A behavioral gene identification method for network attack organizations

Through the behavioral gene recognition method of network attack organizations, the behavioral gene knowledge graph and large language model are used to identify complex network attacks, which solves the shortcomings of the existing technology in recognition accuracy and dynamic adaptability, and achieves efficient identification and prevention of complex threats .

CN118611983BActive Publication Date: 2024-11-12INFORMATION & COMMNUNICATION BRANCH STATE GRID JIANGXI ELECTRIC POWER CO +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411037261.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-12
Estimated Expiration
2044-07-31

AI Technical Summary

Technical Problem

When facing complex cyber attacks, existing network security technologies have insufficient recognition accuracy and poor dynamic adaptability, making it difficult to fully cover and defend against advanced threats, and have a large delay in threat hunting, so they cannot correct and analyze deviations in time.

Method used

Using the behavioral gene recognition method of cyberattack organizations, the behavioral gene knowledge map is constructed by collecting attack intelligence data, malicious behavior sequences are extracted and homologous recognition models are trained, and advanced behavioral gene information is analyzed in combination with large language models to achieve alignment and matching to identify target dependence patterns. , and construct threat search query prompt words to identify attack behavior.

Benefits of technology

It significantly improves the identification accuracy and coverage of complex network attacks, reduces threat response delay, supports real-time mining of dynamic threat modes, improves the security system's ability to adapt to complex threats, and provides more solid information security guarantees.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118611983B_ABST
    Figure CN118611983B_ABST
Patent Text Reader

Abstract

The present invention provides a method for identifying a behavior gene of a network attack organization, including: collecting basic data including a behavior gene knowledge graph of an attack organization and system process data; extracting process behavior sequence data in the system process data to perform behavior gene homology reasoning to predict a target attack organization; retrieving and organizing high-level behavior gene information of the target attack organization and then converting it into a behavior dependency pattern set, analyzing the behavior logs of known attack events to obtain corresponding behavior dependency instances, aligning and matching the behavior dependency instances with the behavior dependency pattern set to identify the target dependency pattern; constructing threat search query prompts based on the target dependency pattern for interaction with a large language model to identify attack behaviors that are semantically consistent with the behavior gene of the target dependency pattern. The application of this method can deconstruct and identify behavior gene data in real time, can perform multi-dimensional data analysis and high-level behavior gene recognition, and improves recognition accuracy and coverage.
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 behavior gene recognition method for a network attack organization. Background Art

[0002] In the information technology era, national infrastructure, enterprises and individuals are constantly subjected to complex network attacks such as advanced persistent threats (APT). APT attacks are usually launched by highly organized and targeted attack groups, whose behavior is covert and their methods are diverse. On the one hand, there are diverse advanced attack methods, including comprehensive and accurate intelligence collection, social engineering methods, customized malware and the exploitation of zero-day vulnerabilities. On the other hand, attackers usually use camouflage and evasion strategies to enable attack activities to be hidden for a long time in multiple stages, and use system management tools to attack, thereby evading security monitoring. In response to the growing complexity and diversity of APT attacks, many companies and institutions have invested a lot of resources, but still face major challenges.

[0003] In the field of network security, traditional static defense strategies based on signatures and rules have always been the foundation and key components. These strategies mainly include firewalls, intrusion detection systems (IDS) and anti-virus software. However, in the face of increasingly complex and sophisticated attack methods, traditional static defense methods are not flexible and effective enough. To this end, a threat hunting method based on system audit logs has been proposed to provide a more dynamic network security defense strategy. This method can identify potential security threats by collecting and analyzing system activity logs. System activity monitoring tools record and analyze system events including user login attempts, file access, and network communication activities to reveal unusual or suspicious behavior patterns. With the continuous development of technology, some advanced intrusion detection systems based on system audit logs have been improved and applied, significantly improving the detection capabilities of more complex and covert attack behaviors.

[0004] Open Source Cyber ​​Threat Intelligence (OSCTI) describes the process of collecting, analyzing, and distributing information to identify, monitor, and predict cyber threats, providing evidence-based threat knowledge, indicating its great potential in making up for the shortcomings of existing defense strategies. Many researchers have begun to conduct a lot of research and applications around the data collection and integration of open source cyber threat intelligence. In addition, many threat hunting methods based on system audit logs, such as ThreatRaptor, enhance the threat hunting capabilities of system audit logs with attack behavior examples in OSCTI.

[0005] Through the study of existing technologies, it is found that the existing network security technologies have insufficient insight into complex attack strategies, and the delay in threat hunting is large and the dynamic adaptability is poor, which is specifically reflected in the following aspects: Limitation of comprehensive identification ability: Traditional security threat identification models generally lack the ability to fully deconstruct the attack behavior of attack organizations. This one-sided cognition leads to insufficient identification accuracy and inability to deeply understand the complexity and variability of attack patterns, so it is difficult to fully cover and defend against advanced persistent threats in practice. Insufficient technology integration and data mining: The current security analysis framework has obvious defects in multi-technology integration. A single algorithm or model is difficult to process heterogeneous data from multiple sources and accurately extract the multi-level implicit features of attack behaviors from them. The isolation of this technology limits the depth and breadth of understanding of attack behaviors. Challenges of real-time and dynamic response: Existing methods often face large delays when extracting and analyzing process system behavior data, which is not conducive to rapid response and threat hunting. Moreover, the lack of effective dynamic learning and real-time feedback mechanisms leads to the inability to timely correct deviations in the threat analysis process. Limitations of threat hunting capabilities: Existing threat databases are often static and rigid, and lack the ability to mine threat patterns in real time and dynamically, and cannot effectively adapt to the complex and changing security threat environment.

[0006] Therefore, it is necessary to provide a network attack behavior identification method that can solve the defects of existing network security technologies in attack behavior identification and defense. Summary of the invention

[0007] The purpose of the present invention is to provide a behavioral gene identification method for a network attack organization, so as to improve the problem that the existing network security technology has insufficient insight into complex attack strategies.

