A method and device for identifying data security risks based on multi-dimensional feature fusion
By using a hybrid model of convolutional neural networks and long short-term memory networks, and combining static and dynamic features, a multi-dimensional feature fusion system is constructed. This solves the problems of incomplete feature capture, insufficient real-time performance, and high false alarm rate in existing technologies for data security risk identification, and achieves real-time risk identification with high accuracy and low false alarm rate.
Patent Information
- Application Number
- CN202611060120.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-16
- Publication Date
- 2026-11-10
AI Technical Summary
Existing data security risk identification methods suffer from problems such as one-sided feature capture, insufficient real-time performance, and high false alarm rate. They are difficult to effectively identify complex attacks and adapt to multi-source heterogeneous data, and lack a unified feature representation framework.
A hybrid model based on convolutional neural networks and long short-term memory networks is adopted, combining static and dynamic features. Key features are screened through information gain algorithm to construct a multi-dimensional feature fusion system to identify data security risks.
It achieves real-time risk identification with high accuracy (95.6%) and low false alarm rate (4.4%), significantly improving the initiative and efficiency of data security protection.
Smart Images

Figure CN123074754A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of information security technology, specifically a data security risk identification method, apparatus, device, medium, and product program based on multi-dimensional feature fusion. Background Technology
[0002] With the deepening of digital transformation across various industries, data has become a key production factor, leading to the rapid development of data security risk identification technologies. Existing technologies are mainly divided into three categories: static matching methods based on rule bases (such as the Snort intrusion detection system), detection methods based on single machine learning models (such as Support Vector Machine (SVM) and Random Forest), and post-event tracing methods based on log auditing. Among these, rule base methods rely on manually defined features, making them difficult to handle unknown attacks; single-model methods often use single data sources such as network traffic or user behavior, resulting in limitations in feature dimensions; and log auditing methods, due to data processing delays, cannot meet real-time protection requirements.
[0003] The aforementioned methods suffer from three major limitations: First, feature capture is one-sided. Traditional methods often rely on single-dimensional features (such as IP address and access frequency), ignoring the contextual relationships and dynamic changes in data, making it difficult to identify complex attacks. Second, real-time performance is insufficient. Batch-based risk analysis models (such as daily log summary analysis) have a detection delay of 5-10 minutes, failing to prevent ongoing risky behavior. Third, the false positive rate remains high. Existing models have low tolerance for abnormal fluctuations in normal behavior; in the financial sector, the false positive rate is generally over 15%, leading to a waste of security personnel's response resources. Furthermore, the heterogeneous formats of multi-source data (such as structured logs, unstructured traffic packets, and semi-structured API call records) further exacerbate the difficulty of risk feature fusion. Traditional methods lack a unified feature representation framework, resulting in "data silos."
[0004] In recent years, although deep learning technology has been introduced into the field of risk identification (such as malicious code detection based on Convolutional Neural Networks (CNNs)), most research remains focused on single-task optimization and has not yet formed an end-to-end, multi-dimensional risk assessment system. For example, an existing solution models user behavior sequences using a Long Short-Term Memory (LSTM) model, but fails to incorporate static attributes such as data sensitivity levels and access permissions, resulting in a detection rate of less than 60% for permission abuse risks. Therefore, developing a risk identification method that can integrate multi-source heterogeneous features and combines real-time performance with accuracy has become an urgent industry need. Summary of the Invention
[0005] The purpose of this application is to provide at least one method and device for identifying data security risks based on multi-dimensional feature fusion, in order to solve at least a part of the above-mentioned technical problems.
[0006] To address the aforementioned technical problems, at least one embodiment of this application provides a data security risk identification method based on multi-dimensional feature fusion, comprising: The data to be acquired for identifying security risks is divided into static features and dynamic features; The security risks of the data to be identified are determined based on the static features, the dynamic features, and the pre-generated security risk identification model. The security risk identification model is generated by training on historical data of the static features and historical data of the dynamic features, and is used to identify local correlation features of historical data of the static features and long-term and short-term dependencies of historical data of the dynamic features.
[0007] In some embodiments, the step of generating the security risk identification model includes: An initial model is generated based on convolutional neural networks and long short-term memory networks; The historical data of the static features and the historical data of the dynamic features are fused to generate a fused feature vector; The dimensionality of the fused feature vector is reduced based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The initial model is trained based on the fused feature vector after dimensionality reduction to generate the security risk identification model.
[0008] In some embodiments, the dimensionality reduction of the fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors includes: The first weight of the current fused feature vector is determined based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The fused feature vectors are filtered based on the first weight and a preset threshold.
[0009] In some embodiments, the first weight of the current fused feature vector is determined based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors, and is performed by the following formula:
[0010] in, , Let i be the information gain corresponding to the i-th fused feature vector. The information entropy of all fused feature vectors, The information entropy under the current fused feature vector condition is... This is the first weight.
[0011] In some embodiments, a data security risk identification method based on multi-dimensional feature fusion further includes: Local correlation features of the static features are extracted using a one-dimensional convolutional neural network; wherein the kernel size of the one-dimensional convolutional neural network is 3, and it includes 2 convolutional layers; The long-short-term dependencies of the dynamic features are captured by a bidirectional long short-term memory network; wherein the hidden layer dimension of the bidirectional long short-term memory network is 128 and it is a 2-layer stacked structure.
