Financial risk monitoring method and system, electronic equipment and medium

Abstract

The invention provides a financial risk monitoring method and system, electronic equipment and a medium, and the method comprises the steps: collecting financial data of a target customer from a plurality of different data sources, fusing the financial data of the different data sources, and obtaining a fused financial data set; constructing a financial risk knowledge graph based on the fused financial data set; calculating a real-time risk score of the target customer based on the financial risk knowledge graph and a pre-constructed risk scoring model; and generating risk early warning information based on the real-time risk score and a pre-constructed risk early warning rule. Financial risk monitoring can be efficiently and accurately carried out, and the risk prevention and control capability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk monitoring, and in particular to a financial risk monitoring method, system, electronic equipment and medium. Background Art

[0002] Currently, existing financial risk monitoring systems have numerous shortcomings in data integration, risk identification, real-time performance, visualization, functional coverage, scalability, security, and user experience. These deficiencies limit their effectiveness in financial risk monitoring and make it difficult to meet the increasingly complex needs of financial risk prevention and control. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a financial risk monitoring method, system, electronic equipment and medium, which can monitor financial risks efficiently and accurately and improve risk prevention and control capabilities.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present invention provides a financial risk monitoring method, comprising: collecting financial data of target customers from multiple different data sources, and fusing the financial data from different data sources to obtain a fused financial data set; constructing a financial risk knowledge graph based on the fused financial data set; calculating the real-time risk score of the target customer based on the financial risk knowledge graph and a pre-constructed risk scoring model; and generating risk warning information based on the real-time risk score and pre-constructed risk warning rules.

[0006] Optionally, a financial risk knowledge graph is constructed based on the fused financial dataset, including: extracting entities from the fused financial dataset based on the BERT pre-trained model and the BiLSTM+CRF algorithm model to obtain an entity list; extracting relationships between entities based on the entity list and the fused financial dataset, and performing type labeling and weight calculation on the relationships to obtain a relationship list; extracting attributes of the entities based on the entity list and the fused financial dataset to obtain an attribute list; and constructing a financial risk knowledge graph based on the entity list, relationship list, and attribute list.

[0007] Optionally, a financial risk knowledge graph is constructed based on the entity list and the relationship list, including: constructing an initial financial risk knowledge graph based on the entity list and the relationship list; and using a MuGNN algorithm to perform entity alignment on the initial financial risk knowledge graph to obtain a financial risk knowledge graph.

[0008] Optionally, the real-time risk score of the target customer is calculated based on the financial risk knowledge graph and a pre-built risk scoring model, including: calculating the risk indicator data of the target customer based on the financial risk knowledge graph using a graph computing algorithm; calculating the real-time risk score of the target customer based on the risk indicator data of the target customer and a pre-built risk scoring model.

[0009] Optionally, after generating risk warning information based on real-time risk scores and pre-built risk warning rules, it also includes: classifying the risk warning information and prioritizing the risk warning information according to pre-set risk levels; generating risk disposal suggestions based on the risk warning information and the financial risk knowledge graph, and pushing the risk disposal suggestions to relevant personnel.

[0010] Optionally, after generating risk warning information based on real-time risk scores and pre-built risk warning rules, it also includes: drawing a risk network diagram based on the financial risk knowledge graph and risk indicator data, and visually displaying the risk network diagram; generating a multidimensional risk dashboard based on risk indicator data and risk warning information, and visually displaying the multidimensional risk dashboard; wherein the multidimensional risk dashboard is used to display real-time risk scores, risk warning information and risk transmission paths.

[0011] Optionally, it also includes: regularly acquiring new financial data and user feedback data; updating entities and relationships in the financial risk knowledge graph based on new financial data; training risk scoring models based on new financial data, and optimizing risk warning rules and risk disposal recommendations based on user feedback data.

[0012] In a second aspect, the present invention provides a financial risk monitoring system, comprising: a data acquisition module for collecting financial data of target customers from multiple different data sources, and fusing the financial data from different data sources to obtain a fused financial data set; a knowledge graph construction module for constructing a financial risk knowledge graph based on the fused financial data set; a risk assessment module for calculating the real-time risk score of the target customer based on the financial risk knowledge graph and a pre-built risk scoring model; and a risk warning module for generating risk warning information based on the real-time risk score and pre-built risk warning rules.

[0013] In a third aspect, the present invention provides an electronic device comprising a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the steps of any one of the methods provided in the first aspect above.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program executes the steps of any one of the methods provided in the first aspect.

[0015] The present invention brings the following beneficial effects:

[0016] The above-mentioned financial risk monitoring method, system, electronic device and medium of the present invention first collect the financial data of the target customer from multiple different data sources, and fuse the financial data from different data sources to obtain a fused financial data set; then construct a financial risk knowledge graph based on the fused financial data set; then calculate the real-time risk score of the target customer based on the financial risk knowledge graph and a pre-built risk scoring model; finally, generate risk warning information based on the real-time risk score and the pre-built risk warning rules. In the above-mentioned method, it is possible to integrate financial data from multiple different data sources, construct a financial risk knowledge graph, and calculate the real-time risk score in combination with the pre-built risk scoring model. Finally, risk warning information is generated based on the real-time risk score and the pre-built risk warning rules, thereby enabling efficient and accurate financial risk monitoring, improving risk prevention and control capabilities, and reducing the probability of risky business.

[0017] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the preferred embodiments are specifically listed below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 A flowchart of a financial risk monitoring method provided by an embodiment of the present invention;

[0021] Figure 2 A flowchart of another financial risk monitoring method provided by an embodiment of the present invention;

[0022] Figure 3 A schematic diagram of the structure of a financial risk monitoring system provided by an embodiment of the present invention;

[0023] Figure 4 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] Currently, existing financial risk monitoring systems have many deficiencies in data integration, risk identification, real-time performance, visualization, functional coverage, scalability, security, and user experience:

[0026] (1) Insufficient data integration capabilities: Existing methods are inefficient in data cleaning, fusion, and standardization when processing multi-source heterogeneous data, making it difficult to efficiently integrate financial data.

[0027] (2) Limited accuracy of risk identification: Existing methods have difficulty in capturing complex risk correlations and lack the ability to analyze unstructured data, resulting in low accuracy and comprehensiveness of risk identification.

[0028] (3) Poor real-time performance: The existing system has low computational efficiency when processing large-scale data, making it difficult to achieve real-time risk monitoring and early warning.

