Intelligent malicious code detection method and system based on big data mining analysis

By combining big data mining and analysis with a hybrid detection model, integrating temporal convolutional networks and graph neural networks, and incorporating lightweight sandbox technology, the problem of low efficiency and high false positive rate in existing malware detection methods has been solved. This enables real-time monitoring and accurate identification of malware, thereby improving network security.

CN118965344APending Publication Date: 2024-11-15华能庆阳煤电有限责任公司
View PDF 0 Cites 6 Cited by

Patent Information

Application Number
CN202410838366.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-26
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Existing malware detection technologies rely on a single feature, resulting in low detection efficiency and a high false positive rate, making it difficult to effectively identify disguised or mutated malware.

Method used

A big data mining and analysis approach is adopted. Through multi-source data collection and preprocessing, a hybrid detection model is established. Feature extraction is performed by combining temporal convolutional networks and graph neural networks. Suspicious programs are isolated using a lightweight sandbox. System calls and interactive operations are monitored and recorded, alarm information is generated and processed automatically, and the detection model is optimized.

Benefits of technology

It enables real-time monitoring and rapid response to malicious code, reduces false alarm rates, improves detection efficiency and accuracy, and comprehensively protects network security.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965344A_ABST
    Figure CN118965344A_ABST
Patent Text Reader

Abstract

The invention discloses an intelligent malicious code detection method and system based on big data mining analysis, and the method comprises the steps: collecting and preprocessing multi-source data of a target system, extracting behavior features and structural features of the multi-source data, and building a detection model based on the extracted features; real-time detection is carried out by utilizing a detection model, and when a suspicious program is detected, all system calling and interaction operations of the suspicious program are isolated, monitored and recorded by utilizing a lightweight sandbox; analyzing behaviors of suspicious programs in the sandbox and generating alarm information, performing automatic processing according to the alarm information, and feeding back and optimizing the detection model according to a processing result to realize intelligent detection of malicious codes. According to the method, the malicious codes are comprehensively monitored and accurately identified, the detection efficiency and accuracy are improved, the false alarm rate is reduced, and the network security is effectively protected.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of network security technology, and in particular to a malicious code intelligent detection method and system based on big data mining and analysis. Background Art

[0002] With the rapid development of information technology, the network environment is becoming increasingly complex, and the types and number of malicious codes are also increasing. Traditional malicious code detection methods mainly rely on the surface characteristics of the code or system behavior. However, these methods are often ineffective when facing disguised or mutated malicious codes. Therefore, it is of great practical significance to study a malicious code intelligent detection method based on big data mining and analysis.

[0003] The existing malicious code detection technology has the following main shortcomings:

[0004] Reliance on a single feature: Traditional malicious code detection methods often rely on the surface features of the code or system behavior. This method is often ineffective when facing disguised or mutated malicious code.

[0005] Low detection efficiency: Existing malicious code detection methods often require analyzing each program one by one, resulting in low detection efficiency.

[0006] High false alarm rate: Due to the diversity of malicious code, existing malicious code detection methods often have a high false alarm rate.

[0007] The present invention provides a malicious code intelligent detection method and system based on big data mining and analysis, aiming to effectively solve the problems in the existing malicious code detection technology. Summary of the invention

[0008] In view of the above-mentioned problems, the present invention is proposed.

[0009] Therefore, the technical problem solved by the present invention is that the existing malicious code detection technology relies on a single feature, has low detection efficiency and a high false alarm rate.

[0010] To solve the above technical problems, the present invention provides the following technical solutions: a method for intelligent detection of malicious code based on big data mining and analysis, comprising: collecting and preprocessing multi-source data of the target system, extracting behavioral features and structural features of the multi-source data, and establishing a detection model based on the extracted features; using the detection model for real-time detection, and using a lightweight sandbox to isolate suspicious programs when suspicious programs are detected, monitoring and recording all system calls and interactive operations of the suspicious programs; analyzing the behavior of suspicious programs in the sandbox and generating alarm information, performing automatic processing based on the alarm information, and optimizing the detection model based on the processing result feedback to achieve intelligent detection of malicious code.

[0011] As a preferred solution of the malicious code intelligent detection method based on big data mining and analysis described in the present invention, the multi-source data includes obtaining network traffic data through a network traffic monitoring tool, obtaining file operation records and process, thread information and API call records through system monitoring, obtaining system log data from operating system and application logs, and collecting known malicious code samples from a public malicious code library.

