Financial fraud detection method based on natural language processing and graph neural network
Through natural language processing and graph neural network technology, combined with data preprocessing and graph neural network model, the problems of accuracy and timeliness of existing financial fraud detection methods are solved, and in-depth analysis of financial transaction data is realized, which has improved the The accuracy and efficiency of fraud detection provide financial institutions with a more reliable prevention solution.
Patent Information
- Application Number
- CN202411289326.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-14
- Publication Date
- 2024-12-03
AI Technical Summary
Existing financial fraud detection methods are difficult to fully capture the complexity and diversity of fraudulent behavior, and the detection accuracy is not high. In addition, traditional methods have poor real-time performance when processing large-scale data, which is difficult to meet the requirements of financial institutions for the timeliness of fraud detection.
Using a method based on natural language processing and graph neural network, through data preprocessing and feature extraction, constructing graph neural network models, real-time detection and early warning, and optimizing and iterating the fraud detection model, integrating text features and network features to achieve In-depth mining and analysis of financial transaction data.
It improves the accuracy and efficiency of fraud detection, can accurately identify complex and changeable fraud behaviors, and provides financial institutions with more reliable fraud prevention solutions.
Smart Images

Figure CN119067674A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of financial fraud detection, and specifically is a financial fraud detection method based on natural language processing and graph neural network. Background Art
[0002] With the rapid development of financial services and the widespread application of Internet technology, financial fraud is becoming increasingly rampant, causing huge economic losses and reputation risks to financial institutions and customers. Traditional fraud detection methods often rely on manual review and rule matching, which have problems such as low efficiency, low accuracy, and easy omissions. Therefore, there is an urgent need for an intelligent and automated fraud detection system to cope with increasingly complex fraud scenarios.
[0003] At present, a variety of technical solutions have emerged in the field of financial fraud detection, including statistical-based detection methods, machine learning-based detection methods, and deep learning-based detection methods. Among them, deep learning-based detection methods have attracted widespread attention due to their powerful feature learning and generalization capabilities. However, most of the existing deep learning detection methods focus on the processing of a single data source, such as detection based only on transaction data or detection based only on text data, which makes it difficult to fully capture the characteristics of fraudulent behavior.
[0004] Existing technologies often only use a single data source for detection, which cannot fully reflect the complexity and diversity of fraudulent behavior and has low detection accuracy. Traditional methods find it difficult to effectively extract deep features from text data and associated network data, resulting in limited detection accuracy. Some methods have poor real-time performance when processing large-scale data and are unable to meet the requirements of financial institutions for the timeliness of fraud detection. Summary of the invention
[0005] In view of the above problems, the present invention is proposed to provide a financial fraud detection method based on natural language processing and graph neural network to overcome the above problems or at least partially solve the above problems.
[0006] To achieve the above object, the present invention adopts the following technical solutions:
[0007] A financial fraud detection method based on natural language processing and graph neural network, the method comprising:
[0008] Data preprocessing and feature extraction;
[0009] Build a graph neural network model;
[0010] Build fraud detection models;
[0011] Conduct real-time detection and early warning;
[0012] Optimize and iterate fraud detection models.
[0013] Optionally, data preprocessing and feature extraction include:
[0014] Collecting data, including structured data including transaction amounts, times, and counterparties, and unstructured data including transaction notes, customer chat logs, and social media posts;
[0015] Perform data cleaning on the collected data, including denoising, normalization and missing value processing;
[0016] Perform feature extraction on the collected data, which includes:
[0017] Extract transaction amounts, timing, frequency, and counterparties from structured data;
[0018] Extract keywords, sentiment, and topics from unstructured data.
[0019] Optionally, building a graph neural network model includes:
[0020] Model selection: select a graph neural network model based on the application scenario and data characteristics;
[0021] Feature encoding: encode the nodes and edges in the graph neural network model, and embed text features and statistical features into a low-dimensional vector space;
[0022] Feature propagation and aggregation: feature propagation and aggregation are performed through graph neural network models, and dependencies and interaction patterns between nodes are learned;
[0023] Fraud detection, based on the aggregated node features, classifies and judges whether the node is a fraudulent user or a fraudulent transaction.