[0029] (4) Limited visualization effects: The risk visualization display of the existing system is relatively simple and lacks interactivity and dynamic updating capabilities.

[0030] (5) Incomplete functional coverage: The existing system has a single functional design and lacks coverage of the entire process of financial risk monitoring.

[0031] (6) Insufficient scalability and flexibility: The existing system architecture is closed and difficult to adapt to the rapid changes in financial services and the introduction of new technologies.

[0032] (7) Insufficient security and compliance: The existing system has defects in data security and compliance, which makes it difficult to meet the high requirements of the financial industry.

[0033] (8) Poor user experience: The existing system is complex to operate, has high user learning costs, and lacks personalized functions.

[0034] Based on this, the embodiments of the present invention provide a financial risk monitoring method, system, electronic device and medium, which can monitor financial risks efficiently and accurately and improve risk prevention and control capabilities.

[0035] To facilitate understanding of this embodiment, a financial risk monitoring method disclosed in an embodiment of the present invention is first introduced in detail. The method can be executed by an electronic device such as a smart phone, a computer, a tablet computer, etc. Figure 1 The flowchart of a financial risk monitoring method shown in FIG. 1 illustrates that the method mainly includes the following steps S101 to S104:

[0036] Step S101: collecting financial data of target customers from multiple different data sources, and fusing the financial data from different data sources to obtain a fused financial data set.

[0037] In one implementation, financial data is collected from multiple different data sources, i.e., multi-source heterogeneous data, including but not limited to transaction data, customer data, public opinion data, and industrial and commercial data. The data is then pre-processed, such as by cleaning, conversion, and fusion, to construct a unified financial data set. This specifically involves steps 1 to 3:

[0038] Step 1: Multi-source heterogeneous data collection.

[0039] During the specific implementation, multi-source heterogeneous data (such as transaction data, customer data, public opinion data, industrial and commercial data, etc.) are collected from different data sources through API interfaces, database connections, file imports, etc., and the collected data are preliminarily classified and stored to obtain the original data set.

[0040] Specifically, hierarchical data storage supports direct connection to index databases such as Oracle, Hive, and ElasticSearch, allowing for mixed storage of big data. Parameters such as extraction time range, field selection, and data type definition can be configured. It also supports connecting to local data sources and is compatible with various data formats such as CSV and Excel. It also supports data source upload, deletion, and preview functions. It uses the Spark distributed computing framework for big data computing.

[0041] Step 2: Data cleaning.

[0042] In specific implementation, the financial data in the original data set is cleaned to obtain a cleaned data set. Data cleaning includes but is not limited to: removing duplicate data, missing data and noise data; standardizing the data format (such as date format, currency unit, etc.); and performing pre-processing on text data such as word segmentation and stop word removal.

[0043] Specifically, in the data preprocessing process, stream processing technology and real-time data integration tools are used to complete real-time data integration. Machine learning and natural language processing technologies are then used to automatically identify, clean, and transform data. Data is captured from various data sources in the data warehouse and processed and transformed using data cleansing techniques to conform to the format and requirements of the knowledge graph. Furthermore, through code-free and low-code data access, visual interfaces and graphical tools are used to simplify the data access process. This approach reduces employees' reliance on technical expertise and enables more people to participate in data preprocessing.

[0044] Step 3: Data fusion.

[0045] During the specific implementation, the data from different data sources in the cleaned data set are associated and integrated to build a unified data model, ensure data consistency, and obtain a fused financial data set.

[0046] Specifically, first, data from different data sources is mapped and converted, including: determining the correspondence between fields in different data sources to ensure that the data can match after integration; according to business needs, the data is converted into a unified format, such as standardizing dates, currency units, etc.

[0047] Then, data from different data sources is integrated and loaded, including:

[0048] ETL (Extract, Transform, Load): Extract data from the data source, transform it, and then load it into the target system (such as a data warehouse).

[0049] API integration: Data sharing and integration between systems is achieved through application programming interfaces.

[0050] Data warehouse: stores data in a centralized location and provides a unified view.

[0051] Loading: Loading the consolidated data into the target system to ensure data accessibility and availability.

[0052] Finally, perform data modeling, including but not limited to:

[0053] Paradigm modeling: With normalization as the core, it reduces data redundancy and improves data consistency.

[0054] Dimensional modeling: Built through star, snowflake, or constellation models, suitable for analytical needs in data warehouses.

[0055] Layered framework: Divide data into different layers (such as ODS layer, DWD layer, DWM layer, etc.) for easy management and use.

[0056] Step S102: Construct a financial risk knowledge graph based on the fused financial data set.

[0057] In one embodiment, a financial risk knowledge graph is constructed based on financial domain knowledge, including steps such as entity recognition, relationship extraction, and attribute extraction, to convert financial data into nodes and edges in the knowledge graph to form a financial risk knowledge graph.

[0058] Step S103: Calculate the real-time risk score of the target customer based on the financial risk knowledge graph and the pre-built risk scoring model.

[0059] In one embodiment, a real-time risk score is calculated based on a financial risk knowledge graph using graph computing, machine learning and other technologies.

[0060] Step S104: Generate risk warning information based on the real-time risk score and pre-built risk warning rules.

[0061] In one embodiment, based on real-time risk scoring and pre-set risk warning rules, potential risks are monitored and warned in real time to generate risk warning information. Specifically, the following steps 1 to 2 are included:

[0062] Step 1: Set up risk warning rules.

[0063] During specific implementation, risk warning rules (such as risk score thresholds, risk transmission paths, etc.) are set according to risk scores and historical risk data according to business needs, and the warning rules are stored in the rule engine.

[0064] Step 2: Real-time risk monitoring.

[0065] In specific implementation, risk scores are monitored in real time, and risk warning information is generated when risk warning rules are triggered. For example, when the real-time risk score exceeds the risk score threshold, a risk warning information is generated.

[0066] The above-mentioned financial risk monitoring method of the present invention can integrate financial data from multiple different data sources, construct a financial risk knowledge graph, and calculate real-time risk scores in combination with a pre-built risk scoring model. Finally, risk warning information is generated based on the real-time risk scores and pre-built risk warning rules, thereby enabling efficient and accurate financial risk monitoring, improving risk prevention and control capabilities, and reducing the probability of risky business occurrence.