[0012] In some embodiments, identifying the security risks of the data to be identified based on the static features, the dynamic features, and a pre-generated security risk identification model includes: A fused feature vector is generated based on the static features and the dynamic features; The fused feature vector is divided into multiple attention heads by the pre-generated feature fusion layer in the security risk identification model; The second weight of the fusion feature vector corresponding to each attention head is determined based on the similarity between the fusion feature vector and the query vector. The fused feature vector is optimized based on multiple second weights; The security risks of the data to be identified are determined based on the optimized fusion feature vector.
[0013] In some embodiments, a second weight of the fused feature vector corresponding to each attention head is determined based on the similarity between the fused feature vector and the query vector, and is executed by the following formula:
[0014] in, The second weight is the i-th fused feature vector. The similarity score between the i-th fused feature vector and the query vector is given. m The number of attention heads.
[0015] At least one embodiment of this application also provides a data security risk identification device based on multi-dimensional feature fusion, comprising: The data segmentation module is used to divide the pre-acquired security risk data to be identified into static features and dynamic features; The data security risk identification module is used to identify the security risks of the data to be identified based on the static features, the dynamic features, and the pre-generated security risk identification model; wherein, the security risk identification model is trained and generated based on the historical data of the static features and the historical data of the dynamic features, and the security risk identification model is used to identify the local correlation features of the historical data of the static features and the long-term and short-term dependencies of the historical data of the dynamic features.
[0016] In some embodiments, a data security risk identification device based on multi-dimensional feature fusion further includes: A security risk identification model generation module is used to generate the security risk identification model; the security risk identification model generation module includes: The initial model generation unit is used to generate an initial model based on a convolutional neural network and a long short-term memory network. The first unit for generating the fused feature vector is used to fuse the historical data of the static features and the historical data of the dynamic features to generate a fused feature vector. The dimensionality reduction unit for the fused feature vector is used to reduce the dimensionality of the fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The security risk identification model generation unit is used to train the initial model based on the dimensionality-reduced fusion feature vector to generate the security risk identification model.
[0017] In some embodiments, the fusion feature vector dimensionality reduction unit includes: The first weight determination unit is used to determine the first weight of the current fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The fusion feature vector filtering unit is used to filter the fusion feature vector according to the first weight and a preset threshold.
[0018] In some embodiments, the first weight determination unit performs the following function:
[0019] in, , Let i be the information gain corresponding to the i-th fused feature vector. The information entropy of all fused feature vectors, The information entropy under the current fused feature vector condition is... This is the first weight.
[0020] In some embodiments, a data security risk identification device based on multi-dimensional feature fusion further includes: The local correlation feature extraction module is used to extract the local correlation features of the static features through a one-dimensional convolutional neural network; wherein the kernel size of the one-dimensional convolutional neural network is 3, and it includes 2 convolutional layers; The long-short-term dependency capture module is used to capture the long-short-term dependencies of the dynamic features through a bidirectional long short-term memory network; wherein the hidden layer dimension of the bidirectional long short-term memory network is 128 and it is a 2-layer stacked structure.
[0021] In some embodiments, the data security risk identification module identifies the security risks of the data to be identified based on the static features, the dynamic features, and a pre-generated security risk identification model, including: The second feature vector generation unit is used to generate a fused feature vector based on the static features and the dynamic features; The fusion feature vector partitioning unit is used to divide the fusion feature vector into multiple attention heads through the pre-generated feature fusion layer in the security risk identification model; The second weight determination unit is used to determine the second weight of the fusion feature vector corresponding to each attention head based on the similarity between the fusion feature vector and the query vector. A fusion feature vector optimization unit is used to optimize the fusion feature vector according to multiple second weights; The data security risk identification unit is used to identify the security risks of the data to be identified based on the optimized fusion feature vector.
[0022] In some embodiments, the second weight determination unit performs the following function:
[0023] in, The second weight is the i-th fused feature vector. The similarity score between the i-th fused feature vector and the query vector is given. m The number of attention heads.
[0024] At least one embodiment of this application also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described data security risk identification method based on multi-dimensional feature fusion.
[0025] At least one embodiment of this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described data security risk identification method based on multi-dimensional feature fusion.
[0026] The embodiments of this application provide a data security risk identification method and apparatus based on multi-dimensional feature fusion. First, the pre-acquired data to be identified for security risks is divided into static features and dynamic features. Then, the security risks of the data to be identified are identified based on the static features, dynamic features, and a pre-generated security risk identification model. The security risk identification model is trained and generated based on historical data of static features and historical data of dynamic features, and the security risk identification model is used to identify the local correlation features of historical data of static features and the long-term and short-term dependencies of historical data of dynamic features.
[0027] The method provided in this application can be deployed on an enterprise data security management platform. By accurately identifying risks such as internal data misuse and external intrusion, it can reduce the incidence of data breaches, significantly reduce the workload of ineffective responses, and significantly improve the initiative and efficiency of data security protection. Attached Figure Description
[0028] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.