[0012] As a preferred solution of the malicious code intelligent detection method based on big data mining and analysis described in the present invention, the detection model includes: performing behavioral feature analysis and structural feature analysis on the target program. When analyzing the behavioral features, considering that the program's behavioral pattern has a temporal relationship, a long short-term memory network layer is added to the neural network structure to enhance the ability to capture the temporal pattern in the system call sequence. The main body of the neural network selects a temporal convolutional network, which is expressed as: an input layer, inputting the target program behavior features; a TCN layer, using TCN to extract temporal features, capturing long-distance dependencies by expanding the convolution layer, introducing multi-scale convolution kernels, and weighting the convolutional features through the attention mechanism; a feature aggregation layer, aggregating important features of the output of the TCN layer across time steps, enhancing the expression ability of the features, so that the subsequent LSTM layer can better capture the temporal relationship, expressed as:

[0013]

[0014] LSTM layer, composed of a single LSTM layer through a gating mechanism and memory units, extracts H agg The timing relationship in which H agg represents the output of the feature aggregation layer; T represents the length of the sequence; H TCN Represents the output of the TCN layer; [:, t, :] represents the selection of all filter outputs at the t-th time step in all batches.

[0015] As a preferred solution of the malicious code intelligent detection method based on big data mining and analysis described in the present invention, wherein: the detection model also includes analyzing the structural features of the program sample through a graph neural network. The malicious code usually has a complex calling relationship and dependency structure. The graph structure features are extracted layer by layer through hierarchical graph convolution. The structure of the graph neural network is represented as follows: an input layer, which converts the structural features of the target program into a node feature matrix X and an adjacency matrix A. The node feature matrix X contains the feature information of each node and has a shape of (N, F). N is the number of nodes, representing the basic unit in the program. F represents the number of features of each node. The adjacency matrix A represents the connection relationship between nodes and has a shape of (N, N); a hierarchical graph convolution layer, which is composed of a local convolution layer, a middle convolution layer and a global convolution layer. The local convolution layer preliminarily extracts node features and captures the relationship between directly adjacent nodes. The node feature matrix X is weighted and summed using the adjacency matrix A, and then the output H of the local convolution layer is obtained by linear transformation through the weight matrix. 1 ; The middle convolution layer uses the adjacency matrix A to the output H of the local convolution layer 1 Perform weighted summation to capture the multi-hop neighbor relationship of nodes and extract more complex structural features. The middle convolutional layer outputs H 2 ; The global convolution layer captures the global features of the entire graph structure, extracts global dependencies, and uses the adjacency matrix A to convolutionally transform the output H of the middle convolution layer. 1 Perform weighted summation and skip connection of local convolutional layers, using the deformation of the adjacency matrix A 2 The output H of the local convolutional layer 1 Perform weighted summation to capture the dependencies between distant nodes, and obtain the output H of the global convolutional layer through weighted summation 3 .

[0016] As a preferred solution of the intelligent detection method of malicious code based on big data mining and analysis described in the present invention, the lightweight sandbox isolation includes simulating the target system and application software environment in an isolated sandbox environment to ensure that the suspicious program can exhibit real behavior during execution, loading the suspicious program judged by the detection model into the sandbox and starting execution; during the sandbox execution process, monitoring and recording all system calls and interactive operations of the suspicious program, the system calls include file operations, network communications, process management and memory operations, and the interactive operations include user input and program output; comparing the recorded system calls and interactive operations with known malicious behavior patterns, identifying abnormal system calls and interactive operations, and generating an analysis report to describe the detected vulnerability attacks and potential threats.

[0017] As a preferred solution of the intelligent detection method of malicious code based on big data mining and analysis described in the present invention, the lightweight sandbox isolation includes simulating the target system and application software environment in an isolated sandbox environment to ensure that the suspicious program can exhibit real behavior during execution, loading the suspicious program judged by the detection model into the sandbox and starting execution; during the sandbox execution process, monitoring and recording all system calls and interactive operations of the suspicious program, the system calls include file operations, network communications, process management and memory operations, and the interactive operations include user input and program output; comparing the recorded system calls and interactive operations with known malicious behavior patterns, identifying abnormal system calls and interactive operations, and generating an analysis report to describe the detected vulnerability attacks and potential threats.

