User credit scoring method and system based on multi-source behavior data

Abstract

The invention provides a user credit scoring method and system based on multi-source behavior data, and the method comprises the steps: obtaining user behavior data, so as to construct a standardized multi-module behavior event record; taking a user as a unit, using the behavior event record as a node to construct a behavior graph, and constructing a directed edge according to a timestamp of event occurrence; based on the behavior map, constructing a user behavior path by using a constrained maximum edge weight path search algorithm; performing structure matching on the high-credit user behavior path template set and each user behavior path, calculating a path deviation degree of each user behavior path, and obtaining a minimum deviation score; and in combination with the minimum deviation score and the matched optimal matching path, predicting a risk tag of the user through a credit scoring model based on dichotomy learning, and outputting a customer credit score.

Description

Technical Field

[0001] The present invention belongs to the field of user credit scoring, and in particular relates to a user credit scoring method and system based on multi-source behavioral data. Background Art

[0002] With the ongoing digital transformation of the financial industry, user interactions within the banking system have undergone fundamental changes. Traditional offline, static credit models are gradually giving way to risk assessment mechanisms based on online operational behavior. Under this trend, a user's creditworthiness is no longer determined solely by their repayment history or asset information, but is increasingly reflected in their operational activities across various banking modules, such as electronic know-your-customer (eKYC), document upload and recognition (OCR), electronic loan applications (eLending), transaction review, and anti-money laundering (AML). Each type of operation generates behavioral data with varying structures, granularity, and semantics, distributed across multiple systems. While this multi-source, heterogeneous behavioral data reflects the real-world user interaction process, current credit scoring systems struggle to fully utilize this data due to the lack of unified modeling standards and behavioral perspectives.

[0003] Existing mainstream credit scoring methods rely on structured feature modeling, typically constructing input vectors for classification by organizing basic user information and credit history, and then employing machine learning models for scoring. While these methods perform reasonably well in traditional credit scenarios, they are largely ineffective for "cold-start" users with no credit history, such as first-time account holders. In recent years, while deep learning models have attempted to directly leverage raw data in certain scenarios, they continue to face challenges in the financial sector, such as poor interpretability, difficulty in regulatory audits, and high training data requirements. Furthermore, they generally lack the ability to handle complex interaction patterns such as behavioral chains, causal sequences, and cross-module pathways. Furthermore, with the increasing adoption of modular deployment and cloud-native architectures in financial IT systems, scoring systems face the challenge of flexible deployment across different platforms, supporting real-time computing, and service-based integration. These factors collectively result in the poor performance of existing credit scoring systems in behavior-driven scenarios, hindering their ability to effectively identify risk factors within complex behavioral chains. This limits the accuracy and efficiency of financial institutions in areas such as new customer acquisition, risk early warning, and intelligent auditing. Summary of the Invention

[0004] The purpose of this invention is to design a user credit scoring method and system based on multi-source behavioral data, which solves the technical deficiencies in the face of multi-source behavioral data, behavioral chain modeling, user cold start processing and system deployment flexibility.

[0005] To achieve the above objectives, the present invention provides a user credit scoring method based on multi-source behavioral data in a first aspect, the method comprising:

[0006] Step 1: Obtain user behavior data to construct a standardized multi-module behavior event record, wherein the standardized multi-module behavior event record is a five-tuple, including the user ID, the standardized behavior event type, the timestamp of the event, an optional behavior description text, and the event source module;

[0007] Step 2: Take the user as the unit, use the behavior event type as the node to build the behavior graph, and build the directed edge according to the timestamp of the event; retain the previous high-scoring paths to form a set of high-credit user behavior path templates;

[0008] Step 3: Based on the behavior graph, a constrained maximum edge weight path search algorithm is used to construct a user behavior path; the set of high-credit user behavior path templates is structurally matched with each of the user behavior paths, the path deviation of each user behavior path is calculated, and the minimum deviation score is obtained; the user behavior path corresponding to the minimum deviation score is the optimal matching path;

[0009] Step 4: Combine the minimum deviation score and the matched optimal matching path, predict the user's risk label through a credit scoring model based on binary classification learning, and output the customer credit score; wherein the training samples of the credit scoring model come from historical user data, and the credit score is generated by posterior credit default results or business score labeling.