[0024] Optionally, building a fraud detection model includes:
[0025] Combining text features extracted by natural language processing and network features learned by graph neural networks to build a fraud detection model;
[0026] Use supervised learning or semi-supervised learning methods to train fraud detection models, using historical fraud cases and normal transaction cases as training data;
[0027] Automatically learn the feature representation and discrimination rules of fraudulent behavior, and perform real-time fraud detection on new transaction data.
[0028] Optionally, real-time detection and early warning are performed including real-time data processing and fraud detection.
[0029] Optionally, the real-time data processing includes:
[0030] Process and analyze newly generated transaction data in real time;
[0031] Real-time natural language processing on unstructured data.
[0032] Optionally, the fraud detection includes:
[0033] Leverage trained natural language processing and graph neural network models to perform fraud detection on new data;
[0034] Generate fraud warning information based on the detection results and notify relevant personnel to take corresponding measures.
[0035] Optionally, optimization and iteration of the fraud detection model include:
[0036] Continuously optimize and adjust fraud detection models based on detection results and user feedback;
[0037] Introduce new data sources and algorithms to improve the detection accuracy and generalization ability of fraud detection models.
[0038] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:
[0039] By integrating natural language processing and graph neural network technology, the present invention realizes in-depth mining and analysis of text information and associated networks in financial transaction data, can accurately identify complex and changeable fraud behaviors, improve the accuracy and efficiency of fraud detection, and provide financial institutions with more reliable fraud prevention solutions. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 A flowchart of a financial fraud detection method based on natural language processing and graph neural network is provided in an embodiment of the present application. DETAILED DESCRIPTION
[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0042] See also Figure 1 This embodiment provides a financial fraud detection method based on natural language processing and graph neural network, including:
[0043] S1. Data preprocessing and feature extraction. Data sources include transaction systems, user management systems, customer service systems, etc. within financial institutions, as well as external public data sources such as social media and news websites.
[0044] Data preprocessing and feature extraction include:
[0045] Collect data, including structured data and unstructured data, wherein the structured data includes transaction amount, time, and counterparty, and the unstructured data includes transaction notes, customer chat records, and social media posts.
[0046] Data cleaning is performed on the collected data, which includes denoising, normalization and missing value processing. Denoising: removing duplicate, erroneous and irrelevant data entries. Normalization: unifying data formats, such as date format, currency unit, etc. Missing value processing: using interpolation, deletion or marking of missing values to process incomplete data.
[0047] Perform feature extraction on the collected data, which includes:
[0048] Extract transaction amount, time, frequency, and counterparty features from structured data.
[0049] Natural language processing technology is used to extract features such as keywords, sentiment tendencies and topics from unstructured data.
[0050] S2. Build a graph neural network model.
[0051] Building a graph neural network model includes:
[0052] Model selection: select a suitable graph neural network model according to the application scenario and data characteristics, such as GCN, GAT, GraphSAGE, etc.
[0053] Feature encoding: feature encoding is performed on nodes and edges in the graph neural network model, and text features and statistical features are embedded into a low-dimensional vector space.
[0054] Feature propagation and aggregation, feature propagation and aggregation are performed through graph neural network models, and the dependencies and interaction patterns between nodes are learned.
[0055] Fraud detection, based on the aggregated node features, classifies and judges whether the node is a fraudulent user or a fraudulent transaction.
[0056] S3. Build a fraud detection model.
[0057] Building a fraud detection model involves:
[0058] The fraud detection model is constructed by combining text features extracted by natural language processing and network features learned by graph neural networks.
[0059] The fraud detection model is trained using supervised learning or semi-supervised learning methods, using historical fraud cases and normal transaction cases as training data.
[0060] Automatically learn the feature representation and discrimination rules of fraudulent behavior, and perform real-time fraud detection on new transaction data.
[0061] Training and evaluating the model includes:
[0062] The model is trained in a supervised manner using historical fraud cases and normal transaction cases as training data.
[0063] Cross-validation and other methods are used to evaluate the performance of the model, including indicators such as accuracy, recall, and F1 score.
[0064] Adjust the model structure and parameters according to the evaluation results to optimize the detection effect of the model.
[0065] S4. Conduct real-time detection and early warning.
[0066] Real-time detection and early warning include real-time data processing and fraud detection.
[0067] The real-time data processing includes:
[0068] Newly generated transaction data is processed and analyzed in real time to ensure that fraudulent activities can be detected in a timely manner.
