to calculate the loss function of the second model, i.e., the second loss function, to guide the model optimization. Sample target features refer to the data predicted by the second model based on the sample data.

[0137] Specifically, firstly, sample data is acquired, carrying labeled data. The labeled data is pre-labeled; for example, to predict a user's personal credit risk level, the labeled data would be a customer rating. In other embodiments, the labeled data is based on the actual usage scenario. Then, the sample data is input into a first model. Optionally, before inputting the sample data into the first model, multiple static knowledge graphs need to be constructed based on the sample data. This is because knowledge graphs allow for clearer observation of data changes and are more beneficial for data analysis and prediction. Then, the multiple static knowledge graphs are input into the first model. Preferably, the first model is a graph convolutional neural network (GCNN). A GCNN is a neural network model specifically designed for processing graph-structured data. It utilizes different neurons in a deep neural network to learn the topological and attribute feature information in the graph data, and then integrates them to obtain a more refined feature representation of nodes or substructures, i.e., obtaining the sample features after feature extraction.

[0138] Specifically, after obtaining the sample features, the sample features are input into the second model. Optionally, before inputting the sample features into the second model, the feature vectors corresponding to multiple static knowledge graphs need to be concatenated in chronological order to obtain dynamic sample features. The dynamic sample features are then input into the second model, which makes predictions based on the sample features to obtain the sample target features. For example, the sample features from time t to t+3 are input, and the sample target features at time t+4 are obtained through prediction by the second model.

[0139] In one embodiment, the first model calculates a first objective loss function based on sample features and feature labels in the sample data. The first objective loss function guides the optimization of the first model until the first model completes training. The second model calculates a second objective loss function on the updated feature model branch based on the sample target features and labeled data. The second loss function is used to guide the optimization of the updated feature model branch until the updated feature model branch completes training.

[0140] In one embodiment, before inputting the sample data into the first model, the missing feature labels of the nodes in the sample data can be supplemented. Specifically, the triplet data of the nodes with missing feature labels can be read, and the nodes with missing feature labels can be supplemented according to the labels of the triplet data.

[0141] In the above embodiments, by training the first model and the second model, a first model capable of extracting features from data features in a static knowledge graph and a second model capable of predicting features extracted by the first model can be obtained.

[0142] In one embodiment, after inputting sample data into a first model and extracting sample features from the sample data, the method further includes: calculating a first target loss function based on the sample features and feature labels in the sample data. The first target loss function is used to guide the optimization of the first model until the first model completes training. After inputting the sample features into a second model and predicting the sample features to obtain the sample target features, the method further includes: calculating a second target loss function for the updated feature model branch based on the sample target features and labeled data. The second loss function is used to guide the optimization of the updated feature model branch until the updated feature model branch completes training.

[0143] The first loss function is a function used to guide the optimization of the parameters in the first model. It can be calculated based on the true value and the predicted value, i.e., the feature labels and sample features in the sample data. The second loss function is a function used to guide the optimization of the parameters in the second model. It can be calculated based on the expressed data and the target features of the sample.

[0144] Specifically, the first model calculates the first target loss function based on the extracted sample features and the feature labels in the sample data, i.e., the labels carried by the nodes in the static knowledge graph. Optionally, the first target loss function can be calculated based on the sample features, the feature labels in the sample data, and the cross entropy loss function. It should be noted that in other embodiments, the first target loss function can be calculated based on the sample features, the feature labels in the sample data, and any one of the loss functions.

[0145] Specifically, the second model calculates a second objective loss function based on the labeled data and the target features of the samples. Optionally, the second objective loss function can be calculated based on the labeled data, the target features of the samples, and the `categorical_crossentropy` function. It should be noted that in other embodiments, the second objective loss function can be calculated based on the labeled data, the target features of the samples, and any one of the loss functions. In other embodiments, if it is necessary to train a personal credit risk rating model, then the labeled data of the samples is the actual credit score; therefore, the second objective loss function can be calculated using the labeled data and the target features of the samples.

[0146] In the above embodiments, the first model and the second model can be optimized respectively by calculating the first target loss function and the second loss function, making the parameters of the trained first model and the second model more accurate.

[0147] In one embodiment, such as Figure 6 As shown, a risk management method is provided, including the following steps:

[0148] S702, Obtain the user data to be processed corresponding to the user to be predicted.