[0010] Furthermore, the method further comprises:

[0011] Step 5: Encapsulate the trained credit scoring model as a microservice component, deploy it using Docker, Kubernetes, and the FastAPI framework, provide a standardized HTTP interface, and integrate score caching, call auditing, and dynamic hot update mechanisms.

[0012] Furthermore, the behavior graph is constructed based on users, using behavior event types as nodes, and directed edges are constructed according to the timestamps of the behavior event types, specifically including:

[0013] Obtain the behavior event type of each user; sort the behavior event types of each user in chronological order to form a behavior sequence;

[0014] Based on each user's behavior sequence, a behavior graph is constructed, where each behavior event record is treated as a node. If the timestamp of the event at the second node is greater than the timestamp of the event at the first node, a directed edge is connected, indicating that the first node points to the second node.

[0015] Design the edge weights of the behavior graph , expressed as:

[0016] ;

[0017] in, Indicates the first node The timestamp of the event and the second node The time interval between the timestamps of events; Indicates whether the path module is switched. If the path module is different, it is 1; Indicates whether the process jump is illegal, illegal is 1, legal is 0; Represents the weight coefficients corresponding to the three structural factors, which are tuned by the validation set;

[0018] in, It is a logical function generated by the business preset legal event jump relationship table, which severely punishes illegal jump paths in structure and prevents the formation of misleading high-scoring paths.

[0019] Further, the behavior map is retained before high-scoring paths to form a behavior template set, which is a set of high-credit user behavior path templates, specifically including:

[0020] Path screening is performed based on the edge weights of the behavior graph, and path scores are generated. high-scoring paths constitute a behavior template set; wherein each template path in the template set includes a standard behavior type, a corresponding path module sequence and a path score.

[0021] Furthermore, the path generation goal of constructing the user behavior path based on the behavior graph using the constrained maximum edge weight path search algorithm is:

[0022] ;

[0023] in, is the target path, For the behavior map, is the global path, is the current path; coefficient and Control inhibition weights; 、 For nodes and nodes ; is the current path module sequence, is the next step path module sequence;

[0024] in, is the sum of path edge weights, indicating that paths with close timing, reasonable module flow, and legal jumps are given priority; To punish the repetition of behavior; The penalty path only stays in a single path module or a single process, lacking the real financial process span.

[0025] Furthermore, performing structural matching between the high-credit user behavior path template set and each user behavior path, calculating the path deviation of each user behavior path, and obtaining the minimum deviation score specifically includes:

[0026] Detect whether the behavior event type in the current user behavior path is not in the set of high-credit user behavior path templates. If it does not match, it is counted as 1, and a deviation result is generated to measure the legitimacy deviation at the attribute level;

[0027] Detecting whether the current path module sequence does not conform to the corresponding path template of the high-credit user behavior path template set, indicating that the path template has misplaced actions or skipped steps;

[0028] Mark whether the event node is a node on an illegal structure jump path;

[0029] Combine the deviation results, path template corresponding results and illegal structure jump results to perform matching and generate the path deviation degree of each user behavior path;

[0030] Match all user behavior paths to obtain the minimum deviation score and the corresponding optimal matching path.

[0031] Furthermore, the method combines the minimum deviation score and the optimal matching path to predict the user's risk label through a credit scoring model based on binary classification learning, and outputs the customer's credit score, specifically including:

[0032] The minimum deviation score, each behavioral event type in the optimal matching path, and the risk weight of each behavioral event type are input into a deployable credit scoring model to output a customer credit score; wherein the credit scoring model is a shallow neural network structure.

[0033] Furthermore, the training process of the credit scoring model is as follows:

[0034] Obtain the minimum deviation score and the optimal matching path;

[0035] Construct a training set based on the types of real risk behavior events;

[0036] Performing training based on the training set using a least squares loss function with L2 regularization to optimize model parameters and risk weights for each behavioral event type;

[0037] Use a standard machine learning framework to complete the training process. After the credit scoring model is trained, save the structured credit scoring model.