[0067] In knowledge graph theory, data is divided into three categories: entities, relationships, and attributes. Entities include customers, accounts, institutions, collateral, and so on. Relationships refer to clearly identifiable types of relationships between entities. For example, the relationship between customers and institutions is one of affiliation; the relationship between accounts is one of payment, collection, or fund transfer; and the relationship between customers is one of spouse, legal representative, or parent company. Attributes refer to the relevant characteristics of an entity or relationship. The number and content of these characteristics vary between entities or relationships. For example, account attributes include status, balance, and account opening date; payment attributes include transaction time and amount; and spousal attributes only include establishment time. After data arrives at the system, it is first standardized and stored in the standard layer. The data ETL process breaks down two-dimensional data tables into three elements: entities, relationships, and attributes, according to pre-defined rules. These elements are then combined and stored in a graph database format.

[0068] In one embodiment, for the aforementioned step S102, i.e., when constructing a financial risk knowledge graph based on the fused financial data set, the following methods may be used, including but not limited to:

[0069] First, entities are extracted from the fused financial dataset based on the BERT pre-trained model and the BiLSTM+CRF algorithm model to obtain an entity list.

[0070] In the specific implementation, when performing entity extraction, the following steps are included from step 1 to step 6:

[0071] Step 1: Data preparation.

[0072] Specifically, we first collect the financial data for entity extraction, including text and corresponding entity labels. We then preprocess the data, including tokenization, removing useless characters, and formatting. Finally, we divide the data into training, validation, and test sets.

[0073] Step 2: BERT pre-training.

[0074] Specifically, select a BERT pre-trained model suitable for the task, such as BERT-base, BERT-large, etc. Then use large-scale unlabeled text data to pre-train the BERT model to learn deep representations of language.

[0075] Step 3: Model construction.

[0076] Specifically, we load the pre-trained BERT model as the underlying representation layer of the entity extraction model, then add a BiLSTM layer on the basis of the BERT model to capture the long-distance dependencies between entities; finally, we add a CRF layer on the basis of the BiLSTM layer to consider the constraint relationship between entity labels and improve the accuracy of entity extraction.

[0077] Step 4: Model training.

[0078] Specifically, first select an appropriate loss function, such as the cross-entropy loss function; then select an appropriate optimizer, such as AdamW; then use the training set data to train the model, and update the model parameters through backpropagation and gradient descent algorithms; finally, use the validation set data to verify the model, and adjust the model parameters and hyperparameters based on the verification results.

[0079] Step 5: Model evaluation.

[0080] Specifically, the model is tested and evaluated using the test set data to obtain the final performance of the model, which can be evaluated through evaluation indicators such as precision, recall, F1 value, etc.

[0081] Step 6: Entity extraction.

[0082] Specifically, the text of the entity to be extracted (i.e., the text in the fused financial dataset) is input into the trained model. The model outputs the entity label prediction results for each position in the text. Based on the prediction results, the entities in the text are extracted to obtain an entity list.

[0083] In an embodiment of the present invention, a BERT pre-trained model is used to generate sentence-level text representations. The BERT pre-trained model is also capable of understanding the semantic relationships between multiple words. In addition, BERT can be fine-tuned to adapt to different tasks and datasets, thereby improving the performance and generalization capabilities of the model, giving word feature vectors flexibility, and reducing system overhead. BiLSTM can simultaneously consider forward and backward contextual information to better capture dependencies in sequences. In natural language information such as financial reports, annual reports, and financial news, contextual information is very important for understanding the meaning of the information. Therefore, bidirectional consideration can improve the accuracy and generalization capabilities of the model.

[0084] In addition, in the embodiment of the present invention, natural language processing (NLP) technology can also be used to identify entities (such as names of people, company names, places, etc.) in text data, and to extract and annotate entities in structured data.

[0085] Then, based on the entity list and the fused financial dataset, the relationships between entities are extracted, and the relationships are type-labeled and weighted to obtain a relationship list.

[0086] In specific implementations, the present invention can use OpenAI's open-source GPT2 large language model for pre-training and fine-tuning in the relation extraction part. The GPT model can complete relation extraction through pre-training and fine-tuning. In the pre-training stage, the GPT model has extensive language knowledge and semantic understanding capabilities through large-scale text data training. These knowledge and capabilities can be fully applied in relation extraction tasks. In the fine-tuning stage, the GPT model is fine-tuned and optimized according to the requirements of different relation extraction tasks, thereby achieving more accurate relation extraction.

[0087] Specifically, in this embodiment of the present invention, a GPT model can be pre-trained using large-scale corpus text to empower the model with semantic understanding, text generation, and structure generation capabilities. Then, a dataset from the financial sector is used to fine-tune the pre-trained model based on the needs of the employee empowerment platform. Finally, the fine-tuned model is used to extract relationships between entities (such as transaction relationships and equity relationships) from the integrated financial dataset. These relationships are then labeled and weighted to produce a relationship list.

[0088] Next, based on the entity list and the fused financial dataset, the attributes of the entity are extracted to obtain an attribute list.

[0089] In the specific implementation, when performing attribute extraction, the following steps are included from step 1 to step 7:

[0090] Step 1: Data preparation.

[0091] Specifically, we first collect data with labeled entities, which can be structured or semi-structured. Then, for each identified entity, we annotate its corresponding attributes, such as entity type, description, and relationship. We then preprocess the text data, including tokenization, removing useless characters, and formatting. Finally, we divide the data into training, validation, and test sets.

[0092] Step 2: BERT pre-training.

[0093] Specifically, select a BERT pre-trained model suitable for the task, such as BERT-base, BERT-large, etc. Then use large-scale unlabeled text data to pre-train the BERT model.

[0094] Step 3: Model construction.

[0095] Specifically, we load the pre-trained BERT model, then add a BiLSTM layer on top of the BERT model to capture the long-distance dependencies between entity attributes. We then add a CRF layer on top of the BiLSTM layer to consider the constraints between attribute labels. Finally, we add a classification layer to predict the attribute labels of the entities.

[0096] Step 4: Model training.

[0097] Specifically, first select a loss function, such as the cross-entropy loss function; then select an optimizer, such as AdamW; then use the training set data to train the model, and update the model parameters through backpropagation and gradient descent algorithms; finally, use the validation set data to verify the model, and adjust the model parameters and hyperparameters based on the verification results.

[0098] Step 5: Model evaluation.