[0149] Among them, the users to be predicted refer to customers who need to undergo trust assessment, and the user data to be processed refers to the basic information of the users to be predicted, such as identity information, property status, card information, transaction information, loan information, repayment information, loan application information, loan records, credit card records, quasi-credit card records, special transaction records and query records, as well as customer behavior data, because in addition to information in the financial field, information from the customer's social network plays a key role.

[0150] Specifically, the server can retrieve user data to be processed corresponding to the user to be predicted from the customer information database, and build multiple static knowledge graphs based on the user data to be processed corresponding to the user to be predicted.

[0151] S704, Obtain the target feature data corresponding to the user data to be processed according to the method in any of the above embodiments.

[0152] The target feature data refers to a customer's credit score in a lending scenario. Based on this score, the customer's credit rating can be further determined. Optionally, the credit rating of the user to be predicted can be determined using pre-set evaluation levels and the target feature data. Specifically, the target feature data corresponding to the user data to be processed can be obtained based on multiple static knowledge graphs corresponding to the user to be predicted, according to any of the methods described above.

[0153] In one embodiment, multiple static knowledge graphs corresponding to the user to be predicted are input into a pre-trained personal credit risk rating model. The pre-trained personal credit risk rating model includes a first model and a second model. The specific training process can be referred to any of the above embodiments, and will not be repeated here. The credit score of the customer to be predicted can be obtained through the prediction of the personal credit risk rating model.

[0154] S706, determine the risk level of the user to be predicted based on the target feature data.

[0155] Risk level refers to the quantitative standard for judging customer risk. For example, risk level can be trusted safe user, creditworthy user, credit questionable user, and credit untrustworthy user.

[0156] Specifically, after obtaining the target feature data corresponding to the user data to be processed, the server determines the risk level of the user to be predicted based on the target feature data. Optionally, the credit level of the user to be predicted can be determined by a pre-set evaluation level and the target feature data.

[0157] In one embodiment, a specific combination Figure 7 As shown, Figure 7 As illustrated in one embodiment of risk level assessment, customer information is first collected. Then, multiple static knowledge graphs (i.e., dynamic knowledge graphs) are built based on this information. These static knowledge graphs include two types of graphs: user-loan relationship graphs and user-user relationship graphs. This is because modern transactions are increasingly conducted online rather than face-to-face, making it crucial to analyze customer behavior data to understand customer interests and emotions. Therefore, user-user relationship graphs are built simultaneously with user-loan relationship graphs to gain better customer insights. The dynamic knowledge graphs are then processed by a first model to obtain static features, specifically the node embeddings obtained from the results and internal features of nodes in the graph at each time step t. These static features are then concatenated to obtain dynamic features, which are input into a second model. The dynamic features are then processed by the second model to obtain target feature data, thus completing the prediction of the personal risk level model. The model training process is similar to the model usage process, except that the labeled sample data is known, and the model branches need to be optimized using objective loss functions. Specifically, the first objective loss function optimizes the first model, and the second objective loss function optimizes the second model, until both models are trained. This completes the training of the personal credit risk level model.

[0158] In one embodiment, the predicted target feature data of the predicted user can be placed into a pre-selected data table for data updating.

[0159] In the above embodiments, predictions can be made based on the user data to be processed corresponding to the user to be predicted, and the risk level can be accurately obtained so as to formulate loan interest rates according to the risk level. For banks, this can greatly reduce the risk of customer credit default.

[0160] In one embodiment, determining the risk level of the user to be predicted based on the target feature data includes: obtaining a pre-set evaluation level; and obtaining the risk level of the user to be predicted based on the target feature data and the evaluation level.

[0161] Among them, the rating level refers to the quantitative standard used to measure the creditworthiness of the user to be predicted, which can be set according to the actual usage scenario.

[0162] Specifically, the server first obtains a pre-set evaluation level, which includes details of the credit rating corresponding to each credit score. Then, based on the target feature data and the evaluation level, it obtains the predicted risk level of the user. In one embodiment, the pre-set evaluation level is 85-100 for a trusted and safe user, and the target feature data is 92. In this case, the user's risk level is a trusted and safe user.

[0163] In the above embodiments, the risk level of the user to be predicted is obtained by pre-setting the evaluation level. In this way, the corresponding loan interest rate can be formulated for the customer based on the risk level and the characteristics of the loan product.

