Method and system for constructing threat detection and response model based on traceability graph

By building a threat detection and response model based on traceability map, combining tag propagation, attenuation mechanism and network security knowledge graph, the problem that existing technology is difficult to adapt to the multi-change threat environment is solved, and more accurate threat detection and response is achieved, improving Efficiency of cybersecurity threat detection and response.

CN119051943AInactive Publication Date: 2024-11-29GUANGZHOU UNIVERSITY

Patent Information

Application Number
CN202411167740.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing threat detection methods are difficult to adapt to the changing system environment, cannot dynamically adjust according to the new attack methods, and lack the ability to propose specific optimization and mitigation suggestions for the nodes with attack characteristics in the graph.

Method used

By obtaining the original log event, building an initial traceability map, defining the event data structure, obtaining tag values ​​based on the preset tag propagation table, performing tag attenuation control, obtaining detection strategies, annotating the initial traceability map, building a network security knowledge map, and inputting it into the threat The detection and response model is trained to form the final threat detection and response model.

Benefits of technology

It significantly improves the accuracy and depth of network security threat detection and response, and can adjust the tag propagation and attenuation mechanism according to specific security needs and system environment, identify attack paths and key attack nodes more accurately, provide targeted response strategies, and improve security Personnel’s work efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119051943A_ABST
    Figure CN119051943A_ABST
Patent Text Reader

Abstract

The invention provides a method and a system for constructing a threat detection and response model based on a traceability graph. The method comprises the following steps: constructing an initial traceability graph; defining a data structure in the initial traceability graph, calling out a first preset rule according to the data structure, and obtaining a label value according to the first preset rule; performing attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value, and obtaining a corresponding detection strategy from a preset database according to the updated label value; acquiring attack behavior information corresponding to each updated label value according to a detection strategy, and labeling the initial traceability graph according to the attack behavior information to obtain a final traceability graph; and constructing a network security knowledge graph, and inputting the network security knowledge graph and the final traceability graph into the initial threat detection and response model for training to obtain a final threat detection and response model. According to the invention, the accuracy and response efficiency of network security threat detection can be improved.
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 protection technology, and in particular to a method and system for constructing a threat detection and response model based on a traceability graph. Background Art

[0002] In recent years, the number of cyber attacks has increased significantly, and attacks on critical information infrastructure of various systems have become more and more frequent. At the same time, the methods used in the attack process have become more and more advanced, and advanced persistent threats (APT) with clear goals, long duration, good concealment, and advanced technology have also emerged.

[0003] Existing threat detection methods generally require network security personnel to use traceability graph technology to identify attackers' tactical and technical procedures. However, this detection method relies on static rules and has a delayed update, and cannot be dynamically adjusted according to new attack methods. It also lacks the ability to propose specific optimization and mitigation suggestions for nodes in the graph with attack characteristics. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for constructing a threat detection and response model based on a traceability graph, aiming to solve the problem that traditional threat detection methods are difficult to apply to system environments with multiple changing threats, that is, they cannot be dynamically adjusted according to new attack methods.

[0005] In a first aspect, the present invention provides a method for constructing a threat detection and response model based on a provenance graph, the method comprising:

[0006] Obtaining original log events, and constructing an initial traceability graph based on the original log events;

[0007] Define the data structure of each event in the initial traceability graph, retrieve the first preset rule corresponding to each event from the preset label propagation table according to the data structure, and obtain the label value of each event according to the first preset rule;

[0008] Performing attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value, and acquiring a corresponding detection strategy from a preset database according to the updated label value;

[0009] Acquire attack behavior information corresponding to each updated label value according to the detection strategy, and annotate the initial traceability graph according to the attack behavior information to obtain a final traceability graph;

[0010] A network security knowledge graph is constructed, and the network security knowledge graph and the final traceability graph are input into an initial threat detection and response model for training to obtain a final threat detection and response model.

[0011] Furthermore, the step of obtaining the original log event and constructing an initial traceability graph according to the original log event includes:

[0012] The original log event includes at least one event, and information is extracted from the original log event to obtain at least one system entity corresponding to each event and the relationship between the system entities;

[0013] An initial traceability graph is constructed with the system entities as edges and the relationships between the system entities as nodes.