[0099] Specifically, the model is tested and evaluated using the test set data to obtain the final performance of the model.

[0100] Step 6: Attribute extraction.

[0101] Specifically, for the entities in the entity list, the trained attribute extraction model is used to predict their attributes, and the attributes of the entities are extracted based on the prediction results.

[0102] Step 7: Integrate the extracted attributes with the corresponding entities to form a complete entity-attribute pair.

[0103] Finally, a financial risk knowledge graph is constructed based on the entity list, relationship list, and attribute list.

[0104] In the specific implementation, the entity list, relationship list and attribute list are stored in a graph database (such as Neo4j, TigerGraph, etc.) to build a financial risk knowledge graph.

[0105] In one embodiment, after obtaining the financial risk knowledge graph data through entity extraction, relationship extraction, and attribute extraction, the data needs to be further completed and filtered. Based on this, when constructing a financial risk knowledge graph based on an entity list and a relationship list, the following methods can be used, including but not limited to: first, construct an initial financial risk knowledge graph based on the entity list and the relationship list; then, use the MuGNN algorithm to perform entity alignment on the initial financial risk knowledge graph to obtain a financial risk knowledge graph.

[0106] In practice, MuGNN is a multi-granularity graph neural network model for extracting entity relationships from knowledge graphs. Its main feature is the use of a multi-granularity graph representation learning algorithm, which represents entities and relationships in the knowledge graph as a multi-level graph structure. Each level of the graph structure corresponds to a different granularity, capturing semantic information and relationships at different levels. The input information of the MuGNN model constitutes the knowledge graph after entity and relationship extraction, which contains information about entities and relationships. The MuGNN model represents the knowledge graph as a multi-level graph structure, with each level of the graph structure corresponding to a different granularity. At each granularity, the MuGNN model uses MG-GCN for feature extraction and representation learning. At the same time, the MuGNN model uses an adaptive attention mechanism to fuse features from different granularities.

[0107] Specifically, the entity alignment of the initial financial risk knowledge graph using the MuGNN algorithm includes the following process:

[0108] (1) Data preparation.

[0109] Specifically, the attributes and relationships of entities are extracted from the initial financial risk knowledge graph as the input of MuGNN.

[0110] (2) Construct a multimodal graph.

[0111] Specifically, a node is first created for each entity. Node features can include text descriptions, category labels, attribute values, and more. Edges are then created based on relationships in the knowledge graph. Edge features can represent the type and strength of the relationship. Finally, different types of information (such as text and numerical values) are fused into the graph, providing multimodal features for each node and edge.

[0112] (3) MuGNN model training.

[0113] Specifically, the MuGNN model consists of multiple graph convolutional layers, attention mechanisms, and fusion layers. A loss function, including alignment loss and regularization terms, is designed based on the entity alignment task. The MuGNN model is then trained using the prepared data, learning node representations by iteratively optimizing the loss function.

[0114] In this embodiment of the present invention, the MuGNN model can be trained using a multi-task learning strategy to improve the model's generalization and effectiveness. This strategy can share model parameters across multiple tasks, allowing the model to handle multiple tasks simultaneously. Compared to traditional single-granularity graph neural network models, the MuGNN model performs well in the knowledge graph entity alignment task.

[0115] (4) Entity alignment.

[0116] Specifically, the MuGNN model is used to learn the representation vector of each entity, and the similarity between the representation vectors of entities from different sources is calculated. Metrics such as cosine similarity and Euclidean distance can be used, and based on the similarity threshold or sorting results, it is determined which entities are aligned, that is, they represent the same real-world objects.

[0117] Furthermore, in an embodiment of the present invention, a financial risk knowledge graph can be constructed through a knowledge graph platform. After the financial risk knowledge graph is constructed, it is necessary to be able to quickly query the association relationship and visualize the association relationship. Therefore, in an embodiment of the present invention, a graph database such as Neo4J, ArangoDB or OrientDB is built into the knowledge graph platform to meet the query and display requirements. Among them, the present invention adopts a new open source graph database Arangodb, which supports flexible data models, such as document (Document), graph (Graph) and key-value pair (Key-Value) storage. At the same time, ArangoDB is also a high-performance database that uses SQL-like queries or JavaScript extensions to build high-performance applications. Based on Arangodb, the knowledge graph platform can provide users with an interface-friendly visual interface that can intuitively express the relationship between various entities.

[0118] Specifically, the knowledge graph platform provided by this invention offers the following capabilities: data mining and processing using various methods, including SQL, Python notebooks with Pyspark, and data processing components; interactive data source synchronization; and interactive workflow configuration, which can string SQL scripts, Python scripts, and machine learning models together into workflows for scheduled execution. All knowledge graph construction processes will be developed using the functionality provided by the knowledge graph platform, supporting batch processing of historical data to build networks, scheduled batch execution of newly added data, and synchronizing the completed knowledge graph data to the graph database.

[0119] Specifically, graph computing is a crucial step in generating knowledge graphs. Risk group extraction primarily involves unsupervised and semi-supervised learning within relational networks. For example, graph segmentation and graph clustering fall under unsupervised learning algorithms, while risk propagation falls under semi-supervised learning. The system of the present invention provides a rich set of knowledge graph algorithm components, such as PageRank, Louvain, LPA, connected subgraphs, strongly connected subgraphs, in-degree statistical analysis, and node2vec. These components can be used to construct model graphs for training / prediction using a drag-and-drop approach. Furthermore, the platform's provided Python notebook allows access to a wider range of Python-based graph algorithm libraries, enabling more flexible processing and analysis of graph data. Subsequently, interactive workflow configuration allows Python scripts and machine learning models to be linked together into workflows for scheduled execution. Development using the functionality provided by the knowledge graph platform allows for batch community discovery algorithm calculations on historical data and scheduled batch runs on newly added data to discover new risk groups and synchronize group risk labels to a graph database for query, retrieval, and visualization. By integrating graph computing technology with the graph database, the constructed knowledge graph can be visualized.