[0164] In one embodiment, combined Figure 8 , Figure 8 This is a schematic diagram illustrating the training and prediction of a personal credit risk rating model in one embodiment, including the following steps:

[0165] Step 1: Construct a customer information database based on a dynamic knowledge graph. Acquire basic customer information, such as identity information, financial status, card information, transaction information, loan disbursement information, repayment information, loan application information, loan records, credit card records, quasi-credit card records, special transaction records, and query records. Beyond financial information, customer social network information plays a crucial role. Modern transactions are increasingly completed online rather than face-to-face, making it increasingly necessary to analyze customer behavior data to understand customer interests and emotions. By analyzing the semantic meaning of textual customer data, such as customer service feedback, social media reviews, and survey feedback, customer tags are created to build user profiles. Simultaneously, knowledge graph technology is used to build a customer relationship graph for better customer insights. This includes customer interest insights (interest in loan products) for personalized product recommendations and targeted marketing, and customer attitude insights (satisfaction with the company and services, suggestions for improvement) for rapid response to customer issues, improved customer experience, strengthened customer relationships, and increased customer loyalty. The above analyses are all based on static relationship graphs and do not consider that the graph structure changes over time, but these changes are themselves related to risk. Therefore, time-series information needs to be incorporated when constructing a knowledge graph to create a dynamic one. The main function of a knowledge graph is to analyze relationships, especially deep relationships.

[0166] Step Two: Feature Extraction Based on Graph Convolutional Neural Networks and Recurrent Neural Networks. The dynamic knowledge graph obtained in Step One is decomposed into two parts for processing. First, features are extracted from the static knowledge graph. Then, the node structures and internal features obtained at different time points are concatenated to learn dynamic characteristics.

[0167] Static knowledge graphs, as a type of graph structure, can be used for feature extraction using a first model, which in this embodiment is a graph convolutional neural network (GCN). First, the dynamic knowledge graph obtained in step one is used as input to the GCN. This graph structure has N nodes, each with its own features. These features form an N×D matrix X, and the relationships between nodes form an N×N matrix A, also known as the adjacency matrix. Matrix X and matrix A are the inputs to the model. The first model can obtain static features corresponding to multiple static knowledge graphs, and these static features are then concatenated to obtain dynamic features.

[0168] The graph convolutional neural network described above can obtain the structure and internal features of each graph node at time t, and then input the dynamic features into the second model. For processing temporal information, LSTM is a good choice; therefore, in this embodiment, the second model is LSTM. The input of LSTM includes not only current information but also past information, meaning it can be used to connect previous information to the current task, such as using past information to infer understanding of current information. Furthermore, compared to basic RNNs, LSTM solves the problem of long-term dependencies. The problem of long-term dependencies arises because as the intervals increase, RNNs lose the ability to learn connections between such distant information. The target feature data can be predicted using the LSTM model.

[0169] Step 3: Training the Personal Credit Risk Rating Model. Using the customer features obtained in Step 2 as input to the convolutional neural network and the customer's personal credit risk rating as labeled data, the model is trained to obtain the personal credit risk rating model.

[0170] Step Four: Prediction by the Personal Credit Risk Rating Model. When processing personal loan applications, it is first necessary to understand the customer's personal credit risk level. At this point, simply inputting the customer's corresponding dynamic knowledge graph into the model will allow the prediction of the customer's personal credit risk level.

[0171] Step Five: Intelligent Interest Rate Pricing. In personal loan business, when a customer selects a loan product, the customer's personal credit risk level obtained in Step Four, combined with the characteristics of the loan product, allows for the formulation of a corresponding loan interest rate for that customer. Simultaneously, follow-up investigations and feedback are conducted on the customer, and the credit information is updated in real time to further optimize the model.

[0172] In the above embodiments, temporal information is incorporated when extracting key features from customer information, and the customer's personal credit risk level is assessed using information from the knowledge graph at different times. The processing of the dynamic knowledge graph mainly employs graph convolutional neural network (GCNN) and recurrent neural network (RNN) techniques. Features that integrate temporal information and are processed by GCNN and LSTM can significantly improve the accuracy of personal credit risk assessment, thereby enhancing the precision of intelligent loan interest rate pricing. For banks, this can greatly reduce the risk of customer loan defaults. Furthermore, differentiated pricing for different customer groups can increase the willingness of new customers to apply, reactivate dormant customers, promote repeat loans from existing customers, strengthen the ability to manage customers with refined products, and improve the overall business contribution of customers. In addition, addressing the problem that the storage and processing of financial data cannot meet current needs, a static knowledge graph corresponding to multiple time points—that is, a dynamic knowledge graph—is constructed to store customer information at different times. The graph incorporating temporal information improves the accuracy of subsequent personal credit risk assessment and intelligent loan interest rate pricing. In addition, after setting the corresponding loan interest rate for the customer, the customer's data is updated in the graph in real time through subsequent information tracking to further optimize and improve the model performance; third, the GCN network and LSTM network are combined as feature extractors, which can not only accurately and quickly extract the features of the graph structure, but also incorporate time series information to facilitate analysis.