[0029] Figure 1 This is a flowchart illustrating a data security risk identification method based on multi-dimensional feature fusion, provided in one embodiment of this application. Figure 2 This is a schematic diagram of a second process for a data security risk identification method based on multi-dimensional feature fusion, provided in one embodiment of this application; Figure 3 This is a flowchart illustrating step 300 provided in one embodiment of this application; Figure 4 This is a flowchart illustrating step 303 provided in one embodiment of this application; Figure 5 This is a schematic diagram of the third process of a data security risk identification method based on multi-dimensional feature fusion provided in one embodiment of this application; Figure 6 This is a flowchart illustrating step 200 provided in one embodiment of this application; Figure 7 This is a flowchart illustrating a data security risk identification method based on multi-dimensional feature fusion, provided by a specific embodiment of this application. Figure 8This is a mind map of a data security risk identification method based on multi-dimensional feature fusion provided by a specific implementation of this application; Figure 9 This is a flowchart illustrating step S1 provided in a specific embodiment of this application; Figure 10 This is a flowchart illustrating step S2 provided in a specific embodiment of this application; Figure 11 This is a flowchart illustrating step S3 provided in a specific embodiment of this application; Figure 12 This is a flowchart illustrating step S4 provided in a specific embodiment of this application; Figure 13 This is a block diagram of a data security risk identification device based on multi-dimensional feature fusion, provided in one embodiment of this application; Figure 14 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0031] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0032] It should be noted that the terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Without conflict, the embodiments and features in the embodiments of this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0033] The information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.
[0034] Provide users with corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.
[0035] The acquisition, storage, use, and processing of data in this application all comply with relevant laws and regulations. Specifically: First, the information collected is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data shall comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, not violate public order and good morals, and provide corresponding operation access points for users to choose to authorize or refuse.
[0036] Second, provide users with corresponding operation entry points for them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.
[0037] Example 1: This embodiment presents a data security risk identification method based on multi-dimensional feature fusion, which can be applied to electronic devices with communication, computing, and data storage capabilities. Its specific process can be as follows: Figure 1 As shown, it includes: Step 100: Divide the data to be acquired for security risk identification into static features and dynamic features; Step 200: Identify the security risks of the data to be identified based on the static features, the dynamic features, and the pre-generated security risk identification model; wherein, the security risk identification model is trained and generated based on the historical data of the static features and the historical data of the dynamic features, and the security risk identification model is used to identify the local correlation features of the historical data of the static features and the long-term and short-term dependencies of the historical data of the dynamic features.
[0038] The embodiments of this application provide a data security risk identification method based on multi-dimensional feature fusion. First, the pre-acquired data to be identified for security risks is divided into static features and dynamic features. Then, the security risks of the data to be identified are identified based on the static features, dynamic features, and a pre-generated security risk identification model. The security risk identification model is trained and generated based on historical data of static features and historical data of dynamic features, and the security risk identification model is used to identify the local correlation features of historical data of static features and the long-term and short-term dependencies of historical data of dynamic features.
[0039] This invention aims to address three major technical problems in existing data security risk identification methods: First, the risk characteristics of multi-source heterogeneous data are not fully captured. Traditional methods rely on a single data source (such as network traffic or user logs), ignoring key dimensions such as sensitive data attributes and access context, making it difficult to effectively identify complex attack patterns. Second, the risk assessment model has weak generalization ability. Existing models are mostly designed for specific scenarios (such as web attack detection), and their accuracy drops significantly when faced with new attack methods or cross-scenario data. Third, the accuracy of risk level determination is low. Traditional fixed-threshold determination methods cannot adapt to dynamically changing risk environments, resulting in a high rate of missed detection for high-risk events or a high rate of false detection for low-risk events. In practical applications, the accuracy of level determination is less than 75%.
[0040] Specifically, as shown in Table 1, this invention achieves three major technological breakthroughs through multi-dimensional feature fusion and dynamic evaluation mechanisms: the risk identification accuracy of the method of this invention reaches approximately 95.6%, the false alarm rate is reduced to below 4.4%, and the processing delay is controlled within 2 seconds, meeting the real-time protection requirements.
[0041] Table 1
[0042] Example 2: In some examples, the data to be identified for security risks in step 100 includes four core data sources: system log data (server logs, database audit logs, application logs), network traffic data (TCP / UDP session data, HTTP / HTTPS request packets, abnormal traffic alarms), user behavior data (operation sequence records, permission change logs, login and logout timestamps), and data asset attribute data (data classification and grading labels, storage location information, access control lists).
[0043] Referring to Table 2, this invention constructs a risk feature system from both static and dynamic dimensions. The dimension selection is based on the formation mechanism of data security risks: static features reflect the inherent attributes of the data and the access permissions of the subjects, while dynamic features depict the temporal changes in behavioral patterns. The static dimension includes four types of features: data sensitivity level (e.g., top secret / confidential / sensitive / public), access permission level (e.g., administrator / ordinary user / visitor), data storage location security level (e.g., intranet / DMZ / public network), and data transmission encryption status (e.g., TLS 1.3 / unencrypted). The dynamic dimension includes five types of features: access frequency (number of operations per unit time), access period anomaly (deviation from historical access periods), operation sequence anomaly (state transition probability based on Markov chains), cross-domain access frequency (access frequency between different data domains), and permission change frequency (number of permission escalations per unit time).
[0044] Table 2
[0045] Step 200 proposes a multi-dimensional feature fusion mechanism: breaking through the limitations of traditional single data sources, it innovatively constructs a fusion system of static features (data sensitivity level, permission level) and dynamic features (access sequence, operation time sequence), and solves the "feature island" problem by using a security risk identification model (CNN-LSTM hybrid model) and information gain weighting, thus significantly improving the feature representation capability.
[0046] Specifically, step 200 constructs a risk feature system from static and dynamic dimensions, uses a security risk identification model (deep learning model) to extract local correlation features of static features and temporal dependency features of dynamic features, filters key features through information gain algorithm, outputs a fused feature vector, and uses this to identify the security risks of the data to be identified.