[0120] Specifically, when building a knowledge graph, various graph algorithms must be employed based on the business scenario and project objectives. The platform must also develop relevant knowledge graph algorithms. Numerous analytical methods and graph algorithms are available for use in identifying anomaly risks within knowledge graphs, such as community discovery algorithms (Louvain, LPA clustering, Clique) and node importance algorithms (PageRank, Hits). In recent years, new academic research has applied deep learning algorithms to graph data, resulting in graph deep learning algorithms such as Node2vec, Struct2vec, GNNs, and GCNs. These algorithms fundamentally assume that graph data lacks labeled data and learn graph features from the graph's inherent structure, enabling them to detect anomaly structures. Semi-supervised graph learning algorithms (LPA classification, belief propagation, and GCNs) can utilize a small number of fraudulent node labels, combined with information about the graph's relational structure, to probabilistically infer the fraud probability of other node entities. Overall, graph algorithms and semi-supervised learning algorithms can serve as complementary solutions to rule-based strategies and supervised learning, addressing the shortcomings of these approaches.

[0121] In one embodiment, for the aforementioned step S103, i.e., calculating the real-time risk score of the target customer based on the financial risk knowledge graph and the pre-built risk scoring model, the following methods may be used, including but not limited to:

[0122] First, based on the financial risk knowledge graph, graph computing algorithms are used to calculate the risk indicator data of target customers.

[0123] During the specific implementation, graph computing algorithms (such as PageRank, community discovery algorithms, etc.) are used to analyze the risk nodes and paths in the financial risk knowledge graph, and calculate risk indicator data such as enterprise-related risks, industry risks, and regional risks.

[0124] Then, based on the target customer's risk indicator data and the pre-built risk scoring model, a real-time risk score for the target customer is calculated.

[0125] In specific implementation, historical risk data and risk indicator data can be used in advance to train a risk scoring model using machine learning algorithms (such as random forest and XGBoost), and the model can be verified and optimized. Feature extraction and preprocessing are then performed on real-time data, and the trained risk scoring model is used to calculate the real-time risk score.

[0126] In one embodiment, after generating risk warning information based on real-time risk scores and pre-built risk warning rules, it also includes: classifying the risk warning information and prioritizing the risk warning information according to pre-set risk levels; generating risk disposal suggestions based on the risk warning information and the financial risk knowledge graph, and pushing the risk disposal suggestions to relevant personnel.

[0127] During specific implementation, after generating risk warning information, the risk warning information can also be classified and prioritized to obtain a warning information list; then, risk disposal suggestions (such as freezing accounts, strengthening monitoring, etc.) are generated based on the risk warning information and the financial risk knowledge graph, and the risk disposal suggestions are pushed to relevant personnel.

[0128] Specifically, in terms of risk warning, the knowledge graph connects customer data throughout the entire business life cycle of online loan applications, channels, risk control, approvals, contracts, lending, daily transactions, and post-loan behaviors in the business scenarios of the financial industry. In terms of data organization, the association graph abstracts entities and relationships from business data, reorganizes the data generated by the business in the form of a relationship network, and constructs complex relationships between entities such as people (ID cards), equipment, mobile phone numbers, bank accounts, business orders, and addresses. Based on the constructed association graph, with the help of graph theory, graph data mining, and graph deep learning algorithms, the relationship network in the association graph is subjected to topological structure analysis, group discovery, abnormal risk group mining, similar risk group mining, and risk propagation to identify business risks with complex relationship properties.

[0129] By combining knowledge graph technology with deep learning, using malicious fraud accounts as the analysis source, we mine relational attributes from multiple dimensions, fuse multi-source data to build a domain knowledge graph, and leverage AI algorithms to intelligently identify strongly correlated accounts, infer hidden relationships from complex networks, and identify fraud rings. We comprehensively construct a knowledge graph focusing on business flow relationships in the time dimension, geographic location aggregation relationships in the spatial dimension, "joint prevention and control" relationships in the platform dimension, business fact relationships, and device dimensions such as software and hardware.

[0130] Specifically, by implementing the knowledge graphs of various business fields on the basic big graph, we have accurately and completely constructed the underlying customer relationship network graph, account relationship network, capital flow network graph, group faction network graph, etc. at the level of all financial institutions. We have integrated the three-stage progressive data analysis methodology of graph analysis, graph model, and artificial intelligence graph algorithm to change the traditional thinking of single-layer customers, accounts and expert rule warnings. We have traced the relationship between members of more than three layers with multiple dimensions such as common telephone numbers, IP addresses, and login devices, and intuitively displayed the overall picture of each risk group. We can conduct macro analysis of upstream and downstream, as well as micro analysis of accounts in a certain link of the network group.

[0131] By using graph-based abnormal pattern recognition and artificial intelligence graph algorithms to form an effective early warning model, we can efficiently identify suspected money laundering gangs, gambling and fraud gangs, and risky group customers from within customer groups. This enables automated, intelligent, and multi-dimensional display, identification, and mining, significantly improving the accuracy and efficiency of risk group identification.

[0132] By querying specific conditions, you can view information about groups with different risk types and levels for specific product lines and events. This allows you to quantify the number of people in each group, their standard rate, and their risk score, providing a basis for business personnel to make decisions. This method also allows you to monitor changes in specific groups through the risk group monitoring function, providing a reference for further verification.

[0133] The advantage of the knowledge graph risk control solution is that it can effectively prevent and control various high-risk abnormal behaviors in various business processes such as application approval, repayment after approval, account opening and transaction, and provide real-time query, visual risk monitoring, risk control strategy management, targeted case assignment for high-risk groups and other services for institutional business operations and risk control personnel, thereby dynamically locating potential fraud and high-risk groups, tracing the means of committing crimes from the source, systematically predicting evolutionary trends and accurately quantitatively evaluating the scope and impact of high-sharing operations such as fraud, efficiently exploring various risks, and enhancing the ability to prevent unknown risks.

[0134] In one embodiment, the above method can also visualize the risks, specifically including:

[0135] (1) Draw a risk network diagram based on the financial risk knowledge graph and risk indicator data, and visualize the risk network diagram.

[0136] In specific implementation, visualization tools (such as D3.js, ECharts, etc.) can be used to draw a risk network diagram based on the financial risk knowledge graph and risk indicator data, and the risk network diagram can be displayed through the risk network visualization interface, so that users can interactively explore the risk network.

[0137] (2) Generate a multidimensional risk dashboard based on risk indicator data and risk warning information, and visualize the multidimensional risk dashboard; the multidimensional risk dashboard is used to display real-time risk scores, risk warning information, and risk transmission paths.

[0138] In specific implementation, a multi-dimensional risk dashboard can be designed to display risk scores, warning information, risk transmission paths, etc. Users can customize the dashboard content. Specifically, a multi-dimensional risk dashboard can be generated based on risk indicator data and risk warning information, and displayed through the risk dashboard interface.