[0038] In a second aspect of the present invention, a user credit scoring system based on multi-source behavioral data is provided, the system comprising:

[0039] A task initialization module is used to obtain user behavior data to construct a standardized multi-module behavior event record. The standardized multi-module behavior event record is a five-tuple, including the user ID, the standardized behavior event type, the timestamp of the event, an optional behavior description text, and the event source module;

[0040] The scoring graph module is used to construct a behavior graph based on the user, using the behavior event type as the node, and constructing directed edges according to the timestamp of the event; high-scoring paths to form a set of high-credit user behavior path templates;

[0041] A risk path construction module is configured to construct a user behavior path based on the behavior graph using a constrained maximum edge weight path search algorithm; perform structural matching between the set of high-credit user behavior path templates and each of the user behavior paths, calculate the path deviation of each user behavior path, and obtain a minimum deviation score; wherein the user behavior path corresponding to the minimum deviation score is the optimal matching path;

[0042] A credit scoring module is used to combine the minimum deviation score and the matched optimal matching path to predict the user's risk label through a credit scoring model based on binary classification learning, and output the customer credit score; wherein the training samples of the credit scoring model are derived from historical user data, and the credit score is generated by posterior credit default results or business score labeling.

[0043] Furthermore, the system further comprises:

[0044] The model encapsulation module is used to encapsulate the trained credit scoring model into a microservice component, which is deployed using Docker, Kubernetes, and the FastAPI framework. It provides a standardized HTTP interface and integrates scoring caching, call auditing, and dynamic hot update mechanisms.

[0045] The beneficial technical effects of the present invention are at least as follows:

[0046] This invention addresses the technical deficiencies of existing credit scoring systems in terms of multi-source behavioral data, behavioral chain modeling, user cold start processing, and system deployment flexibility, and proposes a novel credit scoring method and system for comprehensive modeling and deployment. This method first introduces a unified behavioral data representation mechanism to normalize and represent user operation records in different business modules, making various behaviors comparable and fusible. Secondly, a behavioral chain model with a cross-module, cross-time, and semantic structure is constructed to depict the user's interaction path in multiple business processes, thereby discovering potential behavioral deviation patterns and operational anomaly characteristics to support more discriminative and predictive credit scoring judgments. Furthermore, this invention also proposes a scoring capability encapsulation solution for system integration scenarios, which supports the deployment of scoring capabilities as service-oriented components in the bank's core system or digital platform, enabling rapid access and invocation of the scoring model by multiple business systems. This system is particularly well-suited to the requirements of data isolation, module decoupling, and microservice operation under the fintech architecture, and possesses excellent scalability and engineering implementation capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0048] Figure 1 This is a flow chart of a user credit scoring method based on multi-source behavioral data in the present invention. DETAILED DESCRIPTION

[0049] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0050] In one or more embodiments, Figure 1 As shown, a user credit scoring method based on multi-source behavioral data is disclosed, and the method includes the following S1 to S5:

[0051] S1. Obtain user behavior data to construct a standardized multi-module behavior event record. The standardized multi-module behavior event record is a five-tuple, including the user ID, the standardized behavior event type, the timestamp of the event, the optional behavior description text, and the event source module.

[0052] Specifically, during the standardization process, text-based results from the electronic identity authentication module, such as "Authentication passed" and "Face verification failed," are converted into standard labels such as "KYC_PASS" and "KYC_FAIL" through a rule dictionary and keyword matching. Numerical data from the AML module, such as transaction amounts, is normalized using a standardization function to ensure numerical comparability across events.

[0053] Furthermore, the following formula is used to standardize the numerical features:

[0054] ;

[0055] in, Represents raw numerical data (such as transaction amount); Indicates the mean value of the field in the training data; Indicates the standard deviation of the field in the training data; Represents the standardized result, which is used to eliminate the impact of differences in the numerical scales of different modules.

[0056] The system finally expresses each processed event record as a five-tuple ,in:

[0057] Indicates the user ID, provided by the system log or module interface; Indicates the standardized behavioral event type (such as "KYC_FAIL", "UPLOAD_SUCCESS"); Indicates the timestamp of the event, which is uniformly converted to Unix timestamp; Optional behavior description text, only for log readability; Indicates the event source module (such as "OCR module", "AML module", etc.).

[0058] Among them, all standardized event records will be stored in a behavioral event set. Each record has a unified structure, which is convenient for rapid conversion into node form during subsequent graph structure construction.

[0059] For example, the original three records are: OCR upload successful; eKYC authentication failed; AML rule 5 triggered.

[0060] After standardization, it is converted into:

[0061] User ID: user123, Action Type: UPLOAD_SUCCESS, Time: 2025-05-12 10:05:01, Source Module: OCR;

[0062] User ID: user123, Action Type: KYC_FAIL, Time: 2025-05-12 10:07:43, Source Module: eKYC;

[0063] User ID: user123, Behavior type: AML_TRIGGERED_RULE_5, Time: 2025-05-12 10:15:22, Source module: AML.