[0008] In a first aspect, the present invention provides a method for identifying a network attack organization with behavioral genes, including: collecting basic data, the basic data including an attack organization behavioral gene knowledge graph and system process data obtained by capturing attack intelligence data for threat knowledge extraction and updating; extracting malicious behavior sequences from the attack organization behavioral gene knowledge graph to construct and train a primary behavioral gene homology recognition model, extracting process behavior sequence data from the system process data and inputting the primary behavioral gene homology recognition model to perform behavioral gene homology reasoning on the corresponding process behavior sequence data, thereby predicting a target attack organization that matches the characteristics of the malicious behavior sequence; using the attack organization behavioral gene knowledge graph to retrieve and organize high-level behavioral gene information of the target attack organization, applying a large language model to convert the high-level behavioral gene information into a set of behavioral dependency patterns, analyzing the behavior logs of known attack events in the system process data to obtain corresponding behavioral dependency instances, using a large language model to align and match the behavioral dependency instances with the set of behavioral dependency patterns to obtain an alignment matching result, and comparing the alignment matching results to identify the target dependency pattern; constructing a threat search query prompt word according to the target dependency pattern, and using the threat search query prompt word to interact with the large language model to identify attack behaviors that are semantically consistent with the target dependency pattern behavioral genes.

[0009] The beneficial effect of the behavioral gene identification method of the network attack organization provided by the present invention is that it adopts multi-dimensional data analysis and advanced behavioral gene identification technology to significantly improve the identification accuracy and coverage of complex network attacks. By utilizing advanced machine learning algorithms and natural language processing technology, it can effectively integrate and deeply analyze the implicit features in heterogeneous data from multiple sources. Through precise deviation correction and timely feedback mechanism, the delay of threat response is greatly reduced, ensuring a rapid and accurate response to emerging threats. Combined with the design of dynamic database and retrieval grammar, the present invention supports real-time mining of dynamic threat patterns and enhances the ability of security systems to adapt to increasingly complex security threats. The method combines the latest research results in multiple sub-fields such as artificial intelligence, big data analysis, machine learning and network behavior analysis, and provides a new technical path for modern network security threat identification and defense. By implementing the technical solution of the present invention, deep insight and effective prevention of complex attack strategies can be achieved in the field of network security, providing a more solid technical guarantee for information security.

[0010] In a possible embodiment, malicious behavior sequences are extracted from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, including: extracting malicious behavior sequences from the attack organization behavior gene knowledge graph; performing feature engineering on the malicious behavior sequences and dividing the results into a training set and a validation set according to a preset ratio, and inputting the results into a random forest classification model for training and validation, so as to obtain a primary behavior gene homology recognition model for predicting a target attack organization that matches the features of the malicious behavior sequence.

[0011] In another possible embodiment, a large language model is used to convert high-level behavioral gene information into a set of behavioral dependency patterns, including: constructing high-level behavioral gene projection prompt words, which are used to guide the large language model to understand and convert high-level behavioral gene information; and using high-level behavioral gene projection prompt words to interact with the large language model to obtain a set of behavioral dependency patterns derived from the high-level behavioral gene information.

[0012] In other possible embodiments, a large language model is used to align and match the behavior dependency instance with the behavior dependency pattern set to obtain an alignment matching result, including: constructing a constraint feedback prompt word, the constraint feedback prompt word is used to guide the large language model to evaluate the matching degree between the behavior dependency instance and the behavior dependency pattern set from a preset perspective; using the constraint feedback prompt word to interact with the large language model to obtain the alignment matching result of the behavior dependency instance and the behavior dependency pattern set.

[0013] The alignment matching result includes the matching degree between the behavior dependency instance and the behavior dependency pattern set; comparing the alignment matching results to identify the target dependency pattern includes: comparing the matching degree to identify the target dependency pattern with the highest matching degree with the behavior dependency instance.

[0014] Constructing threat search query prompt words according to the target dependency pattern, including: the threat search query prompt words include the target dependency pattern and a preset system database query statement template; the system database query statement template combines a structured query language or a time series database query language.

[0015] In a second aspect, the present invention further provides a behavioral gene identification device for a network attack organization, comprising:

[0016] A data collection unit is used to collect basic data, including attack organization behavior gene knowledge graph and system process data obtained by capturing attack intelligence data to extract threat knowledge and updating;

[0017] The behavior gene homology recognition unit is used to extract malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, extract process behavior sequence data from the system process data and input it into the primary behavior gene homology recognition model to perform behavior gene homology reasoning on the corresponding process behavior sequence data, thereby predicting a target attack organization that matches the characteristics of the malicious behavior sequence;

[0018] A behavior gene information alignment unit is used to retrieve and organize high-level behavior gene information of a target attack organization using an attack organization behavior gene knowledge graph, convert the high-level behavior gene information into a behavior dependency pattern set using a large language model, analyze the behavior logs of known attack events in system process data to obtain corresponding behavior dependency instances, align and match the behavior dependency instances with the behavior dependency pattern set using a large language model to obtain an alignment matching result, and compare the alignment matching results to identify the target dependency pattern;

[0019] The threat hunting unit is used to construct threat hunting query prompts according to the target dependency pattern, and use the threat hunting query prompts to interact with the large language model to identify attack behaviors that are consistent with the target dependency pattern behavior gene semantics.

[0020] Among them, the behavior gene homology recognition unit extracts malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, including: extracting malicious behavior sequences from the attack organization behavior gene knowledge graph; performing feature engineering on the malicious behavior sequences and dividing the results into training sets and validation sets according to a preset ratio, and inputting them into a random forest classification model for training and validation, so as to obtain a primary behavior gene homology recognition model for predicting the target attack organization that matches the characteristics of the malicious behavior sequence.

[0021] The behavioral gene information alignment unit applies a large language model to convert high-level behavioral gene information into a set of behavioral dependency patterns, including: constructing high-level behavioral gene projection prompt words, which are used to guide the large language model to understand and transform high-level behavioral gene information; and using high-level behavioral gene projection prompt words to interact with the large language model to obtain a set of behavioral dependency patterns derived from the high-level behavioral gene information.