[0018] As a preferred solution of the malicious code intelligent detection method based on big data mining and analysis described in the present invention, the automated processing includes: performing automated processing according to the analysis results of the sandbox, and when a suspicious program shows signs of deleting system files, tampering with system configurations, frequent network attacks, and rapid self-propagation in the sandbox, it is judged as a strong malicious program, and the executable files and related files of the strong malicious program in the system are immediately deleted, and the residual data and configurations left by the malicious program are cleaned up, and a cleanup report is generated and sent to the staff. If the strong malicious program has been run in the system, a first-level alarm is issued to remind the staff to handle it immediately and give an analysis report; when a suspicious program shows signs of affecting system processes, relying on key system components, or being unable to be completely confirmed in the sandbox, a first-level alarm is issued to remind the staff to handle it immediately and give an analysis report. When malicious behavior is identified, it is judged as a malicious program, and the malicious program and its related processes in the system are isolated in a safe environment. A secondary alarm is sent to the staff, with a detailed behavior analysis report and current processing status, and detailed information about the isolation and alarm is recorded; when the suspicious program in the sandbox has no similarities with the known malicious behavior patterns, the suspicious program is detected again through the detection model. If it passes the detection, it is judged as a non-threatening program. If it still fails the detection, the suspicious program continues to be executed in the sandbox, more behavior data is collected, and a report is generated to dispatch staff for manual analysis. If the manual analysis result is a malicious program, it is recorded and fed back to the known malicious behavior pattern, and a suspicious program reminder is set in the system, and an alarm is issued and the staff is questioned before running.

[0019] In the second aspect, the present invention also provides an intelligent malicious code detection system based on big data mining and analysis, including: a detection module, which collects multi-source data of the target system and performs preprocessing, and monitors the system in real time through a hybrid detection model constructed by a temporal convolutional network and a graph neural network; an analysis module, which isolates the suspicious programs judged by the detection model using a lightweight sandbox, monitors and records all system calls and interactive operations of the suspicious programs, compares them with known malicious behavior patterns and generates an analysis report; a processing module, which automatically processes the suspicious programs according to the analysis report generated by the sandbox simulation, optimizes the detection model according to the processing result feedback, and realizes intelligent malicious code detection.

[0020] In a third aspect, the present invention further provides a computing device, comprising: a memory and a processor;

[0021] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the malicious code intelligent detection method based on big data mining and analysis are implemented.

[0022] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the intelligent malicious code detection method based on big data mining and analysis.

[0023] Beneficial effects of the present invention: The method of the present invention achieves real-time monitoring and rapid response to malicious code by establishing a hybrid detection model based on the extracted features. A lightweight sandbox is used to isolate suspicious programs, monitor and record all system calls and interactive operations of suspicious programs, and effectively reduce the false alarm rate. By analyzing the behavior of suspicious programs in the sandbox and generating alarm information, automatic processing is performed based on the alarm information, and the detection model is optimized based on the feedback of the processing results, intelligent detection of malicious code is achieved. Comprehensive monitoring and accurate identification of malicious code is achieved, detection efficiency and accuracy are improved, false alarm rate is reduced, and network security is effectively protected. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0025] Figure 1 An overall flow chart of a malicious code intelligent detection method based on big data mining and analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0027] Example 1

[0028] Reference Figure 1 , is an embodiment of the present invention, and provides a malicious code intelligent detection method based on big data mining analysis, including:

[0029] S1: Collect and preprocess multi-source data of the target system, extract the behavioral and structural features of the multi-source data, and establish a detection model based on the extracted features.

[0030] Collect data from various data sources such as network traffic, system logs, file operation records, etc., and use big data platforms (such as Hadoop and Spark) to store and manage data.

[0031] Multi-source data includes obtaining network traffic data through network traffic monitoring tools, obtaining file operation records and process, thread information and API call records through system monitoring, obtaining system log data from operating system and application logs, and collecting known malicious code samples from public malicious code libraries.

[0032] Preprocess the collected multi-source data, clean the original data, remove noise and useless information, and normalize the data to ensure that data from different sources can be processed and analyzed consistently, extract useful information from the original data, and remove irrelevant noise and erroneous data.

[0033] Furthermore, the extraction of behavioral and structural features is the core step in building a malicious code detection model, which can help the model identify malicious behaviors and code patterns.