[0139] In an embodiment of the present invention, a knowledge graph is displayed through a visual interface, and specific nodes and relationships can be displayed according to the node and relationship types; nodes displayed in the graph can be added or deleted; data details can be displayed, and the details have a timeline dragging function; batch highlighting, batch faded display, batch normal display, batch deletion display and other batch operations can be switched; support is provided for displaying graphs according to a specific structure; support is provided for displaying super large points whose associated nodes exceed a certain threshold, etc. By supporting the visual display of graphs and event timelines, business and time filtering of data is supported. In a relationship network, the relationship between two or more entities can be set to query, and drill up and drill down can be achieved, and related events and relationships can be quickly viewed to achieve highly available visual graph analysis. At the same time, using visual graph construction configuration, graph management and configurable knowledge display functions, business personnel can specify analysis objects according to needs, quickly generate graphs, and analyze the associated relationships of specified objects.

[0140] This implementation also supports graph language queries: custom query fields, support for combining multiple fields to query the graph to display all its relationships, and specific graph pattern retrieval. It can support large-scale graph query and analysis of hundreds of billions of edges, providing millisecond-level real-time point and edge precision queries; sub-second attribute filtering queries; second-level multi-layer path topology queries; and a rich set of built-in graph algorithms.

[0141] This implementation can also perform 10+ layers of deep link analysis, support queries such as retrieval paths and shortest paths, support forward, reverse or bidirectional traversal, and can traverse tens of millions of nodes per second to achieve deep link analysis.

[0142] Furthermore, the above method also includes: regularly acquiring new financial data and user feedback data; updating entities and relationships in the financial risk knowledge graph based on the new financial data; training the risk scoring model based on the new financial data, and optimizing risk warning rules and risk disposal recommendations based on user feedback data.

[0143] During the specific implementation, new financial data will be collected regularly, entities and relationships in the knowledge graph will be updated regularly, and the knowledge graph will be optimized and reconstructed; the risk scoring model will be retrained regularly, model performance will be optimized, and warning rules and disposal recommendations will be adjusted based on user feedback.

[0144] For ease of understanding, the present invention also provides a schematic diagram of a financial risk monitoring method, see Figure 2 As shown, including:

[0145] 1. Data collection and preprocessing: Collect data from multiple heterogeneous data sources, clean, convert and fuse them, and generate unified structured data.

[0146] 2. Knowledge graph construction: Build a financial risk knowledge graph through entity recognition, relationship extraction, and attribute extraction.

[0147] 3. Risk indicator calculation: Based on the knowledge graph, financial risk indicators are calculated using graph computing and machine learning techniques.

[0148] 4. Risk warning and disposal: Generate risk warning events and disposal suggestions based on risk indicators and user-defined warning rules.

[0149] 5. Visual display: Provide an interactive visual interface to display risk networks, risk indicators and warning events.

[0150] The above-mentioned method provided by the embodiment of the present invention includes a complete process from data collection to risk warning, which improves the problems of insufficient data integration capabilities, limited risk identification accuracy, and poor real-time performance in the existing technology. Through efficient data processing, accurate risk analysis and real-time warning mechanism, it can provide comprehensive and reliable risk monitoring services for financial institutions. Specifically, the above-mentioned method provided by the embodiment of the present invention has the following beneficial effects:

[0151] (1) Adopt automated data cleaning and fusion algorithms, reduce manual intervention, and adopt a unified data standardization framework to ensure the consistency and quality of multi-source data, thereby significantly improving data integration efficiency, reducing data processing time, supporting real-time data updates, and ensuring the timeliness of knowledge graphs.

[0152] (2) Utilize entity recognition, relationship extraction, and graph computing technologies to build a multi-level, multi-dimensional financial risk knowledge graph, introduce natural language processing (NLP) and deep learning technologies, and enhance the ability to analyze unstructured data, thereby improving the accuracy and comprehensiveness of risk identification, supporting the identification of complex risk networks, and the identification of emerging risks (such as cryptocurrency risks and cross-border capital flow risks).

[0153] (3) Adopt a distributed graph computing engine to support real-time processing of large-scale data, adopt a real-time risk warning mechanism, and dynamically update knowledge graphs and risk indicators, thereby realizing real-time processing and analysis of large-scale data, improving the real-time nature of risk monitoring, and realizing real-time monitoring of high-frequency transactions and dynamic risks.

[0154] (4) Use interactive visualization tools (such as ECharts and D3.js) to generate risk network diagrams and risk dashboards, and visualize and dynamically update multi-dimensional risk indicators, thereby providing rich risk visualization displays, meeting users' needs for intuitive understanding of complex risk networks, improving user experience, and helping users intuitively understand risk information.

[0155] (5) Provides a graphical interface to support business personnel in customizing graph modeling, adopts an automated graph construction algorithm, reduces manual intervention, and improves the efficiency and flexibility of graph construction.

[0156] (6) Using graph databases (such as Neo4j and HugeGraph) to store knowledge graph data, combined with ES components to store index data and relationship attributes, improves query efficiency.

[0157] (7) Integrate internal and external data to build a unified view of a single customer / portfolio dimension, and achieve full-process risk control before, during, and after the loan.

[0158] (8) The continuous incremental update mechanism of the knowledge graph is adopted to ensure the timeliness and accuracy of the data, support real-time risk monitoring and early warning, improve the decision-making support capabilities of the data center, and enhance the intelligence level of the data center.

[0159] The present invention also provides a knowledge graph platform, which is divided into eight modules: authority management, data management, knowledge construction, project space, knowledge analysis, knowledge mining, knowledge application, and system module.

[0160] (1) Permission Management: The permission system of the knowledge graph platform is based on the traditional RBAC model, which is role-based access control. On this basis, the concept of project organization is added to it, which is closer to the business and more conducive to the control of data permissions.

[0161] (2) Data Management: Data hierarchical storage uses the index database ElasticSearch, HBas e, and Hive for mixed data storage. The Spark distributed computing framework is used for big data computing. At the knowledge storage layer, this module is divided into distributed graph storage, distributed data warehouse, and distributed column database for interactive use, and provides a full-text search engine. The data management module mainly provides data management functions, distributed graph storage, distributed data warehouse, and distributed column database for interactive use, and provides a full-text search engine.