[0047] In some examples, see Figure 2 A data security risk identification method based on multi-dimensional feature fusion also includes: Step 300: Generate the security risk identification model; then, see... Figure 3 Step 300 includes: Step 301: Generate an initial model based on a convolutional neural network and a long short-term memory network; Step 302: Fuse the historical data of the static features and the historical data of the dynamic features to generate a fused feature vector; Static features (dimension 64) and dynamic features (dimension 128) are concatenated to form a fused feature vector (dimension 192).
[0048] Step 303: Reduce the dimensionality of the fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors; Key features are selected from multiple fused feature vectors, and the selected 80-dimensional feature vectors are mapped to 128 dimensions through a fully connected layer as input to the feature fusion layer.
[0049] Step 304: Train the initial model based on the fusion feature vector after dimensionality reduction to generate the security risk identification model.
[0050] Specifically, the safety risk identification model was trained using an early stopping strategy (patience=5), with a batch size of 32, an initial learning rate of 0.001, and converged after 95 iterations.
[0051] In addition, the model training adopts a two-stage strategy: the pre-training stage trains the RBM parameters layer by layer through unsupervised learning, and the fine-tuning stage optimizes the overall network using supervised learning (with manually labeled risk scores), using the following cross-entropy loss function to minimize the prediction error:
[0052] In the formula, Here, N represents the cross-entropy loss value, and N is the total number of training samples. Let be the true label of the i-th sample (0 represents normal, 1 represents risk). The model predicts the probability value for the i-th sample.
[0053] In some examples, see Figure 4 Step 303 includes: Step 3031: Determine the first weight of the current fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors; Step 3032: Filter the fused feature vector according to the first weight and the preset threshold.
[0054] In some examples, step 3031 is performed by the following formula:
[0055] in, , Let i be the information gain corresponding to the i-th fused feature vector. The information entropy of all fused feature vectors, The information entropy under the current fused feature vector condition is... This is the first weight.
[0056] In some embodiments, the importance calculation of the fused feature vectors employs an improved information gain algorithm: , For dataset entropy, Features Entropy under certain conditions. By setting a threshold ( Key features were selected, and 80 core features were ultimately retained.
[0057] In some examples, see Figure 5 A data security risk identification method based on multi-dimensional feature fusion also includes: Step 400: Extract local correlation features of the static features using a one-dimensional convolutional neural network; wherein the kernel size of the one-dimensional convolutional neural network is 3, and it includes 2 convolutional layers; Step 500: Capture the long and short-term dependencies of the dynamic features through a bidirectional long short-term memory network; wherein the hidden layer dimension of the bidirectional long short-term memory network is 128 and it is a 2-layer stacked structure.
[0058] In steps 400 and 500, static features (such as sensitivity level and permission level) are transformed into low-dimensional vectors through the embedding layer and then input into a 1D-CNN network (with a kernel size of 3 and containing 2 convolutional layers) to extract local correlation features; for dynamic features (such as access sequence and operation time sequence), a bidirectional LSTM network (with a hidden layer dimension of 128 and 2 stacked layers) is used to capture their long-term and short-term dependencies.
[0059] In some examples, see Figure 6 Step 200 includes: Step 201: Generate a fused feature vector based on the static features and the dynamic features; Step 202: Divide the fused feature vector into multiple attention heads through the pre-generated feature fusion layer in the security risk identification model; Step 203: Determine the second weight of the fusion feature vector corresponding to each attention head based on the similarity between the fusion feature vector and the query vector; Step 204: Optimize the fused feature vector according to multiple second weights; Step 205: Identify the security risks of the data to be identified based on the optimized fusion feature vector.
[0060] In steps 201 to 205, the security risk identification model adopts a three-layer cascaded architecture: the input layer receives a multi-dimensional fused feature vector (80 dimensions), which is processed by batch normalization and then input into the feature fusion layer; the feature fusion layer adopts a multi-head attention mechanism, dividing the input features into 8 attention heads for parallel computation, and each attention head calculates feature weights (…). ,in Features The similarity score with the query vector is used to weight the features, which are then processed by residual connections and layer normalization to output fused features (128 dimensions). The risk assessment layer uses an improved deep belief network (DBN), which contains three restricted Boltzmann machine (RBM) hidden layers (number of neurons 128→64→32) and one output layer, and outputs a comprehensive risk score of 0-100 through the sigmoid activation function.
[0061] In some examples, step 203 is performed by the following formula:
[0062] in, The second weight is the i-th fused feature vector. The similarity score between the i-th fused feature vector and the query vector is given. m The number of attention heads.
[0063] In some examples, as shown in Table 3, this application classifies the risk level of the data to be identified as having security risks into four levels, setting dynamic thresholds based on a comprehensive risk score (0-100 points): low risk (0-20 points), medium risk (21-40 points), high risk (41-60 points), and severe risk (61-100 points). The characteristics corresponding to each level are as follows: low risk is normal operation (e.g., authorized users accessing public data), medium risk is potential anomaly (e.g., accessing sensitive data outside of working hours), high risk is explicit violation (e.g., unauthorized users attempting to access confidential data), and severe risk is attack behavior (e.g., batch downloading of sensitive files). The dynamic threshold adjustment mechanism uses a sliding window method, updating the thresholds for each level based on historical risk data from the past 30 days using the quartile method (Formula 4-1). ,in For smoothing coefficients, (This is the current window quartile value), ensuring that the threshold adapts to changes in the risk environment.