[0034] Behavioral features include network behavior, system calls, file operations, user behavior, process and thread behavior, among which: network behavior includes source IP, destination IP, port number, protocol type, traffic pattern, etc. Detect malicious traffic and abnormal communication behavior; system calls include the number, type, parameters, etc. of system API calls. Identify the behavior patterns of malicious programs; file operations include file creation, deletion, modification, and reading operations. Detect abnormal file operation behavior; process and thread behavior includes process and thread startup, termination, CPU usage, memory usage, etc. Identify abnormal process and thread activities; user behavior includes user login frequency, file access habits, application usage, etc. Identify abnormal user behavior.

[0035] Structural features include code structure, function call graph and module dependency, where: code structure includes control flow graph (CFG), data flow graph (DFG), etc., to identify the structural dependency of the code; function call graph includes the frequency, sequence, hierarchy, etc. of function calls. Detect abnormal function call patterns; module dependency includes the dependency between code modules. Identify abnormal dependencies between modules.

[0036] Furthermore, a hybrid model is constructed for malicious code detection by using a convolutional neural network as a deep learning model and combining it with a graph neural network (GNN). The convolutional neural network is used to process the behavioral features in program samples, and the convolution layer and pooling layer can extract the spatial and temporal relationships in the features; the graph neural network is used to analyze the structural features of program samples and capture the relationship between nodes (such as functions, basic blocks) and edges (such as calls, data flows).

[0037] When analyzing the behavioral characteristics of a program, considering that the behavioral patterns of the program have a strong temporal relationship, an LSTM layer is added to capture long-distance dependencies and avoid gradient vanishing and explosion problems. However, if an LSTM layer is directly added to a CNN, the feature map output by the convolutional layer is usually three-dimensional (batch size, feature map height, feature map width), and the input expected by the LSTM is one-dimensional time series data. It is necessary to deform the data to adapt it to the format of the LSTM. In the deformation process, it is inevitable to encounter problems such as data loss that affect prediction accuracy. Therefore, consider using a temporal convolutional network (TCN) instead of a graph neural network (GNN). The output of the TCN and the expected input of the LSTM layer are both time series data, and the layers can be directly connected. The structure of the neural network is expressed as:

[0038] Input layer, inputs the behavior characteristics of the target program;

[0039] TCN layer, uses TCN to extract temporal features, captures long-distance dependencies by dilating the convolution layer, introduces multi-scale convolution kernels, and weights the convolutional features through the attention mechanism;

[0040] The feature aggregation layer aggregates important features of the output of the TCN layer across time steps to enhance the expressiveness of the features, so that the subsequent LSTM layer can better capture the temporal relationship, expressed as:

[0041]

[0042] LSTM layer, composed of a single LSTM layer through a gating mechanism and memory units, extracts H agg The timing relationship in which H agg represents the output of the feature aggregation layer; T represents the length of the sequence; H TCN Represents the output of the TCN layer; [:, t, :] represents the selection of all filter outputs at the t-th time step in all batches.

[0043] It should be noted that the LSTM layer here is mainly used to assist in capturing timing features and improve feature expression capabilities. The data processed is also the timing features extracted by the TCN layer. Therefore, only a single LSTM layer is required. There is no need to stack multiple LSTM layers like the overall LSTM network to form a deep network to capture complex timing relationships and long-distance dependencies.

[0044] The main purpose of the feature aggregation layer is to aggregate important features across time steps, enhance the expressiveness of features, and enable the subsequent LSTM layer to better capture temporal relationships. In malicious code detection, features at different time steps may have different importance. By aggregating features across time steps, the weights of important features can be increased.

[0045] Furthermore, the analysis of structural features is achieved through graph neural networks. Malware code usually contains complex calling relationships between multiple functions and modules, and different parts of the malicious code may depend on each other. Through these dependencies, its behavior patterns can be identified. Through hierarchical graph convolution, graph structural features are extracted layer by layer to better capture the multi-level dependencies in the malicious code. The structure of the graph neural network is represented as follows:

[0046] The input layer converts the structural features of the target program into a node feature matrix X and an adjacency matrix A. The node feature matrix X contains the feature information of each node and has a shape of (N, F), where N is the number of nodes, representing the basic unit in the program, and F represents the number of features of each node. The adjacency matrix A represents the connection relationship between nodes and has a shape of (N, N).