[0022] The behavior gene information alignment unit uses a large language model to align and match the behavior dependency instance with the behavior dependency pattern set to obtain an alignment matching result, including: constructing a constraint feedback prompt word, the constraint feedback prompt word is used to guide the large language model to evaluate the matching degree of the behavior dependency instance with the behavior dependency pattern set from a preset perspective; using the constraint feedback prompt word to interact with the large language model to obtain the alignment matching result of the behavior dependency instance and the behavior dependency pattern set.

[0023] The alignment matching result includes the matching degree between the behavior dependency instance and the behavior dependency pattern set; the behavior gene information alignment unit compares the alignment matching results to identify the target dependency pattern, including: comparing the matching degree to identify the target dependency pattern with the highest matching degree with the behavior dependency instance.

[0024] The threat search unit constructs a threat search query prompt word according to the target dependency pattern, including: the threat search query prompt word includes the target dependency pattern and a preset system database query statement template; the system database query statement template combines a structured query language or a time series database query language.

[0025] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the above-mentioned behavioral gene identification method for a network attack organization is implemented.

[0026] In a fourth aspect, the present invention also provides an electronic device, comprising: a processor and a memory; the memory is used to store a computer program; the processor is used to execute the computer program stored in the memory, so that the electronic device executes the above-mentioned behavioral gene identification method of the network attack organization.

[0027] For the beneficial effects of the second to fourth aspects, reference may be made to the description of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart of a method for identifying behavioral genes of a network attack organization provided by an embodiment of the present invention;

[0029] Figure 2 A schematic diagram of the implementation scheme of the behavior gene identification method for a network attack organization provided by an embodiment of the present invention;

[0030] Figure 3 A schematic diagram of the hierarchical structure of network attack organizational behavior genes provided by an embodiment of the present invention;

[0031] Figure 4 A schematic diagram of a behavioral gene identification device for a network attack organization provided by an embodiment of the present invention;

[0032] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0034] To facilitate understanding, the relevant concepts of some technical terms involved in this application are first introduced below.

[0035] Aggressive organizational behavior gene

[0036] The attack organization behavior gene refers to the attack organization's behavioral preference. It is a logical abstraction that includes the attack organization's usual attack strategies, technical means, and attack processes, and includes the system-level malicious traces left by the attack organization in the victim system during the actual attack implementation.

[0037] Specifically, in order to comprehensively analyze the behavioral characteristics of the attacking organization and capture potential threats, the present invention divides the attacking organization's behavioral genes into two levels: advanced behavioral genes and primary behavioral genes.

[0038] Advanced behavioral genes

[0039] Advanced behavioral genes are the attack patterns adopted by the attacking organization. In the MITRE AT&CK framework, the attack patterns are defined as TTPs (Tactics, Techniques and Procedures).

[0040] Primary behavior genes

[0041] Primary behavior genes are the attack traces left by the attack organization when performing malicious activities in the victim system. In this solution, primary behavior genes are defined as the threat characteristics contained in the behavior instances composed of a series of IoCs (Indicators of Compromises). The primary behavior threat characteristics can be identified and detected from the behavior instances in two data forms. On the one hand, the sequence numerical pattern shown by the behavior instance, such as the activity of a process P in a short period of time is monitored as the sequence pattern (Files operation: Files Creation:3 / Files Delection:1; Process Operation: Executable Launched: 3 / code injection: 2; Registry Key Operations: Registry Key Modified: 2; Mutex Operations: Mutex Creation: 2; DNSQueries: 2; HTTP Sessions: 2). Through the comprehensive analysis of the above dynamic characteristics, a series of obvious malicious behavior indicators can be determined, including multiple reading and modification of key system files, launching unauthorized executable files, tampering with system processes and registries, interfering with system security services, and communicating with suspicious network entities. This behavior sequence suggests a possible attack scenario. On the other hand, in the behavior instance, the interdependence between multiple IoC entities constitutes the behavior dependency data, such as "An unknown process Process_X first opens the system core file system32.dll for reading, and then creates and writes a file named temp_payload.exe to the temporary directory. Process_X then modifies the registry HKEY_LOCAL_MACHINE\Software\Microsoft\Windows\CurrentVersion\Run, adding a new entry to make temp_payload.exe run automatically when the system starts. Process_X creates a global mutex to ensure that only one instance of temp_payload.exe is running at any time." The behavior dependency composed of the above multiple IoC entities (Process_X, temp_payload.exe, etc.) provides an empirical data type with time dependency information and entity type information, describing a process that opens key system files, creates temporary files, and modifies the registry to ensure its own persistence. However, the analysis of behavior dependency data requires processing massive system audit information, which consumes a lot of time and is difficult to execute in real time.Since behavior dependency data analysis requires a lot of time to process system audit information, this solution divides the analysis process into two parts in order to maximize the mining of primary behavior gene features and optimize the analysis latency: lightweight behavior sequence data that is easy to obtain through real-time auditing, which is quickly analyzed through efficient machine learning models; and behavior dependency data that needs to be further processed to provide support for the constraint feedback analysis process of advanced behavior genes.

[0042] In view of the problems existing in the prior art, an embodiment of the present invention provides a method for identifying behavioral genes of a network attack organization.

[0043] This embodiment provides a method for identifying behavioral genes of a network attack organization. Figure 1 and Figure 2 , the method comprising:

[0044] S101: Collect basic data, which includes capturing attack intelligence data to extract threat knowledge and updating the attack organization behavior gene knowledge graph and system process data.

[0045] In S101, in a possible embodiment, the attack intelligence data is captured to extract threat knowledge and update to obtain the attack organization behavior gene knowledge graph, including: using automated tools to periodically capture attack intelligence from multiple open source network platforms, using natural language processing (NLP) technology to deepen the extraction of threat knowledge, and synchronously updating the attack organization behavior gene knowledge graph. Among them, the automated tool used to capture attack intelligence can be a web crawler or other tool. The attack organization behavior gene knowledge graph uses graph database technology to store attack patterns and instance behaviors, and integrates multiple open source threat intelligence sources including Alpha, FraunhoferFKIE, Alibaba Cloud Vulnerability Library and VirusTotal to form a semantically rich and structurally consistent high-quality data resource library.