[0064] S2. Take the user as the unit, use the behavior event record as the node to build the behavior graph, and build the directed edge according to the timestamp of the event; retain the previous high-scoring paths to form a set of high-credit user behavior path templates.

[0065] Specifically, this step aims to record the standardized multi-module behavior events output from step 1. A structured behavioral graph is constructed to capture the user's operational chain, behavioral logic, and module transitions within a multi-module financial process. This graph not only preserves the chronological order of behaviors but also incorporates business-specific constraints such as module switching, abnormal jumps, and repeated behaviors. It is then used to extract well-structured behavioral path templates from high-credit users, serving as a reference for subsequent path generation and deviation modeling steps.

[0066] The system uses each user As a unit, sort the behavioral event records in chronological order to form a behavioral sequence. Then, construct a behavioral map , where each is a node, if , then connect the directed edges .

[0067] To enhance the graph structure's ability to truly express business processes, the system defines the following innovative edge weight functions to quantify the continuity and rationality of behaviors between events:

[0068] ;

[0069] in, Indicates behavioral events and time interval; Indicates whether the module is switched, if the module is different, it is 1; Indicates whether the process jump is illegal (1 for illegal, 0 for legal); Represents the weight coefficients corresponding to the three structural factors, which are tuned by the validation set (such as , , );

[0070] in It is a logical function generated by the business's preset legal event jump relationship table. Its design ensures that illegal jump paths (such as skipping KYC and directly lending) are severely punished in structure to prevent the formation of misleading high-scoring paths.

[0071] After the graph is built, from the high-credit user group Extract frequent, stable, and cross-module complete behavioral paths from the graph collection as a reference template collection Path screening uses the following scoring function:

[0072] ;

[0073] in, represents the edge weight defined above, which measures the fluency of the behavior; represents the number of unique modules appearing in the path, indicating the process coverage; Indicates the number of illegal jump events (from the number of edges); Indicates the module coverage bonus coefficient (e.g. ), is the jump penalty coefficient (e.g. ); The higher the path score, the more continuous the behavior, the richer the modules, and the more reasonable the structure. high-scoring paths, forming a behavioral template set .

[0074] For example, the path example is: "UPLOAD_IMAGE" → "KYC_PASS" → "LOAN_APPLY" → "AML_PASS"; "OCR_FAIL" → "REUPLOAD" → "KYC_PASS" → "LOAN_APPLY".

[0075] Each template path contains standard behavior types Sequence, corresponding to module sequence , for subsequent model comparison.

[0076] Further output: behavior graph of each user ; High-credit user behavior path template collection , including scores, module sequences and behavior labels.

[0077] S3. Based on the behavior graph, a constrained maximum edge weight path search algorithm is used to construct a user behavior path; the high-credit user behavior path template set is structurally matched with each of the user behavior paths, the path deviation of each user behavior path is calculated, and a minimum deviation score is obtained; wherein the user behavior path corresponding to the minimum deviation score is the optimal matching path.

[0078] Specifically, this step further generates the user's current behavior path based on the previous behavior graph structure modeling and reference path template extraction, and calculates the degree of deviation from the reference path set. Deviation score, as the core risk feature of this patent, is a key indicator for distinguishing "users within the process" from "users outside the process." Different from traditional risk scores based on field comparison or black box models, this step uses the behavior graph as input, combined with the standard process template structure, through graph path search + structure alignment + scenario regularization, to build a set of interpretable, quantifiable, and business-verifiable behavior deviation system, ultimately forming the ability to identify high-risk behavior chains.

[0079] First, based on the graph , using the constrained maximum edge weight path search algorithm to construct the user behavior path , the goal is to identify its "complete chain of currently executed behaviors". To avoid misidentifying short jump paths or single module loop paths as main paths, structural integrity and repeated behavior penalties are introduced in the search process. The path generation goals are:

[0080] ;

[0081] in:

[0082] The first item is the sum of the path edge weights, which comes from the edge weight definition in step 2. It indicates that the paths with close time, reasonable module flow, and legal jumps are given priority.