[0047] The hierarchical graph convolution layer consists of a local convolution layer, a middle convolution layer, and a global convolution layer. The local convolution layer initially extracts node features, captures the relationship between directly adjacent nodes, uses the adjacency matrix A to perform weighted summation on the node feature matrix X, and then performs a linear transformation through the weight matrix to obtain the output H of the local convolution layer. 1 .

[0048] The middle convolutional layer uses the adjacency matrix A to convolutionally transform the output H of the local convolutional layer. 1 Perform weighted summation to capture the multi-hop neighbor relationship of nodes and extract more complex structural features. The middle convolutional layer outputs H 2 .

[0049] The local convolution layer and the middle convolution layer are in a progressive relationship. The middle convolution layer only accepts and processes the output of the local convolution layer, and uploads the processed output result to the global convolution layer.

[0050] The global convolutional layer captures the global features of the entire graph structure, extracts global dependencies, and uses the adjacency matrix A to convolutionally transform the output H of the middle convolutional layer. 1 Perform weighted summation and skip connection of local convolutional layers, using the deformation of the adjacency matrix A 2 The output H of the local convolutional layer 1 Perform weighted summation to capture the dependencies between distant nodes, and obtain the output H of the global convolutional layer through weighted summation 3 .

[0051] It should be noted that the behavior characteristics of malicious code may span multiple calls or operations. Therefore, when the global convolution layer processes the output of the local convolution layer, the initial adjacency matrix A is no longer used, but is transformed into A 2 It captures dependencies between distant nodes, combines the output of the local convolutional layer with the output of the global convolutional layer through jump connections, and captures both the local details and the global structure of the program, improving the completeness and accuracy of feature representation, which helps to more accurately identify malicious behavior.

[0052] Furthermore, the detection model is a hybrid model constructed by a time series convolutional network and a graph neural network. The data in the form of a two-dimensional matrix output by the time series convolutional network is incompatible with the graph structure data format output by the graph neural network. By splicing and fusing the features extracted by the time series convolutional network and the graph neural network, the fused features are classified using a fully connected layer, and the detection results are output, which are expressed as:

[0053] Hfused=concat(Hagg,H 3 )

[0054] Y=σ(WHfused+b)

[0055] Among them, H fused represents the concatenated features; Y represents the judgment result of the detection model; σ represents the Softmax activation function; W represents the weight; and b represents the bias.

[0056] By fusing the behavioral features extracted by TCN and the structural features extracted by GNN, the model can capture the features of the program more comprehensively and improve the accuracy of malicious code detection. After feature fusion, the relationship between different features is comprehensively considered to enhance the feature expression ability, which helps to identify malicious behavior more accurately.

[0057] S2: Use the detection model to detect in real time. When a suspicious program is detected, use a lightweight sandbox to isolate it and monitor and record all system calls and interactive operations of the suspicious program.

[0058] Furthermore, lightweight sandbox technology is a technology that runs suspicious programs in an isolated environment to monitor and analyze the behavior of programs. By executing suspicious programs in an isolated environment, it can avoid affecting the host system and record the system calls and interactive operations of the program in detail.

[0059] Set up an isolated sandbox environment to ensure complete isolation from the host system, configure the sandbox environment to simulate common operating systems and software environments, simulate the target system and application software environment in the isolated sandbox environment, ensure that suspicious programs can exhibit real behavior during execution, load suspicious programs identified by the detection model into the sandbox and start executing them.

[0060] During the sandbox execution process, all system calls and interactive operations of suspicious programs are monitored and recorded. System calls include file operations, network communications, process management and memory operations, and interactive operations include user input and program output.

[0061] Compare recorded system calls and interactions with known malicious behavior patterns, identify abnormal system calls and interactions, and generate analysis reports describing detected vulnerability attacks and potential threats.

[0062] Through lightweight sandbox technology, suspicious programs can be safely run and analyzed in an isolated environment, and all their system calls and interactive operations can be recorded. Combined with behavioral analysis, abnormal behaviors related to known and unknown vulnerabilities can be detected and detailed analysis reports can be generated.

[0063] S3: Analyze the behavior of suspicious programs in the sandbox and generate warning information, perform automatic processing based on the warning information, optimize the detection model based on the processing result feedback, and realize intelligent detection of malicious code.