[0046] Collecting system process data includes: collecting original security event information including system logs, network traffic and application behavior logs through monitoring tools and sensor technology. The collected system process data is cleaned and formatted in the pre-processing stage, and then imported into the structured data lake, which can lay a solid foundation for further in-depth analysis of the behavioral genes of cyber attack organizations.

[0047] S102: extract malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, extract process behavior sequence data from the system process data and input it into the primary behavior gene homology recognition model to perform behavior gene homology reasoning on the corresponding process behavior sequence data, thereby predicting a target attack organization that matches the characteristics of the malicious behavior sequence.

[0048] In S102, in a possible embodiment, malicious behavior sequences are extracted from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, including: extracting malicious behavior sequences from the attack organization behavior gene knowledge graph; performing feature engineering on the malicious behavior sequences and dividing the results into a training set and a validation set according to a preset ratio, and inputting the results into a random forest classification model for training and validation, so as to obtain a primary behavior gene homology recognition model for predicting a target attack organization that matches the features of the malicious behavior sequence.

[0049] Exemplarily, in this embodiment, the random forest model is used for behavior sequence homology metric analysis. Specifically, the malicious behavior sequence features in the primary behavior gene of the attack organization are learned through the random forest model, so as to construct and train a primary behavior gene homology recognition model, which can be used to infer and classify the real-time system process behavior sequence to be tested input in the subsequent link, so as to obtain the homology probability of the sequence to be tested with respect to each attack organization.

[0050] In this embodiment, the primary behavior gene homology recognition model training process includes: first, extracting malicious behavior sequences in the primary behavior gene data from the attack organization preference profile derived from the attack organization behavior gene knowledge graph.

[0051] For example, a malicious behavior sequence of an attack organization is as follows:

[0052] {"operations": [{"count": 379, "type": "registry_keys_opened"}, {"count": 320, "type": "files_opened"}, {"count": 66, "type": "registry_keys_set"}, {"count": 2, "type": "ip_traffic"}, {"count": 1, "type": "files_deleted"}, {"count": 7, "type": "processes_terminated"}, {"count": 1, "type":"windows_hidden"}, {"count": 6, "type": "files_written"}, {"count": 9, "type": "mutexes_created"}, {"count": 2, "type": "files_dropped"}, {"count":3, "type": "dns_lookups"}, {"count": 5, "type": "registry_keys_deleted"}, {"count": 1, "type": "mutexes_opened"}], "have_behavior": true, "behavior": ["windows_operation", "processes_operation", "mutexes_operation", "networks_operation", "files_operation", "registry_keys_operation"]}

[0053] Then, feature engineering is performed on the behavior sequence according to the degree of harm of the malicious behavior. Specifically, a weighted method is used for processing: the weight is multiplied by the count of the operation (op_count) to calculate the weighted operation: if the operation type (op_type) contains "mutex", the weight is set to 3; if the operation type contains "registry" or "service", the weight is set to 2. The purpose of this step is to optimize the feature representation of the malicious behavior sequence to enhance the model's ability to distinguish the behavioral characteristics of different attack organizations.

[0054] Finally, the data processed by feature engineering is divided into a training set and a validation set according to a preset ratio, and input into a random forest classification model for training and validation. The training goal is to predict the target attack organization that matches the characteristics of the malicious behavior sequence. The ratio of dividing the data processed by feature engineering into a training set and a validation set is set according to the actual situation. In a specific embodiment, the data processed by feature engineering is divided into a training set and a validation set at a ratio of 8:2.

[0055] The process behavior sequence data extracted from the system process data is input into the primary behavior gene homology recognition model obtained after training. It can perform homology reasoning on the pattern representation of the known malicious behavior sequence of the attack organization, and output the attack organization that matches the input process behavior sequence and its matching probability, thereby realizing the recognition of homology sequences. For example, if the behavior sequence of a process is highly correlated with the behavior pattern of a specific attack organization, the model will return the high matching probability corresponding to the attack organization.

[0056] Step S102 provides a detailed threat detection method, which can accurately track the malicious activities performed by malware and camouflage tools used by the attacking organization, and enhance the system's fine-grained perception of threats.

[0057] S103: Use the attack organization behavior gene knowledge graph to retrieve and organize the high-level behavior gene information of the target attack organization, apply the large language model to convert the high-level behavior gene information into a set of behavior dependency patterns, analyze the behavior logs of known attack events in the system process data to obtain the corresponding behavior dependency instances, use the large language model to align and match the behavior dependency instances with the behavior dependency pattern set to obtain the alignment matching results, and compare the alignment matching results to identify the target dependency pattern.

[0058] In S103, in a possible embodiment, the attack organization behavior gene knowledge graph is used to retrieve and organize the behavior preference profile of the attack organization, which contains detailed high-level behavior gene information. Specifically, it covers the TTPs (Tactics, Techniques, and Procedures) IDs defined in the MITRE ATT&CK framework and detailed attack pattern descriptions of other threat intelligence sharing platforms. For example, in order to deeply understand and standardize the attack pattern, the detailed behavior description of specific TTPs is retrieved and summarized on the official website of MITRE, which may include attack steps, tools used, affected systems, and typical indicators.

[0059] The large language model is used to transform high-level behavioral gene information into a set of behavioral dependency patterns, including: constructing high-level behavioral gene projection prompts, which are used to guide the large language model to understand and transform high-level behavioral gene information; using high-level behavioral gene projection prompts to interact with the large language model to obtain a set of behavioral dependency patterns derived from high-level behavioral gene information. In this process, the large language model is guided to identify a specific set of behavioral patterns starting from high-level abstract concepts, so as to facilitate further analysis and identification of potential threats.