[0083] Item 2: Punishing repeated behaviors, such as repeated "loan applications" and "KYC failures";

[0084] Third: The penalty path is limited to a single module (such as OCR) or a single process, lacking the full scope of real financial processes;

[0085] coefficient and Control the suppression weights, which are generally set to 0.7 and 0.9 respectively, and are tuned based on historical training samples.

[0086] The system will then With each Perform structural matching and calculate path deviation Due to the non-standard path fluctuations, step skipping, and reverse order of user behavior, the traditional edit distance cannot meet the business accuracy of financial processes. Therefore, the following module offset matching function with a structural regularization term is designed:

[0087] ;

[0088] in, Indicates the current reference path Bit event in user path The allowed window on the event location, increasing tolerance; Represents a reference event If it does not appear in the current window, it is recorded as deviation; Indicates that the corresponding modules do not match, indicating that the module behavior is misplaced or skipped; Represents a marker event Whether it is a node on the illegal structure jump path (such as AML before KYC) is determined by the predefined flow chart; Control module jump impact strength, generally set to 1.2; This is a strong penalty item for illegal paths, and its value can be 2.0 or even higher.

[0089] Furthermore, this function integrates the flexibility of path matching position, rationality of module matching, and penalty for illegal behavior path, truly simulating whether the user is running under the "standard process" and accurately depicting the degree of deviation. Perform the above matching to get the final minimum deviation score:

[0090] ;

[0091] Among them, the higher the score, the more the user behavior deviates from the standard process, which is manifested as process jumping, structural missing, and illegal cross-module behavior, and is subsequently used as the core feature in the scoring model.

[0092] Output: Minimum deviation score : Quantify the difference between user behavior chain and standard process; optimal matching path : The reference path that is closest to the current behavior, which can be used for auditing, interpretable output, and score correction.

[0093] S4. Combining the minimum deviation score and the matched optimal matching path, predict the user's risk label through a credit scoring model based on binary classification learning, and output the customer credit score; wherein the training samples of the credit scoring model are derived from historical user data, and the credit score is generated by posterior credit default results or business score labeling.

[0094] Specifically, the goal of this step is to build a credit scoring model that can learn the user behavior risk scoring rules, and the input is the user behavior deviation score output in the previous step. and its matching reference path By introducing the structural attention factor and process deviation regularization mechanism, a scoring model with structural expression ability is constructed. In this model, is considered as the extracted “behavioral risk factor”, while The behavior labels in are embedded as structural path labels to assist the behavior attention mechanism in the modeling process. The final output is a trained structured scoring model. , used in the subsequent deployment of the scoring system.

[0095] The modeling task is a binary classification problem, namely, predicting whether a user is a "high-risk user." Training samples are derived from historical user data, and their labels are generated by a posteriori credit default results or business ratings. This paper uses a deployable shallow neural network structure for modeling, with the input layer consisting of the following:

[0096] First dimension: minimum deviation score ;

[0097] Second dimension: reference path Each key behavior label The appearance state is input as a vector , the dimension is , is the set of all behavior tags, Indicates that the user path has appeared , otherwise 0;

[0098] The third dimension: risk weight for each tag , set by the expert system or business knowledge, for example, "KYC_FAIL" is set to 1.5 and "UPLOAD_IMAGE" is set to 0.7.

[0099] Furthermore, the present invention defines the final model output as follows:

[0100] ;

[0101] in: Model parameters representing the deviation factors; Indicates the The risk weight of each reference behavior; Indicates whether the user has performed the action; represents the bias term; Represents an activation function (such as Sigmoid or ReLU).

[0102] The innovation of this formula lies in integrating process structure label information into modeling features in the form of "path attention + risk weighting", and combining deviation as a risk regularizer to achieve the integration of graph structure path and modeling algorithm.

[0103] The model training process is as follows:

[0104] For each historical user, execute steps 1 to 3 to obtain ;

[0105] Construct a training set based on real risk events (such as whether it is overdue);

[0106] Use the least squares loss function with L2 regularization for training and optimize the parameters and risk-weighting parameters (if not fixed);

[0107] Use standard machine learning frameworks (such as PyTorch and TensorFlow) to complete the training process;

[0108] After model training is completed, save the structured scoring model , including parameter weight, structure definition and risk interpretation modules.

[0109] Output: Credit scoring model , containing the trained parameter set; model structure configuration file, which is used to load and use it when the scoring function is subsequently called.