[0064] Furthermore, automated processing is performed based on the analysis results of the sandbox. When a suspicious program shows signs of deleting system files, tampering with system configurations, frequent network attacks, and rapidly spreading itself in the sandbox, it is judged to be a strong malicious program. The executable files and related files of the strong malicious program in the system are immediately deleted, and the residual data and configurations left by the malicious program are cleaned up. A cleanup report is generated and sent to the staff. If the strong malicious program has already run in the system, a level one alarm is issued to remind the staff to immediately handle it and give an analysis report.

[0065] When a suspicious program shows signs of affecting system processes, relying on key system components, or cannot be fully confirmed as malicious in the sandbox, it is judged as a malicious program, and the malicious program and its related processes in the system are isolated in a safe environment. A secondary alarm is sent to the staff, with a detailed behavior analysis report and current processing status, and detailed information on the isolation and alarm is recorded.

[0066] When a suspicious program shows no similarities with known malicious behavior patterns in the sandbox, the detection model is used to detect the suspicious program again. If it passes the detection, it is judged as a non-threatening program. If it still fails the detection, the suspicious program continues to be executed in the sandbox, more behavioral data is collected, and a report is generated to schedule staff for manual analysis. If the manual analysis result is a malicious program, it is recorded and fed back into the known malicious behavior pattern. A suspicious program reminder is set in the system, and an alarm is issued and staff are questioned before running.

[0067] This embodiment also provides a malicious code intelligent detection system based on big data mining and analysis, including: a detection module, which collects multi-source data of the target system and performs preprocessing, and monitors the system in real time through a hybrid detection model constructed by a temporal convolutional network and a graph neural network; an analysis module, which isolates suspicious programs judged by the detection model using a lightweight sandbox, monitors and records all system calls and interactive operations of suspicious programs, compares them with known malicious behavior patterns and generates an analysis report; a processing module, which automatically processes suspicious programs according to the analysis report generated by the sandbox simulation, optimizes the detection model according to the processing result feedback, and realizes intelligent detection of malicious code.

[0068] This embodiment also provides a computing device, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the intelligent detection method of malicious code based on big data mining and analysis as proposed in the above embodiment.

[0069] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the malicious code intelligent detection method based on big data mining and analysis proposed in the above embodiment is implemented.

[0070] The storage medium proposed in this embodiment and the intelligent detection method for malicious code based on big data mining and analysis proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0071] Through the above description of the implementation method, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccessMemory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0072] Example 2

[0073] The following is an embodiment of the present invention, which provides a malicious code intelligent detection method based on big data mining and analysis. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0074] The experimental environment is configured as follows: Hadoop and Spark are used as data collection and storage platforms, the experimental operation platform is a desktop computer with 3.40GHz Intel(R)Xeon(R)Gold 5318Y CPU, NVIDIA GeForce GTX 4090GPU, and 128GB RAM, the programming language used in the experiment is Python, and the deep learning frameworks used are TensorFlow and PyTorch.

[0075] In terms of data acquisition, the source IP, destination IP, port number, protocol type, traffic pattern, etc. are obtained through the Wireshark monitoring tool, system event records are extracted from Windows EventViewer and Linux system logs, and file creation, deletion, modification, reading and other operation records are obtained through system API calls. Known malicious code samples are collected from public malicious code libraries (such as VirusShare and MalwareBazaar).

[0076] The sample data is divided into five groups, and the trained detection model is used to extract behavioral features and structural features. The temporal convolutional network (TCN) and long short-term memory network (LSTM) are used to extract behavioral features. The spatial and temporal relationships in the features are extracted and combined with the temporal features to enhance the expressiveness of the features. The graph neural network (GNN) is used to analyze the structural features of the program samples to capture the relationship between nodes (such as functions, basic blocks) and edges (such as calls, data flows). Some of the data are shown in the following table. Label 1 represents a suspicious program and 0 represents a safe program.

[0077] Table 1 Experimental data table

[0078]

[0079] Simulate the target system and application software environment in an isolated environment to ensure that suspicious programs exhibit real behaviors during execution. Use lightweight sandbox technology (such as Cuckoo Sandbox) to isolate suspicious programs and record all their system calls and interactive operations in detail. The suspicious program operations over a period of time are shown in the following table.