[0060] Exemplarily, this embodiment provides a specific example of applying a large language model to convert high-level behavior gene information into a set of behavior dependency patterns: Consider a set of high-level behavior genes of a certain attack organization, which are attack strategies refined and abstracted by TTP: email phishing (TTP ID: T1566), exploiting public vulnerabilities (TTP ID: T1190). In order to guide the large language model to understand and transform the above high-level behavior genes, the following prompt words (Prompt) can be designed: "Based on the following high-level attack pattern: 1) Sending an email containing a malicious link through email phishing, 2) Exploiting a public vulnerability to execute malicious code, describe a possible attack behavior sequence and identify the behavior dependency pattern therein. Consider the specific actions that the attacker may take and the victim's system response, and how to detect signs of attack from these activities."

[0061] In response to the above prompt, the large language model may generate the following behavioral dependency patterns: Email sending stage: The attacker sends an email containing a malicious link to the victim. The dependency patterns at this stage may include: email sending behavior, use of the email server, method of embedding the malicious link, etc. Link clicking stage: The victim clicks on the malicious link in the email. The dependency patterns at this stage may involve: the victim's browser activity, link redirection behavior, user interaction pattern, etc. Malicious code execution stage: The malicious code is executed by exploiting public vulnerabilities on the victim's system. The dependency patterns at this stage may include: vulnerability exploitation technology, loading and execution methods of malicious code, changes in system permissions, circumvention methods of system monitoring tools, etc. System control stage: The attacker gains control of the victim's system. The dependency patterns at this stage may include: installation of backdoors, data theft, communication with the attacker's command and control server (C2), etc.

[0062] Analyze the behavior logs of known attack events in the system process data to obtain the corresponding behavior dependency instances.

[0063] Using a large language model to align and match the behavior dependency instance with the behavior dependency pattern set to obtain an alignment matching result, specifically including: constructing a constraint feedback prompt word, the constraint feedback prompt word is used to guide the large language model to evaluate the matching degree of the behavior dependency instance with the behavior dependency pattern set from a preset perspective; using the constraint feedback prompt word to interact with the large language model to obtain the alignment matching result of the behavior dependency instance with the behavior dependency pattern set. Exemplarily, evaluating the matching degree of the behavior dependency instance with the behavior dependency pattern set from a preset perspective includes: evaluating from the perspectives of behavior temporal consistency, correlation between IoC entities, and logical coherence of attack steps.

[0064] This embodiment also provides a specific example of using a large language model to align and match behavior dependency instances with behavior dependency pattern sets: consider a known network attack scenario, where an attacker uses email phishing to send emails with malicious attachments. After the victim downloads and opens the attachment, the malware is activated, and the attacker then uses the software to perform a series of malicious actions in the victim's system.

[0065] Step 1: Collect system behavior logs

[0066] The security team collects behavioral logs from the affected systems, which record a series of activities such as email reception, attachment download, file opening, malware activation, etc. These activities constitute actual behavioral dependency instances of primary behavioral genes.

[0067] Step 2: Generate behavioral dependency patterns

[0068] A set of candidate behavioral dependency patterns for email phishing attacks has been obtained through large language models (LLMs), which describe the possible behavioral dependencies from email reception to malware activation.

[0069] Step 3: Design Constraint Feedback Prompt

[0070] In order to align the behavior dependency pattern with the actual behavior dependency instance, the following constraint feedback prompt is designed: "Given a series of behavior dependency sequences from email receiving, attachment downloading, file opening to malware activation, evaluate the degree of match of each behavior instance with the known email phishing attack behavior dependency pattern. Consider the consistency of behavior timing, the correlation between IoC entities, and the logical coherence of attack steps."

[0071] Step 4: Alignment of behavior dependency instances with behavior dependency patterns

[0072] In this step, the security team aligns and matches the collected system behavior logs with the candidate set of behavior dependency patterns generated by LLMs. For example: Behavior temporal consistency: Ensure that the behavior sequence recorded in the log is consistent with the candidate behavior dependency pattern in time order, that is, email reception occurs before attachment download, attachment download occurs before file opening, and so on. IoC entity correlation: Check whether each behavior instance recorded in the log involves known indicators (IoC), such as a specific malicious email sender address, known malicious attachment file name, etc., and evaluate the degree of correlation between these IoCs. Logical coherence of attack steps: Evaluate whether the behavior sequence in the log is logically coherent, that is, whether each step of behavior provides a condition or basis for the occurrence of the next step of behavior.

[0073] Implementation example results:

[0074] Through this alignment process, the security team can identify behavior dependency instances that are highly matched with the behavior dependency pattern, thereby confirming the occurrence of the attack and further analyzing the details of the attack. For example, if a behavior dependency fully matches the behavior dependency pattern of an email phishing attack, then it can be determined that the sequence represents an actual attack activity and corresponding countermeasures can be taken accordingly.

[0075] LLMs’ deep semantic understanding capability is used to compare the aligned behavior dependency patterns with the actual behavior instances, thereby identifying the target dependency patterns. That is, the alignment matching results of the behavior dependency patterns and the actual behavior dependency instances are compared to identify the target dependency patterns.

[0076] In a possible embodiment, comparing the alignment matching results to identify the target dependency pattern includes: the alignment matching results include the matching degree of the behavior dependency instance and the behavior dependency pattern set; comparing the matching degree to identify the target dependency pattern with the highest matching degree with the behavior dependency instance. Exemplarily, it is assumed that the preliminary behavior dependency pattern includes any form of email link click activity. We can refine this pattern by introducing specific constraints, such as focusing only on emails containing specific keywords (such as "financial report", "urgent update" and other common phishing words), or focusing only on link click activities in emails from unknown or uncommon domain names. By introducing these constraints, we can effectively reduce the risk of misjudging normal email reading and web browsing behaviors as malicious activities.

[0077] Through these steps, the genetic information fusion disambiguation of advanced behavioral genes is completed. The conversion from highly abstract advanced behavioral gene information (such as attack patterns) to specific empirical behavioral data achieves high accuracy in attack behavior analysis. Not only does it integrate and utilize detailed information of attack behavior instances in the analysis process of advanced behavioral gene recognition, but it also realizes automatic intelligent analysis based on the powerful ability of LLMs in processing abstract concepts and complex semantics, greatly improving the efficiency and accuracy of security analysts in assessing potential threats and responding quickly.