[0110] S5. Encapsulate the trained credit scoring model as a microservice component, deploy it using Docker, Kubernetes, and the FastAPI framework, and provide a standardized HTTP interface. It also integrates scoring caching, call auditing, and dynamic hot update mechanisms.

[0111] Specifically, the goal of this step is to convert the scoring model trained in the previous step Service-oriented deployment enables real-time access to financial business systems in a stable, callable, and auditable manner. The deployment process adheres to the core requirements of FinTech systems for high availability, low latency, and compliance tracking. It uses a microservice container architecture to encapsulate model operation logic, builds a standard scoring interface, and ensures the efficiency and security of system operation through mechanisms such as score caching, interface auditing, and dynamic model hot loading.

[0112] The model deployment uses the Docker+Kubernetes+FastAPI framework to implement the scoring microservice. The core deployment process is as follows:

[0113] Model loading: When the service starts, read The structure definition and weight parameters are loaded by calling a framework (such as PyTorch or ONNX). The loading script registers the interface name, version number, and business label of the model (such as "Anti-Fraud Scoring Model v1.0").

[0114] Interface definition: The service exposes the standard HTTP POST interface / score. The interface request body is:

[0115] { "user_id":"u123456", "input_features":"encodedstring", / / Compressed or encrypted model input vector "request_id":"req20250102001" }

[0116] After receiving the data, the service decodes input_features into the model input tensor and calls the model. Get the rating results .

[0117] Rating response structure:

[0118] { "request_id":"req20250102001", "score":0.837, "risk_level":"MEDIUM", "model_version": "v1.0","timestamp":1700000000 }

[0119] Rating caching mechanism: To reduce the overhead of repeated ratings in high-concurrency scenarios, the service has a built-in rating cache module. The cache key is composed of the user ID and the request hash value. After a cache hit, the historical rating is directly returned. The rating response time is:

[0120] ;

[0121] in: Cache retrieval time, typically <1ms; Model inference time (e.g., 5ms~20ms); For cache hit rate, system target setting .

[0122] Call audit mechanism: Every interface call will be recorded in the "Score Audit Log Table". The log content includes:

[0123] Request timestamp;

[0124] User ID (desensitized and encrypted);

[0125] Request ID;

[0126] Response score and risk level;

[0127] Interface time-consuming;

[0128] The model version used. Audit logs are regularly uploaded to the bank's risk control audit platform or ELK log stack to support post-event traceability and compliance review.

[0129] Model hot update mechanism: Model microservices support zero-downtime hot loading of new model versions The new model launch process is as follows:

[0130] Mount the new model structure and weights in the background;

[0131] First, perform a self-test of the model to check the scoring range and response time;

[0132] If passed, the request is routed from Switch to , the original model remains in cold standby and supports rollback;

[0133] Switch version records to write to the version change log.

[0134] Output: deployable scoring microservice container image and image address (e.g. registry.ai / credit-scorer:v1.0); scoring service interface description document (OpenAPI / Swagger format); model call audit log table design; service deployment performance evaluation results (e.g. average response time); , cache hit rate and other indicators).

[0135] In one or more embodiments, a user credit scoring system based on multi-source behavioral data is disclosed, the system comprising:

[0136] A task initialization module is used to obtain user behavior data to construct a standardized multi-module behavior event record. The standardized multi-module behavior event record is a five-tuple, including the user ID, the standardized behavior event type, the timestamp of the event, an optional behavior description text, and the event source module;

[0137] The scoring graph module is used to construct a behavior graph based on users, using behavior event records as nodes, and construct directed edges according to the timestamps of events; high-scoring paths to form a set of high-credit user behavior path templates;

[0138] A risk path construction module is configured to construct a user behavior path based on the behavior graph using a constrained maximum edge weight path search algorithm; perform structural matching between the set of high-credit user behavior path templates and each of the user behavior paths, calculate the path deviation of each user behavior path, and obtain a minimum deviation score; wherein the user behavior path corresponding to the minimum deviation score is the optimal matching path;

[0139] A credit scoring module is used to combine the minimum deviation score and the matched optimal matching path to predict the user's risk label through a credit scoring model based on binary classification learning, and output the customer credit score; wherein the training samples of the credit scoring model are derived from historical user data, and the credit score is generated by posterior credit default results or business score labeling.