[0173] 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.

[0174] Based on the same inventive concept, this application also provides a user data processing apparatus for implementing the user data processing method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more user data processing apparatus embodiments provided below can be found in the limitations of the user data processing method described above, and will not be repeated here.

[0175] In one embodiment, such as Figure 9As shown, a user data processing device is provided, including: a data acquisition module 100, a map construction module 200, a feature extraction module 300, a feature processing module 400, and a target feature calculation module 500, wherein:

[0176] The data acquisition module 100 is used to acquire user data to be processed, which includes initial user data at different time points.

[0177] The graph construction module 200 is used to construct multiple static knowledge graphs from the initial user data at various time points.

[0178] The feature extraction module 300 is used to extract features from multiple static knowledge graphs to obtain static features corresponding to the multiple static knowledge graphs. The static features are used to characterize the feature data of each node in the fused static knowledge graph.

[0179] The feature processing module 400 is used to concatenate the static features corresponding to each time point to obtain dynamic features.

[0180] The target feature calculation module 500 is used to obtain the target feature data corresponding to the user data to be processed based on dynamic features.

[0181] In one embodiment, the above-mentioned map construction module 200 includes:

[0182] The triplet extraction unit is used to extract initial user data at each time point to obtain triplet data.

[0183] The static graph construction unit is used to construct multiple static knowledge graphs based on triplet data.

[0184] In one embodiment, the above-mentioned map construction module 200 further includes:

[0185] The label supplementation unit is used to read the triple data of the node with missing labels when there are nodes with missing labels in the static knowledge graph, and supplement the nodes with missing labels according to the labels of the triple data.

[0186] In one embodiment, the feature extraction module 300 includes:

[0187] The feature matrix extraction unit is used to obtain the feature matrices corresponding to multiple static knowledge graphs.

[0188] The static feature extraction unit is used to fuse various feature matrices to obtain static features corresponding to multiple static knowledge graphs.

[0189] In one embodiment, the feature processing module 400 includes:

[0190] The retention feature extraction unit is used to extract features from dynamic features to obtain retention features.

[0191] The feature extraction unit is used to update the retained features to obtain updated features;

[0192] The feature calculation unit is used to calculate the target feature data based on the updated features.

[0193] In one embodiment, the above-mentioned apparatus further includes:

[0194] The sample acquisition module is used to acquire sample data, which includes labeled data and feature labels.

[0195] The sample feature acquisition module is used to input sample data into the first model, and the first model extracts sample features from the sample data.

[0196] The first model optimization module is used to calculate the first objective loss function based on the sample features and feature labels. The first objective loss function is used to optimize the first model until the first model completes training.

[0197] The sample target feature calculation module is used to input sample features into the second model, and the second model predicts the sample features to obtain the sample target features of the sample features.

[0198] The second model optimization module is used to calculate the second target loss function based on the sample target features and labeled data. The second loss function is used to optimize the second model until the second model completes training.

[0199] In one embodiment, such as Figure 10 As shown, a risk processing device is provided, including: a user data acquisition module 600, a risk prediction module 700, and a risk level judgment module 800, wherein:

[0200] The pending user data acquisition module 600 is used to acquire pending user data corresponding to the users to be predicted.

[0201] The risk prediction module 700 is used to obtain target feature data corresponding to the user data to be processed based on the device in any of the above embodiments.

[0202] The risk level assessment module 800 is used to determine the risk level of the user to be predicted based on the target feature data.

[0203] In one embodiment, the risk level assessment module 800 includes:

[0204] The evaluation level acquisition unit is used to acquire the pre-set evaluation level.

[0205] The risk assessment unit is used to determine the risk level of the user to be predicted based on the target feature data and the evaluation level.

[0206] Each module in the aforementioned user data processing device and risk processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0207] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores user data to be processed and user data corresponding to users to be predicted. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a user data processing method and a risk processing method.