[0162] For example, the external data list displays basic information about all external data sources on the platform, presented in a list format. Users can quickly find matching data sources by entering filter criteria. The external data source list serves as a portal for environment management and creating new data sources, allowing for editing and deleting external data sources.

[0163] When users click the New Data Source button in the Data Source List, a New Data Source window pops up. This feature allows users to create various data sources, including ORCALE, MySQL, Hive, and DMC. Users need to fill in the basic information for connecting to the corresponding data source in the new interface.

[0164] (3) Knowledge construction: The knowledge construction module provides graph instance management, ontology model management, knowledge extraction, knowledge data access, verification, etc.

[0165] This module can realize the configuration construction of graph data and construct the ontology model in a visual way. The definition of graph metadata information can realize the definition of "entity points" and "edge relationships", thereby constructing the entire graph logical relationship. The system supports a top-down ontology model definition method, and can also define the model from the bottom up from the perspective of the data source. When defining an "entity point", it is necessary to determine the primary key identifier of the entity and the attribute information that needs to be added. The attribute information can be used as information for display or subsequent graph mining. When defining a "relationship edge", it is necessary to determine the primary key identifier of the edge as well as the starting point, end point, direction and other required attribute information of the edge.

[0166] Graph instance management consists of two parts: the graph instance list and ontology design. The graph instance list displays information about existing graph instances. On this page, operators can create new graph instances. Each graph instance corresponds to an ontology and can be independently configured with a data source.

[0167] Create entities and edges of new graph instances. Users can perform visual entity and relationship management in ontology design, create new entities and relationships, and manage corresponding fields and styles. Users are allowed to create, edit, and delete fields, and modify display color, size, and other information.

[0168] Create a new relationship edge in the graph instance. Relationship edges are divided into directed and undirected. There are four types of edges: normal edge, mining edge, detail edge, and summary edge.

[0169] ① Ordinary edge: General business relationship data that needs to be displayed in the graph, such as relationships such as "residence, ownership, purchase", can be imported into GDB.

[0170] ② Mining edges: Based on some existing data and specific business rules, the relational data output after certain calculations needs to be displayed in the graph, such as "concerted actors, same address" and other relationships, which can be imported into GDB.

[0171] ③ Detail edge: General original business relationship data has a large data volume and does not need to be directly displayed in the graph, such as "transfer details, call details" and other relationships. If the data volume of such relationships is average (tens of millions or hundreds of millions), if you do not want to introduce HBase to increase system complexity, you can directly store it in GDB or even ES. If the data volume reaches a massive level of tens of billions or hundreds of trillions, store it in Hbase.

[0172] ④ Summary Edges: Statistical relationships summarized based on detail edges need to be displayed in the graph, such as "transfer summary" and "call summary." Importing these into GDB is sufficient. Clicking a summary edge in the graph to view its details will find its corresponding detail edge data based on the detail_schema and detail_ids fields. The details panel to the left of the summary edge displays its associated detail edge information in a paginated table.

[0173] Field Management: Users can manage entity fields. Internal link configuration determines whether content found on the homepage search results can be redirected to a graphic analysis or profile. For example, if a company has internal links configured for graphic analysis and profiles, when searching on the homepage, clicking on a company can redirect to the graphic analysis or profile.

[0174] The entity identifier in field management is related to graph analysis - advanced search. When you start entity search in graph analysis, the attribute fields of the entity to be searched are automatically selected based on the entity identifier set in field management.

[0175] (4) Project space: The project space is used to manage and further control functions and data permissions in the graph platform. A project includes members, roles, and data. Most functions in the graph platform can only be used based on projects.

[0176] Users can create new projects in the project list, edit and delete existing projects, or select Project Management to further manage the selected project. The system will automatically switch to the selected project. The project list is divided into two tabs. "My Managed Projects" displays projects for which the user has management permissions, allowing management of project members and roles. "Other Projects" displays projects on the platform that the current user cannot directly manage. The display and related operations of this tab depend on the user's configured permissions.

[0177] (5) Knowledge Analysis: Provides basic graph operations, path analysis, relationship expansion, graph settings, and other functions. The graph analysis function provides users with basic graph analysis operations. Currently, it supports graph expansion, short path, and all path search. It also supports graph language query functions where users can query by entering graph analysis language. The current supported language for graph language query is Arango's AQL.

[0178] (6) Knowledge Mining: The graph mining module of the knowledge graph platform provides comprehensive graph mining capabilities, integrating general graph algorithms, encapsulated business graph algorithms, and custom algorithm workflows. It provides functions ranging from data processing, algorithm definition management and publishing, to scenario scheduling, and collaborates with other platforms to complete a one-stop graph construction service from data to application. In the entire knowledge graph construction, the mining platform plays a connecting role, discovering the business value in the underlying data for specific scenarios, providing basic data for subsequent analysis and application, and eliminating the gap between data and business.

[0179] (7) Knowledge application: Knowledge is ultimately applied to the search and analysis of map warehouse knowledge, providing scene exploration capabilities and developing a systematic knowledge base.

[0180] Three search types are supported: full-text search, tag search, and batch search. Search tags can be configured independently. Click the search name to switch between search types. All three types must be configured separately through the homepage search configuration under Project Management to be displayed.

[0181] ① Full-text search: Full-text search can search all business data tables according to the configuration, supporting quick search based on business data classification. The search history is displayed below the input box, and you can click to search directly for the keyword. You can also use the advanced search function on the right to search for multiple keywords.

[0182] ② Tag Search: Tag search enables the retrieval of single or batch entity tags. Using a pre-configured tag dictionary, you can quickly select the tags you want to search for. After selecting a tag, you can also add search keywords to achieve a "tag + keyword" search.

[0183] Click the input box to see the available tags for each group, allowing you to quickly select the desired tag. You can also enter keywords to fuzzy match the tag field, and the selected tags will be displayed below the input box. To include keywords in your search, search for the desired keywords in the input box and click Search.

[0184] ③ Batch search: Batch search uses spaces to separate search keywords, completing the search for batch keywords at once. After completing the initial search, you can also add further search keywords by turning on the append mode.

[0185] The advanced search of full-text search is a combination search of the "and" relationship, and the batch search is a combination search of the "or" relationship.

[0186] ④ Search results display: Search results are mainly divided into two parts: result categories and result cards. Result categories can quickly switch search types, while result cards display an overview of configured search results.