[0078] S104: Construct threat search query prompts according to the target dependency pattern, and use the threat search query prompts to interact with the large language model to identify attack behaviors that are consistent with the target dependency pattern behavior gene semantics.

[0079] In S104, in a possible embodiment, a threat search query prompt word is constructed according to a target dependency pattern, including: the threat search query prompt word includes a target dependency pattern and a preset system database query statement template; the system database query statement template combines a structured query language or a time series database query language.

[0080] Exemplarily, LLMs first receive threat search query prompts written in natural language. These prompts are written by network security experts based on the analysis results of the advanced behavior genes of the attack organization after genetic information fusion and disambiguation (i.e., the target dependency pattern identified by analysis according to step S103) to accurately reflect the actual network security scenarios and needs. This type of prompts also includes template examples of system database query statements, which combine forms such as SQL (Structured Query Language) or TBQL (Time Series Database Query Language) to improve query efficiency and ensure its cross-platform compatibility. Subsequently, LLMs uses its advanced natural language processing capabilities to generate structured query statements adapted to the system log database based on the received prompts. In the query execution phase, the system log database is searched according to these structured query statements to identify potential attack behaviors that are consistent with the semantics of the target dependency pattern behavior genes to achieve augmented threat search.

[0081] Exemplarily, based on the target dependency pattern, targeted threat hunting queries can be designed. For example, a threat hunting query is created to specifically detect activities related to constrained email link clicks in system logs, such as email opening records with specific keywords, link access records from unknown domain names, etc. Further, the threat hunting space can be expanded, and the focus can be expanded from email link clicks to subsequent possible malicious activities, such as abnormal login attempts, unusual network traffic, unauthorized data access, etc., which can all be used as part of the augmented threat hunting pattern.

[0082] Through the implementation of the above-mentioned augmented threat hunting, not only the abstract understanding of attack patterns is refined, and the false alarm probability of threat hunting based solely on behavioral dependency patterns is reduced, but also a more accurate augmented threat hunting strategy can be constructed based on the refined dependency patterns after constraints, making the detection and response to network security threats more efficient and accurate, and further enhancing the search capability and dynamic response efficiency of network security threat intelligence.

[0083] See also Figure 3 In the present invention, the behavioral gene hierarchy of the attacking organization includes a primary behavioral gene layer, an advanced behavioral gene layer, and a gene information fusion and disambiguation layer.

[0084] Among them, in the primary behavior gene layer, the primary behavior gene information structure is refined into two data forms for analysis: the process behavior sequence instance data (divided into Figure 3 The behavior sequence instance layer shown in the figure is handed over to the random forest model for efficient homology analysis; the behavior dependency instance data (divided into Figure 3 The behavior dependency instance layer shown in the figure is processed by a series of processes such as data cleaning, entity extraction, and grammatical dependency analysis, and is used as part of the LLM input in the gene information fusion disambiguation layer to strengthen the constraint feedback analysis of advanced behavior genes. Using such a diversion analysis method to process primary behavior gene data allows the lightweight model to process process behavior sequences in a timely manner, effectively improving the system's threat detection rate and accelerating the perception of system threats, thereby improving the time efficiency of the threat analysis system.

[0085] The genetic information fusion disambiguation layer designed by the present invention processes the attack abstract description (divided into Figure 3 The behavior threat pattern layer shown in the figure is transformed into a behavior dependency pattern based on the system behavior ontology type (divided into Figure 3The behavior dependency pattern layer shown in the figure is then input into the large model together with the system behavior dependency instances in the primary behavior gene information after processing (data cleaning, entity extraction, grammatical dependency parsing), and feedback constraint prompts are constructed for analysis and alignment, so as to realize the use of empirical constraints on the pattern candidate space and expand the threat empirical semantics with the pattern, so as to facilitate the system to identify and respond to threats more accurately. Specifically, the gene information fusion disambiguation layer aims to coordinate the semantic differences between the logical abstraction of the high-level behavior gene layer and the empirical data of the primary behavior gene layer. It realizes the alignment and constraint of the two types of gene information by integrating and aligning the behavior gene information at different levels (i.e., mapping from high-level strategies to specific behaviors), providing the necessary foundation for accurate dynamic threat hunting and feedback, so as to improve the accuracy and efficiency of threat detection. In this way, not only the accuracy of threat detection is improved, but also an innovative technical approach is provided to deal with complex network security threats.

[0086] The multi-dimensional data analysis and advanced behavioral gene identification technology used in the behavioral gene identification method of the network attack organization provided by the present invention significantly improve the identification accuracy and coverage of complex network attacks, and solve the shortcomings of the existing technology in interpreting complex attack information. By utilizing advanced machine learning algorithms and natural language processing technologies, it is possible to effectively integrate and deeply analyze the implicit features in multi-source heterogeneous data, surpassing the limitations of the traditional security analysis framework in technical integration. Through precise deviation correction and timely feedback mechanisms, the delay of threat response is greatly reduced, ensuring a rapid and accurate response to emerging threats. Combined with the design of dynamic databases and retrieval grammars, the present invention supports real-time mining of dynamic threat patterns and enhances the ability of security systems to adapt to increasingly complex security threats. The method combines the research results of multiple sub-fields such as artificial intelligence, big data analysis, machine learning and network behavior analysis, and provides a new technical path for modern network security threat identification and defense. By implementing the technical solution of the present invention, deep insight and effective prevention of complex attack strategies can be achieved in the field of network security, providing a more solid technical guarantee for information security.