[0064] Table 3
[0065] In some examples, this application uses a fuzzy comprehensive evaluation algorithm to determine the risk level, which is implemented in three steps: first, constructing an evaluation matrix. ,in Features For level The membership degree is calculated (using the trapezoidal membership function); secondly, the feature weight vector is calculated. Finally, the overall score for each level is calculated. The level corresponding to the maximum value is taken as the final result. To improve real-time performance, the algorithm employs a pre-computation mechanism: membership function parameters and feature weights are generated offline, while only matrix multiplication and maximum value operations are performed online, with a single decision taking ≤50ms. For high-concurrency scenarios, GPU parallel computing increases the throughput to 1000 decisions / second, meeting the real-time decision requirements of large-scale data centers.
[0066] This application provides a data security risk identification method based on multi-dimensional feature fusion. First, the pre-acquired data to be identified regarding security risks is divided into static features and dynamic features. Then, the security risks of the data to be identified are determined based on the static features, dynamic features, and a pre-generated security risk identification model. The security risk identification model is trained and generated based on historical data of both static and dynamic features, and is used to identify local correlation features of historical data of static features and long-term and short-term dependencies of historical data of dynamic features. Compared with existing technologies, this application has the following advantages: Multi-dimensional feature fusion mechanism: Breaking through the limitations of traditional single data sources, it innovatively constructs a fusion system of static features (data sensitivity level, permission level) and dynamic features (access sequence, operation time sequence). Through the CNN-LSTM hybrid model and information gain weighting, it solves the "feature island" problem and significantly improves feature representation capabilities.
[0067] Dynamic threshold risk level determination: A dynamic threshold adjustment algorithm based on sliding window and fuzzy comprehensive evaluation is proposed to achieve the adaptability of risk level classification. The accuracy of level determination is significantly improved compared with the fixed threshold method, and it can effectively adapt to complex and ever-changing risk environments.
[0068] End-to-end integrated process design: Construct a closed loop across the entire chain from data collection, preprocessing, feature extraction to risk assessment and level determination. Through modular architecture and standardized interface design, seamless connection between each stage is achieved, and system deployment efficiency and maintenance costs are significantly optimized.
[0069] Example 3: For further explanation of the plan, see Figure 7 as well as Figure 8This application also provides a specific implementation of a data security risk identification method based on multi-dimensional feature fusion, which includes the following:
[0070] S1: Collect and preprocess data.
[0071] Collect heterogeneous data from multiple sources, including log data, network traffic data, user behavior data, and data asset attribute data. Clean, standardize, and reduce the dimensionality of the collected data to output a standardized feature dataset.
[0072] Specifically, log data (system logs, application logs), network traffic data (TCP / UDP packets), user behavior data (operation sequences, permission change records), and data asset attributes (sensitivity level, storage location) are accessed through a distributed acquisition framework. After cleaning, standardization, and dimensionality reduction, a feature dataset with a unified format is formed.
[0073] More specifically, step S1 adopts a distributed architecture design, covering four core data sources: system log data (server logs, database audit logs, application logs), network traffic data (TCP / UDP session data, HTTP / HTTPS request packets, abnormal traffic alarms), user behavior data (operation sequence records, permission change logs, login / logout timestamps), and data asset attribute data (data classification and grading tags, storage location information, access control lists). The collection method dynamically adapts to the data type: log data is captured in real-time using the Filebeat lightweight collector; network traffic is captured at high speed using mirrored ports combined with PF_RING technology; user behavior data is connected to the business system in real-time via API interfaces; and data asset attributes are updated through a combination of periodic scanning and manual annotation. To ensure real-time performance, the module adopts an edge computing architecture, performing initial filtering at the data generation node (e.g., collecting only logs containing sensitive operation keywords), achieving high-throughput data transmission through a Kafka message queue, controlling end-to-end latency to within 500ms, and supporting stable collection of over 100,000 data records per second.
[0074] See Figure 9 Data preprocessing includes three key steps: data cleaning, standardization, and dimensionality reduction. The data cleaning stage employs a multi-strategy combination: for missing values and continuous features, a weighted imputation method based on K-nearest neighbors is used. ,in (Assuming weights for nearest neighbor samples), categorical features are filled with the mode.
[0075] For outliers, an improved Z-score algorithm is used ( ,in (For absolute median) Identify and remove extreme values that deviate from 3 times the standard deviation.
[0076] The standardization process uses the Z-score method (Formula 1: ,in The characteristic mean, The standard deviation of the features is used. The eigenvalues are mapped to a distribution interval with a mean of 0 and a standard deviation of 1, eliminating dimensional differences. In the dimensionality reduction stage, principal component analysis (PCA) and feature importance pre-screening are combined. First, multicollinear features (VIF>10) are removed using the variance inflation factor (VIF). Then, PCA is used to compress the remaining features to 60% of the original dimension, retaining over 95% of the cumulative variance contribution. The preprocessed data shows a 40% improvement in signal-to-noise ratio and a 50% reduction in feature dimensionality, providing high-quality input for subsequent feature extraction.
[0077] exist Figure 9 In this process, after quality verification, multi-source data undergoes three core steps: cleaning (filling in missing values and removing outliers), standardization, and dimensionality reduction, ultimately outputting a standardized feature dataset. Key processing nodes (quality verification, cleaning, and dimensionality reduction) are highlighted with bold borders to ensure controllable data processing quality.
[0078] S2: Extract multi-dimensional risk features.
[0079] A risk feature system is constructed from static and dynamic dimensions. A deep learning model is used to extract local correlation features of static features and temporal dependency features of dynamic features. Key features are selected through information gain algorithm and a fused feature vector is output.