[0080] Table 2 Sandbox record table

[0081] Sample No. Number of file operations Number of process starts Number of thread starts Network communication times Sample 2 12 5 8 15 Sample 3 15 6 9 18 Sample 5 14 7 10 20

[0082] The experimental data table shows that the behavioral and structural characteristics of test objects 2, 3, and 5 are significantly higher than those of other objects. Through real-time detection, the model successfully identified and isolated the object, avoiding its potential threat to the system. In the sandbox environment, the behavioral and timing characteristics of test objects 2, 3, and 5 are abnormal, further verifying their malicious behavior. The combination of real-time detection and sandbox isolation effectively improves system security.

[0083] It can be seen that the hybrid model combining TCN and GNN can capture both behavioral and structural features. Through feature aggregation and fusion, the model enhances the feature expression capability at different time steps and structural levels.

[0084] Through comparative analysis of experimental data, compared with the existing technology, the present invention achieves higher detection accuracy and response speed through real-time detection, lightweight sandbox isolation, comprehensive behavior recording and analysis, multi-level feature extraction and fusion, dynamic update and model optimization, etc., providing a strong guarantee for system security.

[0085] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A malicious code intelligent detection method based on big data mining and analysis, characterized in that: include: Collect and preprocess multi-source data of the target system, extract behavioral and structural features of the multi-source data, and establish a detection model based on the extracted features; Use detection models to detect in real time. When suspicious programs are detected, use lightweight sandbox isolation to monitor and record all system calls and interactive operations of suspicious programs. Analyze the behavior of suspicious programs in the sandbox and generate warning information, perform automatic processing based on the warning information, optimize the detection model based on the processing result feedback, and realize intelligent detection of malicious code.

2. The method for intelligent detection of malicious code based on big data mining analysis as claimed in claim 1, characterized in that: The multi-source data includes obtaining network traffic data through network traffic monitoring tools, obtaining file operation records and process, thread information and API call records through system monitoring, obtaining system log data from operating system and application logs, and collecting known malicious code samples from public malicious code libraries.

3. The method for intelligent detection of malicious code based on big data mining analysis as claimed in claim 2, characterized in that: The detection model includes behavioral feature analysis and structural feature analysis of the target program. In the behavioral feature analysis, considering the temporal relationship of the program's behavior pattern, a long short-term memory network layer is added to the neural network structure to enhance the ability to capture the temporal pattern in the system call sequence. The main body of the neural network selects a temporal convolutional network, which is expressed as: Input layer, inputs the behavior characteristics of the target program; TCN layer, uses TCN to extract temporal features, captures long-distance dependencies by dilating the convolution layer, introduces multi-scale convolution kernels, and weights the convolutional features through the attention mechanism; The feature aggregation layer aggregates important features of the output of the TCN layer across time steps to enhance the expressiveness of the features, so that the subsequent LSTM layer can better capture the temporal relationship, expressed as: LSTM layer, composed of a single LSTM layer through a gating mechanism and memory units, extracts H agg The timing relationship in which H agg represents the output of the feature aggregation layer; T represents the length of the sequence; H TCN Represents the output of the TCN layer; :,t,: represents the selection of all filter outputs at the tth time step in all batches.

4. The method for intelligent detection of malicious code based on big data mining analysis as claimed in claim 3, characterized in that: The detection model also includes analyzing the structural features of program samples through graph neural networks. Malicious codes usually have complex call relationships and dependency structures. Through hierarchical graph convolution, graph structural features are extracted layer by layer. The structure of the graph neural network is represented as follows: Input layer, converts the structural features of the target program into node feature matrix X and adjacency matrix A. The node feature matrix X contains the feature information of each node and has a shape of (N, F), where N is the number of nodes, representing the basic unit in the program, and F represents the number of features of each node. The adjacency matrix A represents the connection relationship between nodes and has a shape of (N, N); The hierarchical graph convolution layer consists of a local convolution layer, a middle convolution layer, and a global convolution layer. The local convolution layer initially extracts node features, captures the relationship between directly adjacent nodes, uses the adjacency matrix A to perform weighted summation on the node feature matrix X, and then performs a linear transformation through the weight matrix to obtain the output H of the local convolution layer. 1 ; The middle convolutional layer uses the adjacency matrix A to convolutionally transform the output H of the local convolutional layer. 1 Perform weighted summation to capture the multi-hop neighbor relationship of nodes and extract more complex structural features. The middle convolutional layer outputs H 2 ; The global convolutional layer captures the global features of the entire graph structure, extracts global dependencies, and uses the adjacency matrix A to convolutionally transform the output H of the middle convolutional layer. 1 Perform weighted summation and skip connection of local convolutional layers, using the deformation of the adjacency matrix A 2 The output H of the local convolutional layer 1 Perform weighted summation to capture the dependencies between distant nodes, and obtain the output H of the global convolutional layer through weighted summation 3 .