[0014] Furthermore, the step of defining the data structure of each event in the initial traceability graph, retrieving the first preset rule corresponding to each event from the preset label propagation table according to the data structure, and obtaining the label value of each event according to the first preset rule includes:

[0015] The data structure includes an event type and an event subject type corresponding to the event type, the event type includes creation, reading, writing, loading, execution, and injection, and the event subject type includes benign subject, suspicious subject, and suspicious environment subject;

[0016] If the event type is creation, the first preset rule under all event subject types is the same, specifically: x.dtag = s.dtag

[0017] If the event type is read, the first preset rule under all event subject types is the same, specifically: s.dtag = min(s.dtag, x.dtag)

[0018] If the event type is write, when the event subject type is benign subject, the first preset rule is: x.dtag=min(s.dtag+a b ,x.dtag); when the event subject type is a suspicious subject, the first preset rule is: x.dtag = min(s.dtag,x.dtag); when the event subject type is a suspicious environment subject, the first preset rule is x.dtag = min(s.dtag + a e ,x.dtag);

[0019] If the event type is loading, the first preset rule under all event subject types is the same, specifically: s.stag=min(s.stag,x.itag), s.dtag=min(s.dtag,x.dtag);

[0020] If the event type is execution, when the event subject type is a benign subject, the first preset rule is: s.stag=x.itag, s.dtag=<1.0,1.0>; when the event subject type is a suspicious subject, the first preset rule is: s.stag=min(x.itag,susp_env), s.dtag=min(s.dtag,x.dtag); when the event subject type is a suspicious environment subject, the first preset rule is: s.dtag==x.itag, s.dtag=min(s.dtag,x.dtag);

[0021] If the event type is injection, the first preset rule under all event subject types is the same, specifically: s′.stag=min(s′.stag,s.itag), s′.dtag=min(s.dtag,s′.dtag);

[0022] Among them, x.dtag represents the data tag value of the executed object under the corresponding event, s.dtag represents the data tag value of the entity under the corresponding event, s.stag represents the subject tag value of the entity under the corresponding event, s′.stag represents the data tag value injected into another entity under the corresponding event, and a b 、a e Both represent correction values, and x.itag represents the integrity label value of the executed object under the corresponding event.

[0023] Furthermore, the step of performing attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value includes:

[0024] For all events whose event subject type is benign subject, label decay is performed according to the following formula:

[0025] s″.dtag=max(s.dtag,d b ·s.dtag+(1-d b )·T qb )

[0026] For all events whose event subject type is suspicious environment subject, label decay is performed according to the following formula:

[0027] s″.dtag=max(s.dtag,d e ·s.dtag+(1-d e )·T qe )

[0028] Among them, s″.dtag represents the updated tag value, d b ,de They represent the attenuation coefficients under benign subjects and suspicious environment subjects, respectively, and their values ​​are between 0 and 1. qb and T qe They represent the expected label values ​​under benign subjects and suspicious environment subjects respectively.

[0029] Furthermore, the step of acquiring a corresponding detection strategy from a preset database according to the updated tag value includes:

[0030] Obtain the integrity label value of the entity or the executed object according to the updated label value, and filter out the integrity label values ​​that meet the suspicious level from all integrity label values;

[0031] The corresponding detection strategy information is obtained from a preset database according to the integrity tag value of the suspicious level, and the detection strategy information includes name information, description information and operation information.

[0032] Furthermore, the steps of constructing a network security knowledge graph, and inputting the network security knowledge graph and the final traceability graph into an initial threat detection and response model for training to obtain a final threat detection and response model include:

[0033] Collecting network security data, including attack patterns, vulnerability types, and attack strategies;

[0034] A network security knowledge graph including relationships between entities is constructed based on the network security data.

[0035] In a second aspect, the present invention provides a system for constructing a threat detection and response model based on a provenance graph, the system comprising:

[0036] A data acquisition module is used to acquire original log events and construct an initial traceability graph based on the original log events;

[0037] A label propagation execution module, used to define the data structure of each event in the initial traceability graph, and retrieve the first preset rule corresponding to each event from the preset label propagation table according to the data structure, and obtain the label value of each event according to the first preset rule;

[0038] A label attenuation execution module, used to perform attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value, and obtain a corresponding detection strategy from a preset database according to the updated label value;

[0039] A labeling module, used to obtain attack behavior information corresponding to each updated label value according to the detection strategy, and label the initial traceability graph according to the attack behavior information to obtain a final traceability graph;

[0040] The model training module is used to construct a network security knowledge graph, and input the network security knowledge graph and the final traceability graph into the initial threat detection and response model for training to obtain the final threat detection and response model.

[0041] Furthermore, the data acquisition module is also used for:

[0042] The original log event includes at least one event, and information is extracted from the original log event to obtain at least one system entity corresponding to each event and the relationship between the system entities;

[0043] An initial traceability graph is constructed with the system entities as edges and the relationships between the system entities as nodes.

[0044] In a third aspect, the present invention provides a readable storage medium, which stores one or more programs, which, when executed by a processor, implement the above-mentioned method for constructing a threat detection and response model based on a tracing graph.