[0187] (8) System module: Provides basic system functions such as user management, login, permission management, business monitoring, log management, and access control. Taking business monitoring as an example, the business monitoring module displays the summary data of the access status of various services in the graph platform. Users can filter by time period to view information such as the total number of service visits, duration, and success rate within a fixed time range.

[0188] Regarding the financial risk monitoring method provided in the above embodiment, the present invention provides a financial risk monitoring system, see Figure 3 The structure diagram of a financial risk monitoring system shown in FIG. 1 illustrates that the system mainly includes the following parts:

[0189] The data collection module 301 is used to collect the target customer's financial data from multiple different data sources and fuse the financial data from different data sources to obtain a fused financial data set;

[0190] A knowledge graph construction module 302 is used to construct a financial risk knowledge graph based on the fused financial data set;

[0191] Risk assessment module 303, used to calculate the real-time risk score of the target customer based on the financial risk knowledge graph and the pre-built risk scoring model;

[0192] The risk warning module 304 is used to generate risk warning information based on real-time risk scores and pre-built risk warning rules.

[0193] The above-mentioned financial risk monitoring system of the present invention can integrate financial data from multiple different data sources, construct a financial risk knowledge graph, and calculate real-time risk scores in combination with a pre-built risk scoring model. Finally, risk warning information is generated based on the real-time risk scores and pre-built risk warning rules, thereby enabling efficient and accurate financial risk monitoring, improving risk prevention and control capabilities, and reducing the probability of risky business occurrence.

[0194] It should be noted that the implementation principle and technical effects of the device provided in the embodiment of the present invention are the same as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference can be made to the corresponding content in the aforementioned method embodiment.

[0195] An embodiment of the present invention further provides an electronic device. Specifically, the electronic device includes a processor and a storage device. The storage device stores a computer program, and when the computer program is executed by the processor, it executes the method described in any one of the above embodiments.

[0196] Figure 4 This is a structural diagram of an electronic device provided in an embodiment of the present invention. The electronic device 100 includes: a processor 40, a memory 41, a bus 42 and a communication interface 43. The processor 40, the communication interface 43 and the memory 41 are connected via the bus 42; the processor 40 is used to execute an executable module stored in the memory 41, such as a computer program.

[0197] The memory 41 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The system network element communicates with at least one other network element via at least one communication interface 43 (which may be wired or wireless), and may utilize the Internet, a wide area network, a local area network, a metropolitan area network, or the like.

[0198] The bus 42 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, Figure 4 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0199] Among them, the memory 41 is used to store programs, and the processor 40 executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any embodiment of the above-mentioned embodiment of the present invention can be applied to the processor 40 or implemented by the processor 40.

[0200] Processor 40 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method may be completed by hardware integrated logic circuits or software instructions in processor 40. The above processor 40 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It may implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present invention may be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or the like. The storage medium is located in the memory 41 , and the processor 40 reads the information in the memory 41 and completes the steps of the above method in combination with its hardware.

[0201] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be referred to the previous method embodiment and will not be repeated here.

[0202] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0203] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A financial risk monitoring method, characterized in that: include: Collecting financial data of target customers from multiple different data sources, and fusing the financial data from different data sources to obtain a fused financial data set; Constructing a financial risk knowledge graph based on the fused financial data set; Calculating a real-time risk score for the target customer based on the financial risk knowledge graph and a pre-built risk scoring model; Risk warning information is generated based on the real-time risk score and pre-built risk warning rules.

2. The method according to claim 1, characterized in that Constructing a financial risk knowledge graph based on the fused financial dataset includes: Extract entities from the fused financial dataset based on the BERT pre-trained model and the BiLSTM+CRF algorithm model to obtain an entity list; Based on the entity list and the fused financial dataset, extract the relationships between the entities, and perform type labeling and weight calculation on the relationships to obtain a relationship list; Extracting attributes of the entity based on the entity list and the fused financial data set to obtain an attribute list; A financial risk knowledge graph is constructed based on the entity list, the relationship list and the attribute list.

3. The method according to claim 2, characterized in that Constructing a financial risk knowledge graph based on the entity list and the relationship list includes: Constructing an initial financial risk knowledge graph based on the entity list and the relationship list; The MuGNN algorithm is used to perform entity alignment on the initial financial risk knowledge graph to obtain a financial risk knowledge graph.

4. The method according to claim 1, wherein Calculating the real-time risk score of the target customer based on the financial risk knowledge graph and the pre-built risk scoring model includes: Based on the financial risk knowledge graph, a graph computing algorithm is used to calculate the risk indicator data of the target customer; Calculate the real-time risk score of the target customer based on the risk indicator data of the target customer and a pre-built risk scoring model.

5. The method according to claim 4, characterized in that After generating risk warning information based on the real-time risk score and pre-built risk warning rules, the method further includes: Classify the risk warning information and prioritize the risk warning information according to pre-set risk levels; Generate risk disposal suggestions based on the risk warning information and the financial risk knowledge graph, and push the risk disposal suggestions to relevant personnel.

6. The method according to claim 4, characterized in that After generating risk warning information based on the real-time risk score and pre-built risk warning rules, the method further includes: Drawing a risk network diagram based on the financial risk knowledge graph and the risk indicator data, and visually displaying the risk network diagram; A multidimensional risk dashboard is generated based on the risk indicator data and the risk warning information, and the multidimensional risk dashboard is visually displayed; wherein the multidimensional risk dashboard is used to display real-time risk scores, risk warning information and risk transmission paths.

7. The method according to claim 5, characterized in that Also includes: Regularly obtain new financial data and user feedback data; Updating entities and relationships in the financial risk knowledge graph based on the new financial data; The risk scoring model is trained based on the new financial data, and the risk warning rules and the risk handling suggestions are optimized based on the user feedback data.

8. A financial risk monitoring system, characterized in that: include: A data collection module is used to collect financial data of target customers from multiple different data sources and fuse the financial data from different data sources to obtain a fused financial data set; A knowledge graph construction module, configured to construct a financial risk knowledge graph based on the fused financial data set; A risk assessment module, configured to calculate a real-time risk score for the target customer based on the financial risk knowledge graph and a pre-built risk scoring model; The risk warning module is used to generate risk warning information based on the real-time risk score and pre-built risk warning rules.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the steps of the method according to any one of claims 1 to 7.

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

Images