5. The method for intelligent detection of malicious code based on big data mining and analysis as claimed in claim 4 is characterized by: The detection model also includes a hybrid model constructed by a time series convolutional network and a graph neural network. The data in the form of a two-dimensional matrix output by the time series convolutional network is incompatible with the graph structure data format output by the graph neural network. The features extracted by the time series convolutional network and the graph neural network are spliced ​​and fused, and the fused features are classified using a fully connected layer, and the detection result is output, which is expressed as: H fused =concat(H agg ,H 3 ) Y=σ(WH fused +b) Among them, H fused represents the concatenated features; Y represents the judgment result of the detection model; σ represents the Softmax activation function; W represents the weight; and b represents the bias.

6. The method for intelligent detection of malicious code based on big data mining analysis as claimed in claim 5, characterized in that: The lightweight sandbox isolation includes simulating the target system and application software environment in an isolated sandbox environment to ensure that the suspicious program can show real behavior during execution, and loading the suspicious program judged by the detection model into the sandbox and starting execution; During the sandbox execution process, all system calls and interactive operations of suspicious programs are monitored and recorded. System calls include file operations, network communications, process management and memory operations, and interactive operations include user input and program output. Compare recorded system calls and interactions with known malicious behavior patterns, identify abnormal system calls and interactions, and generate analysis reports describing detected vulnerability attacks and potential threats.

7. The method for intelligent detection of malicious code based on big data mining analysis as claimed in claim 6, characterized in that: The automated processing includes: performing automated processing according to the analysis results of the sandbox; when a suspicious program shows signs of deleting system files, tampering with system configurations, frequent network attacks, and rapid self-propagation in the sandbox, it is judged as a strong malicious program, and the executable files and related files of the strong malicious program in the system are immediately deleted, and the residual data and configurations left by the malicious program are cleaned up, and a cleanup report is generated and sent to the staff; if the strong malicious program has been run in the system, a first-level alarm is issued to remind the staff to immediately handle it and give an analysis report; When a suspicious program shows signs of affecting system processes, relying on key system components, or cannot be fully confirmed as malicious in the sandbox, it is judged as a malicious program, and the malicious program and its related processes in the system are isolated in a safe environment. A secondary alarm is sent to the staff, with a detailed behavior analysis report and current processing status, and detailed information on isolation and alarm is recorded; When a suspicious program shows no similarities with known malicious behavior patterns in the sandbox, the detection model is used to detect the suspicious program again. If it passes the detection, it is judged as a non-threatening program. If it still fails the detection, the suspicious program continues to be executed in the sandbox, more behavioral data is collected, and a report is generated to schedule staff for manual analysis. If the manual analysis result is a malicious program, it is recorded and fed back into the known malicious behavior pattern. A suspicious program reminder is set in the system, and an alarm is issued and staff are questioned before running.

8. A malicious code intelligent detection system based on big data mining analysis using the method according to any one of claims 1 to 7, characterized in that: include, The detection module collects and preprocesses multi-source data of the target system, and monitors the system in real time through a hybrid detection model built using a temporal convolutional network and a graph neural network. The analysis module isolates suspicious programs identified by the detection model using a lightweight sandbox, monitors and records all system calls and interactive operations of suspicious programs, compares them with known malicious behavior patterns, and generates analysis reports; The processing module automatically processes suspicious programs according to the analysis report generated by the sandbox simulation, optimizes the detection model based on the processing result feedback, and realizes intelligent detection of malicious code.

9. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Citation Information

Cited By

  • Multilayer certificate verification method and system based on digital certificate

    CN119766517A

  • Equipment access control method and system based on security sandbox

    CN120050075A

  • A secure sandbox-based device admission control method and system

    CN120050075B

  • Data anomaly detection method based on multi-scale feature extraction

    CN120316629A

  • Ransomware automatic acquisition and analysis method and system based on large language model

    CN120567457A