[0080] Specifically, a feature system is constructed from static dimensions (data sensitivity level, access subject permissions) and dynamic dimensions (access frequency, operation sequence abnormality). A CNN-LSTM hybrid model is used to extract local and temporal features, and key features are screened through information gain algorithm.
[0081] Static dimension features include data sensitivity level, access subject permission level, data storage location security level, and data transmission encryption status. Dynamic dimension features include access frequency, access time period anomaly, operation sequence anomaly, cross-domain access frequency, and permission change frequency.
[0082] More specifically, a risk characteristic system is constructed from both static and dynamic dimensions. The selection of dimensions is based on the formation mechanism of data security risks: static characteristics reflect the inherent attributes of the data and the access permissions of the subjects, while dynamic characteristics depict the temporal changes in behavioral patterns. The static dimension includes four types of characteristics: data sensitivity level (e.g., top secret / confidential / sensitive / public), access permission level (e.g., administrator / regular user / visitor), data storage location security level (e.g., intranet / DMZ / public network), and data transmission encryption status (e.g., TLS 1.3 / unencrypted). The dynamic dimension includes five types of characteristics: access frequency (number of operations per unit time), access period anomaly (deviation from historical access periods), operation sequence anomaly (state transition probability based on Markov chains), cross-domain access frequency (access frequency between different data domains), and permission change frequency (number of permission escalations per unit time).
[0083] See Figure 10 The static feature vector (64 dimensions) and dynamic feature vector (128 dimensions) output by the model are concatenated to form a fused feature matrix (192 dimensions). The training data for the CNN-LSTM hybrid model contains two types of features and risk labels: static features (such as data sensitivity level and user permission level) and dynamic features (such as the time sequence of continuous user operations), with labels indicating whether the operation is an abnormal risk. During training, the static features are first transformed into low-dimensional dense vectors through an embedding layer and input into a 1D-CNN network to extract local correlation features such as "sensitivity level + permission"; the dynamic features are captured by a bidirectional LSTM network to capture long-short-term time dependencies such as "login → download top-secret files"; the static feature vector output by the CNN and the dynamic feature vector output by the LSTM are concatenated to form the fused feature vector, which is then output as a risk probability through a fully connected layer. Finally, the model is iteratively trained using the Adam optimizer and the binary cross-entropy loss function, combined with an early stopping strategy to prevent overfitting.
[0084] exist Figure 10 In this process, static and dynamic features are extracted using CNN and LSTM networks respectively, then concatenated to form a fused feature matrix. Information gain is then used to filter key features. The multi-dimensional fusion logic is highlighted through subgraphs, demonstrating the collaborative extraction mechanism of static and dynamic features.
[0085] S3: Generate a security risk identification model.
[0086] A hierarchical security risk identification model is constructed, which includes an input layer, a feature fusion layer, and a risk assessment layer. The feature fusion layer uses an attention mechanism to dynamically allocate feature weights, and the risk assessment layer outputs a comprehensive risk score based on a deep belief network.
[0087] See Figure 11We designed a feature fusion layer based on the attention mechanism, dynamically allocated the weights of features of different dimensions, and combined it with an improved deep belief network (DBN) to build a risk assessment model and output a comprehensive risk score.
[0088] Referring to Table 4, the training dataset is a mixture of a public dataset and internal enterprise log data (1 million samples, 30% positive samples). The SMOTE algorithm is used to address class imbalance (oversampling minority class samples to achieve balance). Data augmentation employs feature perturbation: Gaussian noise is added to numerical features (…). The algorithm randomly replaces categorical features (with a replacement probability of 5%). Comparative optimization tests show that the Adam optimizer performs best in both convergence speed and accuracy, with its parameters set as follows: The learning rate uses a cosine annealing strategy (initially 0.001, decaying by 10% every 10 rounds). The model convergence condition is set as the validation set loss not decreasing for 5 consecutive rounds or reaching the maximum number of iterations (200 rounds).
[0089] Table 4
[0090] exist Figure 11 In this model, the fused features are preprocessed by the input layer and then output as a risk score through the feature fusion layer and the risk assessment layer. After performance evaluation, the optimized parameters are fed back. Key components of the model (feature fusion layer and risk assessment layer) are labeled with sub-icons to highlight the core roles of the attention mechanism and the DBN network.
[0091] S4: Dynamically determine the risk level through a security risk identification model.
[0092] Based on the fuzzy comprehensive evaluation algorithm, the risk level threshold is dynamically adjusted by combining historical risk data, and the risk score is mapped to four levels: low, medium, high, and severe, triggering the corresponding early warning mechanism.
[0093] Specifically, see Figure 12 The system combines a comprehensive risk score with a sliding window update of historical data to obtain the threshold. It then calculates scores for each level using fuzzy membership and a feature weight matrix, ultimately matching and outputting the risk level. The dynamic threshold adjustment logic is highlighted in a subgraph, demonstrating the mechanism by which the threshold adaptively updates with historical data.
[0094] Steps S1 to S4 are closely linked through data flow: the preprocessed data serves as the input for feature extraction, the extracted multi-dimensional features are calculated by the fusion model to obtain the risk score, and finally the risk level is output through the dynamic judgment algorithm, forming a complete closed loop of "data-feature-assessment-judgment".
[0095] Figure 8The entire chain from data access to risk output is presented: after receiving heterogeneous data from multiple sources, the data input layer processes the data through four core stages, ultimately outputting the risk level result and triggering the corresponding handling mechanism. An abnormal data feedback loop is set up in the data acquisition and preprocessing stage to ensure the quality of the input data.