[0087] In addition, the beneficial effects that can be achieved by the present invention can also be specifically embodied in the following aspects: the present invention deconstructs and separates the primary behavioral gene data, realizes the combination of lightweight behavioral sequence data captured by real-time audit and random forest model, and constructs an efficient and accurate primary behavioral gene homology identification method. Different from the offline analysis and batch processing in the prior art, the real-time deconstruction and identification method of the present invention can instantly process and respond to real-time security events, providing faster threat detection and response capabilities. The fusion disambiguation layer of the present invention designs a prompt design specific to attack pattern recognition, which is a unique method for obtaining optimized information matching from behavioral gene abstraction and specific instance behavior, especially in effectively aligning and constraining abstract attack patterns with actual monitored behavioral instances, providing a more in-depth and systematic solution strategy. The present invention realizes highly adaptable real-time and flexible threat hunting through the construction of dynamic augmented threat search patterns and the design of dynamic constraint feedback mechanisms. Compared with traditional technologies, the difference is that it no longer relies on manually updated static rule bases, but adopts automatic learning and adaptive mechanisms, making threat detection and response more intelligent and timely.

[0088] See also Figure 4 This embodiment also provides a behavior gene identification device for a network attack organization, which is used to implement the above method embodiment. The device includes: a data collection unit 201, which is used to collect basic data, and the basic data includes the attack organization behavior gene knowledge graph and system process data obtained by capturing attack intelligence data to extract threat knowledge and updating.

[0089] The behavior gene homology recognition unit 202 is used to extract malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, extract process behavior sequence data from the system process data and input it into the primary behavior gene homology recognition model to perform behavior gene homology reasoning on the corresponding process behavior sequence data, thereby predicting a target attack organization that matches the characteristics of the malicious behavior sequence.

[0090] The behavior gene information alignment unit 203 is used to use the attack organization behavior gene knowledge graph to retrieve and organize the high-level behavior gene information of the target attack organization, apply the large language model to convert the high-level behavior gene information into a behavior dependency pattern set, analyze the behavior logs of known attack events in the system process data to obtain corresponding behavior dependency instances, use the large language model to align and match the behavior dependency instances with the behavior dependency pattern set to obtain an alignment matching result, and compare the alignment matching results to identify the target dependency pattern.

[0091] The threat search unit 204 is used to construct a threat search query prompt word according to the target dependency pattern, and use the threat search query prompt word to interact with the large language model to identify attack behaviors that are consistent with the target dependency pattern behavior gene semantics.

[0092] Among them, the behavior gene homology recognition unit 202 extracts malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, including: extracting malicious behavior sequences from the attack organization behavior gene knowledge graph; performing feature engineering on the malicious behavior sequences and dividing the results into training sets and validation sets according to a preset ratio, and inputting them into a random forest classification model for training and validation, so as to obtain a primary behavior gene homology recognition model for predicting the target attack organization that matches the characteristics of the malicious behavior sequences.

[0093] The behavior gene information alignment unit 203 applies a large language model to convert high-level behavior gene information into a set of behavior dependency patterns, including: constructing high-level behavior gene projection prompt words, which are used to guide the large language model to understand and convert high-level behavior gene information; and using high-level behavior gene projection prompt words to interact with the large language model to obtain a set of behavior dependency patterns derived from the high-level behavior gene information.

[0094] The behavior gene information alignment unit 203 uses a large language model to align and match the behavior dependency instance with the behavior dependency pattern set to obtain an alignment matching result, including: constructing a constraint feedback prompt word, the constraint feedback prompt word is used to guide the large language model to evaluate the matching degree between the behavior dependency instance and the behavior dependency pattern set from a preset perspective; using the constraint feedback prompt word to interact with the large language model to obtain the alignment matching result of the behavior dependency instance and the behavior dependency pattern set.

[0095] The alignment and matching result of the behavior dependency instance and the behavior dependency pattern set includes the matching degree between the behavior dependency instance and the behavior dependency pattern set; the behavior gene information alignment unit 203 compares the alignment and matching results to identify the target dependency pattern, including: comparing the matching degree to identify the target dependency pattern with the highest matching degree with the behavior dependency instance.

[0096] The threat search unit 204 constructs a threat search query prompt word according to the target dependency pattern, including: the threat search query prompt word includes the target dependency pattern and a preset system database query statement template; the system database query statement template combines a structured query language or a time series database query language.

[0097] All relevant contents of each step involved in the above method embodiment can be referred to the functional description of the corresponding functional module, and will not be repeated here.

[0098] In other embodiments of the present application, the present application discloses an electronic device, such as Figure 5 As shown, the electronic device 300 may include: one or more processors 301; a memory 302; a display 303; one or more applications (not shown); and one or more computer programs 304. The above components may be connected via one or more communication buses 305. The one or more computer programs 304 are stored in the above memory and configured to be executed by the one or more processors 301. The one or more computer programs 304 include instructions. The above instructions may be used to execute the following: Figure 1 , Figure 4 And each step in the corresponding embodiment.

[0099] Through the description of the above implementation methods, technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0100] Each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units.

[0101] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as flash memory, mobile hard disk, read-only memory, random access memory, disk or optical disk.

[0102] The above is only a specific implementation of the embodiment of the present application, but the protection scope of the embodiment of the present application is not limited thereto, and any changes or replacements within the technical scope disclosed in the embodiment of the present application should be included in the protection scope of the embodiment of the present application. Therefore, the protection scope of the embodiment of the present application should be based on the protection scope of the claims.

Claims

1. A behavioral gene identification method for a network attack organization, characterized in that: include: Collect basic data, including attack organization behavior gene knowledge graph and system process data obtained by capturing attack intelligence data to extract threat knowledge and updating; Extract malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, extract process behavior sequence data from system process data and input it into the primary behavior gene homology recognition model to perform behavior gene homology reasoning corresponding to the process behavior sequence data, so as to predict a target attack organization that matches the characteristics of the malicious behavior sequence, wherein extracting malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model includes: extracting malicious behavior sequences from the attack organization behavior gene knowledge graph, performing feature engineering processing on the malicious behavior sequences and dividing the obtained results into a training set and a validation set according to a preset ratio and inputting them into a random forest classification model for training and validation, so as to obtain a primary behavior gene homology recognition model for predicting a target attack organization that matches the characteristics of the malicious behavior sequence; Using the attack organization behavior gene knowledge graph to retrieve and organize high-level behavior gene information of the target attack organization, the high-level behavior gene information includes attack mode information adopted by the attack organization, applying a large language model to convert the high-level behavior gene information into a behavior dependency pattern set, analyzing the behavior logs of known attack events in the system process data to obtain corresponding behavior dependency instances, using a large language model to align and match the behavior dependency instances with the behavior dependency pattern set to obtain an alignment matching result, and comparing the alignment matching results to identify the target dependency pattern; A threat search query prompt word is constructed according to the target dependency pattern, and the threat search query prompt word is used to interact with a large language model to identify attack behaviors that are semantically consistent with the behavior gene of the target dependency pattern.