[0140] Furthermore, the system further comprises:

[0141] The model encapsulation module is used to encapsulate the trained credit scoring model into a microservice component, which is deployed using Docker, Kubernetes, and the FastAPI framework. It provides a standardized HTTP interface and integrates scoring caching, call auditing, and dynamic hot update mechanisms.

[0142] It is worth noting that the specific workflow of the user credit scoring system based on multi-source behavioral data provided by the embodiment of the present invention is the same as the workflow of the user credit scoring method based on multi-source behavioral data described in the above embodiment, and will not be repeated here.

[0143] An embodiment of the present invention further provides a user credit scoring device based on multi-source behavioral data, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the above-mentioned embodiment of the user credit scoring method based on multi-source behavioral data are implemented, for example: Figure 1 or, the processor implements the functions of the modules in the above-mentioned system embodiments when executing the computer program.

[0144] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the user credit scoring device based on multi-source behavioral data.

[0145] The user credit scoring device based on multi-source behavioral data can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The user credit scoring device based on multi-source behavioral data can include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the user credit scoring device based on multi-source behavioral data can also include input / output devices, network access devices, buses, and the like.

[0146] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor serves as the control center of the user credit scoring device based on multi-source behavioral data, connecting the various components of the device using various interfaces and circuits.

[0147] The memory can be used to store the computer programs and / or modules. The processor implements the various functions of the user credit scoring device based on multi-source behavioral data by running or executing the computer programs and / or modules stored in the memory and accessing the data stored in the memory. The memory may primarily include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function, and the data storage area may store data generated based on the operation of the air conditioning controller. Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a Smart Memory Card (SMC), a Secure Digital (SD) card, a Flash Card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0148] If the module integrated into the user credit scoring device based on multi-source behavioral data is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can also implement all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard drive, a magnetic disk, an optical disk, computer memory, read-only memory (ROM), random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium.

[0149] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0150] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A user credit scoring method based on multi-source behavioral data, characterized in that: The method comprises: Step 1: Obtain user behavior data to construct a standardized multi-module behavior event record, wherein the standardized multi-module behavior event record is a five-tuple, including the user ID, the standardized behavior event type, the timestamp of the event, an optional behavior description text, and the event source module; Step 2: Take the user as the unit, use the behavior event record as the node to build the behavior graph, and build the directed edge according to the timestamp of the event; retain the previous high-scoring paths to form a set of high-credit user behavior path templates; Step 3: Based on the behavior graph, a constrained maximum edge weight path search algorithm is used to construct a user behavior path; the set of high-credit user behavior path templates is structurally matched with each of the user behavior paths, the path deviation of each user behavior path is calculated, and the minimum deviation score is obtained; the user behavior path corresponding to the minimum deviation score is the optimal matching path; Step 4: Combining the minimum deviation score and the matched optimal matching path, predict the user's behavioral event type through a credit scoring model based on binary classification learning, and output the customer credit score; wherein the training samples of the credit scoring model are derived from historical user data, and the behavioral event types of the training samples are generated by posterior credit default results or business score labeling.

2. A user credit scoring method based on multi-source behavioral data according to claim 1, characterized in that: The method further comprises: Step 5: Encapsulate the trained credit scoring model as a microservice component, deploy it using Docker, Kubernetes, and the FastAPI framework, provide a standardized HTTP interface, and integrate score caching, call auditing, and dynamic hot update mechanisms.

3. The user credit scoring method based on multi-source behavioral data according to claim 1, characterized in that: The behavior graph is constructed based on users, using behavior event types as nodes, and directed edges are constructed according to the timestamps of the events. Specifically, it includes: Obtain the behavior event type of each user; sort the behavior event types of each user in chronological order to form a behavior sequence; Based on each user's behavior sequence, a behavior graph is constructed, where each behavior event record is treated as a node. If the timestamp of the event at the second node is greater than the timestamp of the event at the first node, a directed edge is connected, indicating that the first node points to the second node. Design the edge weights of the behavior graph , expressed as: ； in, Indicates the first node The timestamp of the event and the second node The time interval between the timestamps of events; Indicates whether the path module is switched. If the path module is different, it is 1; Indicates whether the process jump is illegal, illegal is 1, legal is 0; Represents the weight coefficients corresponding to the three structural factors, which are tuned by the validation set; in, It is a logical function generated by the business preset legal event jump relationship table, which severely punishes illegal jump paths in structure and prevents the formation of misleading high-scoring paths.