[0096] The data security risk identification method based on multi-dimensional feature fusion provided in this application adopts a pipeline architecture, with each stage implementing data flow through standardized interfaces. The data acquisition and preprocessing stage, as the starting point, is responsible for transforming raw heterogeneous data into a feature dataset in a unified format; its output directly determines the completeness of subsequent feature extraction. The multi-dimensional feature extraction stage constructs static and dynamic feature matrices based on the preprocessed data, providing core input for risk assessment. The fusion risk assessment model dynamically weights multi-dimensional features using an attention mechanism, outputting a comprehensive risk score from 0 to 100. The dynamic risk level determination stage maps the score to a specific risk level based on a real-time updated threshold system. The process supports horizontal expansion; recognition capabilities can be improved by adding feature dimension modules or optimizing the model structure. Interfaces are reserved at each stage to facilitate functional iteration and integration with third-party systems.
[0097] The data security risk identification method based on multi-dimensional feature fusion provided in this application is mainly applied to scenarios with high data security requirements, such as enterprise data centers, cloud computing platforms, and financial transaction systems. This method integrates the risk features of multi-source heterogeneous data to achieve real-time monitoring and accurate identification of risks such as data leakage, unauthorized access, and abnormal operations, providing technical support for data security protection systems. This invention also involves interdisciplinary fields such as computer network technology, machine learning, data mining, and information security management, and its technical achievements can be widely adapted to the data security protection needs of critical information infrastructures in government, finance, and healthcare.
[0098] In summary, this invention achieves precise and real-time data security risk identification through multi-dimensional feature fusion and dynamic evaluation mechanisms. Experimental results show that the risk identification accuracy reaches 95.63%, a significant improvement over traditional methods; the false alarm rate is controlled below 4.4%, a significant reduction compared to traditional methods; and the average processing latency is reduced to 1.8 seconds, meeting real-time protection requirements. In deployment practice at a large financial enterprise's data center, within three months of system launch, 17 data breaches were successfully identified and blocked. The internal violation warning response time was shortened from 45 minutes to 90 seconds, the monthly incidence of risk events decreased significantly, and the workload of the security team in ineffective responses was reduced by more than 70%, significantly improving the initiative and efficiency of data security protection.
[0099] Example 4: Another embodiment of this application relates to a data security risk identification device based on multi-dimensional feature fusion. The implementation details of this embodiment's data security risk identification device based on multi-dimensional feature fusion are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution. A schematic diagram of this embodiment's data security risk identification device based on multi-dimensional feature fusion can be seen as follows: Figure 13 As shown, the device includes: The data segmentation module 801 is used to segment the pre-acquired security risk data to be identified into static features and dynamic features; The data security risk identification module 803 is used to identify the security risks of the data to be identified based on the static features, the dynamic features, and the pre-generated security risk identification model; wherein, the security risk identification model is trained and generated based on the historical data of the static features and the historical data of the dynamic features, and the security risk identification model is used to identify the local correlation features of the historical data of the static features and the long-term and short-term dependencies of the historical data of the dynamic features.
[0100] In some embodiments, a data security risk identification device based on multi-dimensional feature fusion further includes: A security risk identification model generation module is used to generate the security risk identification model; the security risk identification model generation module includes: The initial model generation unit is used to generate an initial model based on a convolutional neural network and a long short-term memory network. The first unit for generating the fused feature vector is used to fuse the historical data of the static features and the historical data of the dynamic features to generate a fused feature vector. The dimensionality reduction unit for the fused feature vector is used to reduce the dimensionality of the fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The security risk identification model generation unit is used to train the initial model based on the dimensionality-reduced fusion feature vector to generate the security risk identification model.
[0101] In some embodiments, the fusion feature vector dimensionality reduction unit includes: The first weight determination unit is used to determine the first weight of the current fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The fusion feature vector filtering unit is used to filter the fusion feature vector according to the first weight and a preset threshold.
[0102] In some embodiments, the first weight determination unit performs the following function:
[0103] in, , Let i be the information gain corresponding to the i-th fused feature vector. The information entropy of all fused feature vectors, The information entropy under the current fused feature vector condition is... This is the first weight.
[0104] In some embodiments, a data security risk identification device based on multi-dimensional feature fusion further includes: The local correlation feature extraction module is used to extract the local correlation features of the static features through a one-dimensional convolutional neural network; wherein the kernel size of the one-dimensional convolutional neural network is 3, and it includes 2 convolutional layers; The long-short-term dependency capture module is used to capture the long-short-term dependencies of the dynamic features through a bidirectional long short-term memory network; wherein the hidden layer dimension of the bidirectional long short-term memory network is 128 and it is a 2-layer stacked structure.
[0105] In some embodiments, the data security risk identification module identifies the security risks of the data to be identified based on the static features, the dynamic features, and a pre-generated security risk identification model, including: The second feature vector generation unit is used to generate a fused feature vector based on the static features and the dynamic features; The fusion feature vector partitioning unit is used to divide the fusion feature vector into multiple attention heads through the pre-generated feature fusion layer in the security risk identification model; The second weight determination unit is used to determine the second weight of the fusion feature vector corresponding to each attention head based on the similarity between the fusion feature vector and the query vector. A fusion feature vector optimization unit is used to optimize the fusion feature vector according to multiple second weights; The data security risk identification unit is used to identify the security risks of the data to be identified based on the optimized fusion feature vector.
[0106] In some embodiments, the second weight determination unit performs the following function:
[0107] in, The second weight is the i-th fused feature vector. The similarity score between the i-th fused feature vector and the query vector is given. m The number of attention heads.