[0208] Those skilled in the art will understand that Figure 11 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.

[0209] In one embodiment, a computer device is 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-described method embodiments.

[0210] 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.

[0211] 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.

[0212] Those skilled in the art will understand that all or part of the processes in 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. When executed, the computer program can include the processes of the embodiments described above. 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.

[0213] 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.

[0214] 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 patent 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 user data processing method, characterized in that, The method includes: Acquire user data to be processed, which includes initial user data at different points in time; Initial user data at each of the aforementioned time points is extracted to obtain triplet data; Multiple static knowledge graphs are constructed based on the triplet data, with one static knowledge graph corresponding to one time point. When a node in the static knowledge graph has a missing label, the triple data of the node with the missing label is read, and the node with the missing label is supplemented according to the label of the triple data; Each of the static knowledge graphs is subjected to feature extraction to obtain static features corresponding to the static knowledge graphs. The static features are used to characterize the feature data that integrates the features of each node in the static knowledge graph. The dynamic features are obtained by concatenating the static features corresponding to each of the aforementioned time points. The target feature data corresponding to the user data to be processed is obtained based on the dynamic features.

2. The method according to claim 1, characterized in that, The step of extracting features from multiple static knowledge graphs to obtain static features corresponding to multiple static knowledge graphs includes: Based on the multiple static knowledge graphs, the feature matrices corresponding to the multiple static knowledge graphs are obtained; The feature matrices are fused to obtain the static features corresponding to the multiple static knowledge graphs.

3. The method according to claim 1, characterized in that, The step of obtaining the target feature data corresponding to the user data to be processed based on the dynamic features includes: Feature extraction is performed on the dynamic features to obtain the retention features; The retention features are updated to obtain the updated features; The target feature data is calculated based on the updated features.

4. The method according to claim 1, characterized in that, The step of extracting features from multiple static knowledge graphs to obtain static features corresponding to multiple static knowledge graphs is achieved through a pre-trained first model; The process of obtaining the target feature data corresponding to the user data to be processed based on the dynamic features is achieved through a pre-trained second model.

5. The method according to claim 4, characterized in that, The training process for the first model and the second model includes: Acquire sample data, which carries labeled data and feature labels; The sample data is input into the first model to extract sample features from the sample data. A first objective loss function is calculated based on the sample features and the feature labels. The first objective loss function is used to optimize the first model until the first model completes training. The sample features are input into the second model to predict the sample features and obtain the sample target features of the sample features. A second objective loss function is calculated based on the target features of the sample and the labeled data. The second objective loss function is used to optimize the second model until the second model completes training.

6. A risk management method, characterized in that, The method includes: Obtain the user data to be processed corresponding to the user to be predicted; The method according to any one of claims 1-5 obtains the target feature data corresponding to the user data to be processed; The risk level of the user to be predicted is determined based on the target feature data.

7. The method according to claim 6, characterized in that, Determining the risk level of the user to be predicted based on the target feature data includes: Obtain the pre-set evaluation level; The risk level of the user to be predicted is obtained based on the target feature data and the evaluation level.

8. A risk assessment device, characterized in that, The device includes: The data acquisition module is used to acquire user data to be processed, which includes initial user data at different time points. The graph construction module is used to construct multiple static knowledge graphs from the initial user data at each time point, with one static knowledge graph corresponding to one time point. The feature extraction module is used to extract features from the multiple static knowledge graphs to obtain static features corresponding to the multiple static knowledge graphs. The static features are used to characterize the feature data that integrates the features of each node in the static knowledge graph. The feature processing module is used to concatenate the static features corresponding to each time point to obtain dynamic features; The target feature calculation module is used to obtain target feature data corresponding to the user data to be processed based on the dynamic features; The map construction module is specifically used for: Initial user data at each of the aforementioned time points is extracted to obtain triplet data; Multiple static knowledge graphs are obtained by constructing graphs based on the triplet data; The graph construction module is further configured to: when a node in the static knowledge graph has a missing label, read the triple data of the node with the missing label, and supplement the node with the missing label according to the label of the triple data.

9. A risk management device, characterized in that, The device includes: The pending user data acquisition module is used to acquire pending user data corresponding to the users to be predicted. The risk prediction module is used to obtain target feature data corresponding to the user data to be processed by the device according to claim 8; The risk level assessment module is used to determine the risk level of the user to be predicted based on the target feature data.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images