[0045] In a fourth aspect, the present invention provides a computer device, the computer device comprising a memory and a processor, wherein:

[0046] The memory is used to store computer programs;

[0047] When the processor is used to execute the computer program stored in the memory, the above-mentioned method for constructing a threat detection and response model based on the provenance graph is implemented.

[0048] Compared with the prior art, the present invention has the following advantages:

[0049] The present invention combines a label-based provenance graph, a large language model (LLM) and a network security knowledge graph (KG) to achieve detection and response to network security threats, thereby significantly improving the accuracy and depth of network security threat detection and response. Specifically, the label propagation and attenuation mechanism can be adjusted according to specific security requirements and system environment to adapt to threat detection under multiple changes. Secondly, LLM combined with provenance graph technology makes it more accurate to identify attack paths and key attack nodes, providing accurate and specific positioning information for defense strategies. Finally, the integrated knowledge graph not only enriches the model's grasp of knowledge in the field of network security, but also enables the model to provide targeted suggestions and timely response strategies based on the latest threat intelligence, thereby significantly improving the work efficiency of security personnel in network security threat detection and response. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1A flowchart of a method for constructing a threat detection and response model based on a provenance graph proposed in one embodiment of the present invention;

[0051] Figure 2 A schematic diagram of the structure of a system for building a threat detection and response model based on a provenance graph proposed in one embodiment of the present invention.

[0052] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not 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.

[0054] Regarding the technical term "source traceability graph", it refers to a graphical security incident analysis tool that is used to track and analyze the attacker's action path and attack activities in the system, helping security experts understand the source, propagation path and impact scope of security incidents so as to take appropriate countermeasures.

[0055] About the technical term "Large Language Model (LLM)" is a language model with a large number of parameters and complex structure, usually built based on deep learning technology, which can process and generate natural language text. Large language models require a lot of computing resources and data during training, and are usually able to learn deeper language regularities and semantic information. These models include OpenAI's GPT series (such as GPT-4), Google's BERT model, etc.

[0056] About the technical term "Knowledge Graph (KG) of Cyberspace Security" is a structured data representation method used to describe and construct knowledge and information in the field of cybersecurity. It consists of entities (such as malware WannaCry, vulnerability CVE-2021-34527, etc.), attributes (such as the release date of vulnerability CVE-2021-34527 (June 2021), hazard level (high), affected system (Windows operating system), etc.), and relationships between entities (malware WannaCry exploited the vulnerability CVE-2021-34527, etc.). Cybersecurity knowledge graphs are usually compiled from multiple sources such as security reports, attack logs, vulnerability databases, expert knowledge, etc. By building and utilizing cybersecurity knowledge graphs, security experts can better understand the complexity of the cybersecurity field, discover attack patterns and threats, and assist in security decision-making and the formulation of countermeasures.

[0057] The applicant found that existing traceability detection methods are usually poorly adaptable to environmental changes, and their ability to flexibly perceive dynamic attack scenarios and changing threats is not very high. In addition, the existing technology uses LLM questions to detect and respond to threats, but does not fully utilize the traceability graph technology and knowledge graph to improve the recognition accuracy of attack nodes and enrich the LLM's cyberspace security field knowledge. The lack of application of traceability graph technology may result in low accuracy in identifying attack propagation paths and key attack nodes. At the same time, there is no knowledge graph to enhance the knowledge of the large language model in the field of network security. The model's knowledge depth in the field of network security is insufficient, which affects the formulation of effective defense strategies and response measures. And due to the lack of in-depth understanding of specific attack scenarios, the generated response measures may lack pertinence and cannot effectively respond to specific threats.

[0058] Based on this, the present invention proposes a method for constructing a threat detection and response model based on a traceability graph, and sets the label propagation and attenuation mechanism to be adjusted according to specific security requirements and system environment to adapt to threat detection under multiple changes. Secondly, by combining LLM with traceability graph technology, the identification of attack paths and key attack nodes is more accurate, providing accurate and specific positioning information for defense strategies. Finally, the integrated knowledge graph not only enriches the model's grasp of knowledge in the field of network security, but also enables the model to provide highly targeted suggestions and timely response strategies based on the latest threat intelligence, thereby significantly improving the work efficiency of security personnel in network security threat detection and response.

[0059] See also Figure 1 , which shows a method for constructing a threat detection and response model based on a tracing graph according to an embodiment of the present invention, the method comprises steps S101 to S105, wherein:

[0060] Step S101: obtaining an original log event, and constructing an initial traceability graph according to the original log event;

[0061] It should be pointed out that the original log events come from different operating systems and hosts, such as audit logs from Linux and ETW from Windows, etc. In addition, the original log events include at least one event. After collecting the original log events, they are converted into a structured format, such as JSON format, and then cleaned and information extracted to extract important information, thereby obtaining at least one system entity corresponding to each event and the relationship between system entities.