4. A user credit scoring method based on multi-source behavioral data according to claim 3, characterized in that: The behavior map is retained before high-scoring paths to form a behavior template set, which is a set of high-credit user behavior path templates, specifically including: Path screening is performed based on the edge weights of the behavior graph, and path scores are generated. high-scoring paths constitute a behavior template set; wherein each template path in the template set includes a standard behavior type, a corresponding path module sequence and a path score.

5. The user credit scoring method based on multi-source behavioral data according to claim 1, characterized in that: The path generation goal of constructing the user behavior path based on the behavior graph using the constrained maximum edge weight path search algorithm is: ； in, is the target path, For the behavior map, is the global path, is the current path; coefficient and Control inhibition weights; 、 For nodes and nodes ; is the current path module sequence, is the next step path module sequence; in, is the sum of path edge weights, indicating that paths with close timing, reasonable module flow, and legal jumps are given priority; To punish the repetition of behavior; The penalty path only stays in a single path module or a single process, lacking the real financial process span.

6. A user credit scoring method based on multi-source behavioral data according to claim 5, characterized in that: The structural matching of the high-credit user behavior path template set with each of the user behavior paths, calculating the path deviation of each user behavior path, and obtaining the minimum deviation score specifically includes: Detect whether the behavior event type in the current user behavior path is not in the set of high-credit user behavior path templates. If it does not match, it is counted as 1, and a deviation result is generated to measure the legitimacy deviation at the attribute level; Detecting whether the current path module sequence does not conform to the corresponding path template of the high-credit user behavior path template set, indicating that the path template has misplaced actions or skipped steps; Mark whether the event node is a node on an illegal structure jump path; Combine the deviation results, path template corresponding results and illegal structure jump results to perform matching and generate the path deviation degree of each user behavior path; Match all user behavior paths to obtain the minimum deviation score and the corresponding optimal matching path.

7. The user credit scoring method based on multi-source behavioral data according to claim 1, characterized in that: The method combines the minimum deviation score and the optimal matching path, predicts the user's risk label through a credit scoring model based on binary classification learning, and outputs the customer's credit score, specifically including: The minimum deviation score, each behavioral event type in the optimal matching path, and the risk weight of each behavioral event type are input into a deployable credit scoring model to output a customer credit score; wherein the credit scoring model is a shallow neural network structure.

8. The user credit scoring method based on multi-source behavioral data according to claim 7, characterized in that: The training process of the credit scoring model is as follows: Obtain the minimum deviation score and the optimal matching path; Construct a training set based on the types of real risk behavior events; Performing training based on the training set using a least squares loss function with L2 regularization to optimize model parameters and risk weights for each behavioral event type; Use a standard machine learning framework to complete the training process. After the credit scoring model is trained, save the structured credit scoring model.

9. A user credit scoring system based on multi-source behavioral data, characterized in that: The system comprises: A task initialization module is used to obtain user behavior data to construct a standardized multi-module behavior event record. The standardized multi-module behavior event record is a five-tuple, including the user ID, the standardized behavior event type, the timestamp of the event, an optional behavior description text, and the event source module; The scoring graph module is used to construct a behavior graph based on the user, using the behavior event type as the node, and constructing directed edges according to the timestamp of the event; high-scoring paths to form a set of high-credit user behavior path templates; A risk path construction module is configured to construct a user behavior path based on the behavior graph using a constrained maximum edge weight path search algorithm; perform structural matching between the set of high-credit user behavior path templates and each of the user behavior paths, calculate the path deviation of each user behavior path, and obtain a minimum deviation score; wherein the user behavior path corresponding to the minimum deviation score is the optimal matching path; A credit scoring module is used to combine the minimum deviation score and the matched optimal matching path to predict the user's risk label through a credit scoring model based on binary classification learning, and output the customer credit score; wherein the training samples of the credit scoring model are derived from historical user data, and the credit score is generated by posterior credit default results or business score labeling.

10. A user credit scoring system based on multi-source behavioral data according to claim 9, characterized in that: The system further comprises: The model encapsulation module is used to encapsulate the trained credit scoring model into a microservice component, which is deployed using Docker, Kubernetes, and the FastAPI framework. It provides a standardized HTTP interface and integrates scoring caching, call auditing, and dynamic hot update mechanisms.

Images