[0108] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed in this application; however, this does not mean that other units are absent in this embodiment.
[0109] Example 5: Another embodiment of this application relates to an electronic device, such as... Figure 14 As shown, the electronic device specifically includes the following: Processor 1201, memory 1202, communications interface 1203, and bus 1204; The processor 1201, memory 1202, and communication interface 1203 communicate with each other via bus 1204; the communication interface 1203 is used to realize information transmission between server-side devices and user-side devices and other related devices. The processor 1201 is used to call the computer program in the memory 1202. When the processor executes the computer program, it implements all the steps in the data security risk identification method based on multi-dimensional feature fusion in the above embodiments. For example, when the processor executes the computer program, it implements the following steps: The data to be acquired for identifying security risks is divided into static features and dynamic features; The security risks of the data to be identified are determined based on the static features, the dynamic features, and the pre-generated security risk identification model. The security risk identification model is generated by training on historical data of the static features and historical data of the dynamic features, and is used to identify local correlation features of historical data of the static features and long-term and short-term dependencies of historical data of the dynamic features.
[0110] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.
[0111] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.
[0112] Example 6: Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps in the above-described embodiment of the data security risk identification method based on multi-dimensional feature fusion, the steps including: The data to be acquired for identifying security risks is divided into static features and dynamic features; The security risks of the data to be identified are determined based on the static features, the dynamic features, and the pre-generated security risk identification model. The security risk identification model is generated by training on historical data of the static features and historical data of the dynamic features, and is used to identify local correlation features of historical data of the static features and long-term and short-term dependencies of historical data of the dynamic features.
[0113] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, hardware + program embodiments are relatively simple in description because they are fundamentally similar to method embodiments; relevant parts can be referred to the descriptions in the method embodiments.
[0114] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.
[0115] While this application provides method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the method can be executed in the order shown in the embodiments or drawings or in parallel (e.g., in a parallel processor or multi-threaded processing environment).
[0116] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0117] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0118] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0119] This application uses specific embodiments to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A data security risk identification method based on multi-dimensional feature fusion, characterized in that, include: The data to be acquired for identifying security risks is divided into static features and dynamic features; The security risks of the data to be identified are determined based on the static features, the dynamic features, and the pre-generated security risk identification model. The security risk identification model is generated by training on historical data of the static features and historical data of the dynamic features, and is used to identify local correlation features of historical data of the static features and long-term and short-term dependencies of historical data of the dynamic features.
2. The data security risk identification method according to claim 1, characterized in that, The steps for generating the security risk identification model include: An initial model is generated based on convolutional neural networks and long short-term memory networks; The historical data of the static features and the historical data of the dynamic features are fused to generate a fused feature vector; The dimensionality of the fused feature vector is reduced based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The initial model is trained based on the fused feature vector after dimensionality reduction to generate the security risk identification model.
3. The data security risk identification method according to claim 2, characterized in that, The step of reducing the dimensionality of the fused feature vector based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors includes: The first weight of the current fused feature vector is determined based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors. The fused feature vectors are filtered based on the first weight and a preset threshold.
4. The data security risk identification method according to claim 3, characterized in that, Based on the information gain corresponding to the current fused feature vector and the information gains corresponding to all fused feature vectors, the first weight of the current fused feature vector is determined by the following formula: in, , Let i be the information gain corresponding to the i-th fused feature vector. The information entropy of all fused feature vectors, The information entropy under the current fused feature vector condition is... This is the first weight.
5. The data security risk identification method according to claim 2, characterized in that, Also includes: Local correlation features of the static features are extracted using a one-dimensional convolutional neural network; wherein the kernel size of the one-dimensional convolutional neural network is 3, and it includes 2 convolutional layers; The long-short-term dependencies of the dynamic features are captured by a bidirectional long short-term memory network; wherein the hidden layer dimension of the bidirectional long short-term memory network is 128 and it is a 2-layer stacked structure.
6. The data security risk identification method according to claim 1, characterized in that, Identifying security risks in the data to be identified based on the static features, the dynamic features, and the pre-generated security risk identification model includes: A fused feature vector is generated based on the static features and the dynamic features; The fused feature vector is divided into multiple attention heads by the pre-generated feature fusion layer in the security risk identification model; The second weight of the fusion feature vector corresponding to each attention head is determined based on the similarity between the fusion feature vector and the query vector. The fused feature vector is optimized based on multiple second weights; The security risks of the data to be identified are determined based on the optimized fusion feature vector.
7. The data security risk identification method according to claim 6, characterized in that, The second weight of the fusion feature vector corresponding to each attention head is determined based on the similarity between the fused feature vector and the query vector, and is executed by the following formula: in, The second weight is the i-th fused feature vector. The similarity score between the i-th fused feature vector and the query vector is given. m The number of attention heads.
8. A data security risk identification device based on multi-dimensional feature fusion, characterized in that, include: The data segmentation module is used to divide the pre-acquired security risk data to be identified into static features and dynamic features; The data security risk identification module is used to identify the security risks of the data to be identified based on the static features, the dynamic features, and the pre-generated security risk identification model; wherein, the security risk identification model is trained and generated based on the historical data of the static features and the historical data of the dynamic features, and the security risk identification model is used to identify the local correlation features of the historical data of the static features and the long-term and short-term dependencies of the historical data of the dynamic features.
9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the data security risk identification method based on multi-dimensional feature fusion as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the data security risk identification method based on multi-dimensional feature fusion as described in any one of claims 1 to 7.