[0069] Perform real-time natural language processing on unstructured data to extract key information and sentiment.
[0070] The fraud detection includes:
[0071] Leverage trained natural language processing and graph neural network models for fraud detection on new data.
[0072] Generate fraud warning information based on the detection results and notify relevant personnel to take corresponding measures.
[0073] For suspected fraudulent transactions (i.e. transactions with a fraud probability exceeding a preset threshold or belonging to a specific fraud category), an early warning mechanism can be automatically triggered.
[0074] The early warning information includes key information such as transaction details, fraud type, risk level, etc., so that relevant institutions or personnel can conduct further investigation and processing.
[0075] S5. Optimize and iterate the fraud detection model.
[0076] Optimizing and iterating the fraud detection model includes:
[0077] Based on detection results and user feedback, we continuously optimize and adjust the fraud detection model, analyze false positive and missed negative cases, and identify problems and deficiencies in the model.
[0078] Continuously introduce new data sources and transaction data to enrich the model's training samples and feature space. Regularly update and clean the data to keep it fresh and accurate.
[0079] Focus on the latest research results and technological advances in the field of natural language processing and graph neural networks, introduce new data sources and algorithms, and improve the detection accuracy and generalization ability of fraud detection models.
[0080] This embodiment integrates natural language processing and graph neural network technology to achieve in-depth mining and analysis of text information and associated networks in financial transaction data. It can accurately identify complex and changeable fraud behaviors, improve the accuracy and efficiency of fraud detection, and provide financial institutions with more reliable fraud prevention solutions.
[0081] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.
Claims
1. A financial fraud detection method based on natural language processing and graph neural network, characterized in that: The method comprises: Data preprocessing and feature extraction; Build a graph neural network model; Build fraud detection models; Conduct real-time detection and early warning; Optimize and iterate fraud detection models.
2. A financial fraud detection method based on natural language processing and graph neural network as claimed in claim 1, characterized in that: Data preprocessing and feature extraction include: Collecting data, including structured data including transaction amounts, times, and counterparties, and unstructured data including transaction notes, customer chat logs, and social media posts; Perform data cleaning on the collected data, including denoising, normalization and missing value processing; Perform feature extraction on the collected data, which includes: Extract transaction amounts, timing, frequency, and counterparties from structured data; Extract keywords, sentiment, and topics from unstructured data.
3. A financial fraud detection method based on natural language processing and graph neural network as claimed in claim 1, characterized in that: Building a graph neural network model includes: Model selection: select a graph neural network model based on the application scenario and data characteristics; Feature encoding: encode the nodes and edges in the graph neural network model, and embed text features and statistical features into a low-dimensional vector space; Feature propagation and aggregation: feature propagation and aggregation are performed through graph neural network models, and dependencies and interaction patterns between nodes are learned; Fraud detection, based on the aggregated node features, classifies and judges whether the node is a fraudulent user or a fraudulent transaction.
4. A financial fraud detection method based on natural language processing and graph neural network as claimed in claim 1, characterized in that: Building a fraud detection model involves: Combining text features extracted by natural language processing and network features learned by graph neural networks to build a fraud detection model; Use supervised learning or semi-supervised learning methods to train fraud detection models, using historical fraud cases and normal transaction cases as training data; Automatically learn the feature representation and discrimination rules of fraudulent behavior, and perform real-time fraud detection on new transaction data.
5. A financial fraud detection method based on natural language processing and graph neural network as claimed in claim 1, characterized in that: Real-time detection and early warning include real-time data processing and fraud detection.
6. A financial fraud detection method based on natural language processing and graph neural network as claimed in claim 5, characterized in that: The real-time data processing includes: Process and analyze newly generated transaction data in real time; Real-time natural language processing on unstructured data.
7. A financial fraud detection method based on natural language processing and graph neural network as claimed in claim 5, characterized in that: The fraud detection includes: Leverage trained natural language processing and graph neural network models to perform fraud detection on new data; Generate fraud warning information based on the detection results and notify relevant personnel to take corresponding measures.
8. A financial fraud detection method based on natural language processing and graph neural network as claimed in claim 1, characterized in that: Optimizing and iterating the fraud detection model includes: Continuously optimize and adjust fraud detection models based on detection results and user feedback; Introduce new data sources and algorithms to improve the detection accuracy and generalization ability of fraud detection models.