2. The method according to claim 1, characterized in that The high-level behavioral gene information is converted into a set of behavioral dependency patterns using a large language model, including: Constructing high-level behavioral gene projection prompt words, wherein the high-level behavioral gene projection prompt words are used to guide the large language model to understand and transform the high-level behavioral gene information; The high-level behavioral gene projection prompt words are used to interact with a large language model to obtain a set of behavioral dependency patterns derived from the high-level behavioral gene information.

3. The method according to claim 1, characterized in that Using a large language model to align and match the behavior dependency instance with the behavior dependency pattern set to obtain an alignment matching result, including: Constructing a constraint feedback prompt word, wherein the constraint feedback prompt word is used to guide the large language model to evaluate the matching degree between the behavior dependency instance and the behavior dependency pattern set from a preset perspective; The constraint feedback prompt word is used to interact with the large language model to obtain an alignment matching result between the behavior dependency instance and the behavior dependency pattern set.

4. The method according to claim 1, characterized in that: The alignment and matching result includes a matching degree between the behavior dependency instance and the behavior dependency pattern set; Comparing the alignment matching results to identify target dependency patterns includes: The matching degrees are compared to identify the target dependency pattern that has the highest matching degree with the behavior dependency instance.

5. The method according to claim 1, characterized in that Construct threat hunting query hints based on target dependency patterns, including: The threat search query prompt words include a target dependency pattern and a preset system database query statement template; The system database query statement template combines structured query language or time series database query language.

6. A behavioral gene recognition device for a network attack organization, characterized in that: The device comprises: A data collection unit, used to collect basic data, wherein the basic data includes attack organization behavior gene knowledge graph and system process data obtained by capturing attack intelligence data to extract threat knowledge and updating; A behavior gene homology recognition unit is used to extract malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, extract process behavior sequence data from system process data and input it into the primary behavior gene homology recognition model to perform behavior gene homology reasoning corresponding to the process behavior sequence data, so as to predict a target attack organization that matches the characteristics of the malicious behavior sequence, wherein the behavior gene homology recognition unit extracts malicious behavior sequences from the attack organization behavior gene knowledge graph to construct and train a primary behavior gene homology recognition model, including: extracting malicious behavior sequences from the attack organization behavior gene knowledge graph, performing feature engineering processing on the malicious behavior sequences and dividing the obtained results into a training set and a validation set according to a preset ratio, inputting the results into a random forest classification model for training and validation, so as to obtain a primary behavior gene homology recognition model for predicting a target attack organization that matches the characteristics of the malicious behavior sequence; a behavior gene information alignment unit, for using the attack organization behavior gene knowledge graph to retrieve and organize the high-level behavior gene information of the target attack organization, the high-level behavior gene information including the attack mode information adopted by the attack organization, applying a large language model to convert the high-level behavior gene information into a behavior dependency pattern set, analyzing the behavior logs of known attack events in the system process data to obtain corresponding behavior dependency instances, using a large language model to align and match the behavior dependency instances with the behavior dependency pattern set to obtain an alignment matching result, and comparing the alignment matching results to identify the target dependency pattern; The threat search unit is used to construct a threat search query prompt word according to the target dependency pattern, and use the threat search query prompt word to interact with the large language model to identify attack behaviors that are consistent with the behavior gene semantics of the target dependency pattern.

7. The behavioral gene identification device for a network attack organization according to claim 6, characterized in that: The behavior gene information alignment unit uses a large language model to convert the high-level behavior gene information into a behavior dependency pattern set, including: Constructing high-level behavioral gene projection prompt words, wherein the high-level behavioral gene projection prompt words are used to guide the large language model to understand and transform the high-level behavioral gene information; The high-level behavioral gene projection prompt words are used to interact with a large language model to obtain a set of behavioral dependency patterns derived from the high-level behavioral gene information.

8. The behavioral gene identification device for a network attack organization according to claim 6, characterized in that: The behavior gene information alignment unit uses a large language model to align and match the behavior dependency instance with the behavior dependency pattern set to obtain an alignment and matching result, including: Constructing a constraint feedback prompt word, wherein the constraint feedback prompt word is used to guide the large language model to evaluate the matching degree between the behavior dependency instance and the behavior dependency pattern set from a preset perspective; The constraint feedback prompt word is used to interact with the large language model to obtain an alignment matching result between the behavior dependency instance and the behavior dependency pattern set.

9. The behavioral gene identification device for a network attack organization according to claim 6, characterized in that: The alignment and matching result includes a matching degree between the behavior dependency instance and the behavior dependency pattern set; The behavior gene information alignment unit compares the alignment matching results to identify the target dependency pattern, including: The matching degrees are compared to identify the target dependency pattern that has the highest matching degree with the behavior dependency instance.

10. The behavioral gene identification device for network attack organizations according to claim 6, characterized in that: The threat search unit constructs a threat search query prompt word according to the target dependency pattern, including: The threat search query prompt words include a target dependency pattern and a preset system database query statement template; The system database query statement template combines structured query language or time series database query language.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the behavioral gene identification method of a network attack organization according to any one of claims 1 to 5 is implemented.

12. An electronic device, characterized in that: include: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program stored in the memory so that the electronic device executes the behavioral gene identification method for a network attack organization according to any one of claims 1 to 5.