[0062] As an example but not a limitation, system entities can be processes, files, sockets, etc., and the relationships between system entities can be read, write, execute, etc. For example, for an attack event in the original log event: the attacker guides the user to download and execute a malware (malware.exe) through a phishing email. After the malware is executed in the system, a new process (malware1234) is created, and credentials are stolen by reading sensitive files (confidential.txt) in the system. Subsequently, the malware communicates the stolen credentials through a network connection (socket socket1234) to a remote server. In this attack event, processes, files, sockets, and attack events are entities, while execution, reading, communication, etc. are the relationships between these entities, and then the initial traceability graph is constructed with the system entities as edges and the relationships between the system entities as nodes.

[0063] In addition, in order to build a good initial traceability graph, in some embodiments, RML (a declarative RDF mapping language) is used to map and convert the parsed audit data into a well-defined RDF-based traceability graph, and then the graph is compressed to automatically merge repeated RDF outputs with the same subject, attribute, and object, thereby eliminating redundant events. Secondly, the header dictionary triple is used as a graph compression technology, which provides a compact data structure and binary serialization format for RDF.

[0064] Step S102: defining a data structure of each event in the initial traceability graph, and retrieving a first preset rule corresponding to each event from a preset label propagation table according to the data structure, and obtaining a label value of each event according to the first preset rule;

[0065] It should be noted that in this step, in order to effectively manage and monitor the credibility and sensitivity of the data, a label propagation and label decay mechanism is introduced in the initial traceability graph to define the data structure for each node (system entity, such as process) and edge (relationship between entities, such as reading, creating, etc.) and store label information.

[0066] In some embodiments, the data structure includes event types and event subject types corresponding to the event types, the event types including creation, reading, writing, loading, execution, injection, etc., and the event subject types including benign subjects, suspicious subjects, suspicious environment subjects, etc.

[0067] Specifically, if the event type is creation, the first preset rule under all event subject types is the same, specifically: x.dtag = s.dtag

[0068] If the event type is read, the first preset rule under all event subject types is the same, specifically: s.dtag = min(s.dtag, x.dtag)

[0069] If the event type is write, when the event subject type is benign subject, the first preset rule is: x.dtag=min(s.dtag+a b ,x.dtag); when the event subject type is a suspicious subject, the first preset rule is: x.dtag = min(s.dtag,x.dtag); when the event subject type is a suspicious environment subject, the first preset rule is x.dtag = min(s.dtag + a e ,x.dtag);

[0070] If the event type is loading, the first preset rule under all event subject types is the same, specifically: s.stag=min(s.stag,x.itag), s.dtag=min(s.dtag,x.dtag);

[0071] If the event type is execution, when the event subject type is a benign subject, the first preset rule is: s.stag=x.itag, s.dtag=〈1.0,1.0〉; when the event subject type is a suspicious subject, the first preset rule is: s.stag=min(x.itag,susp_env), s.dtag=min(s.dtag,x.dtag); when the event subject type is a suspicious environment subject, the first preset rule is: s.dtag==x.itag, s.dtag=min(s.dtag,x.dtag);

[0072] If the event type is injection, the first preset rule under all event subject types is the same, specifically: s′.stag=min(s′.stag,s.itag), s′.dtag=min(s.dtag,s′.dtag);

[0073] Among them, x.dtag represents the data tag value of the executed object under the corresponding event, s.dtag represents the data tag value of the entity under the corresponding event, s.stag represents the subject tag value of the entity under the corresponding event, s′.stag represents the data tag value injected into another entity under the corresponding event, and a b 、a e Both represent correction values, and x.itag represents the integrity label value of the executed object under the corresponding event.

[0074] Please refer to Table 1 and Table 2 below. The first preset rule is mainly for data operations and code operations. Table 1 is the label propagation rule for data operations, and Table 2 is the label propagation rule for code operations:

[0075] Table 1 == table of label propagation rules for data operations

[0076]

[0077] Table 2 ==Tag propagation rules for code operations

[0078]

[0079]

[0080] In addition, it should be pointed out that stag is used to indicate whether the subject is benign, suspicious, or in a suspicious environment. The subject tag value is usually expressed as = between 0 and 1, with lower values ​​corresponding to higher confidentiality or lower suspicion, and higher values ​​corresponding to higher integrity or lower suspicion. Dtag represents the security attributes of data, such as confidentiality and integrity. Itag is used to mark the integrity of data or subjects and is part of dtag.

[0081] In addition, in some embodiments, data labels generally include two types: confidentiality labels and integrity labels. c represents the confidentiality label. High confidentiality data should be assigned a confidentiality label of =0, while low confidentiality data should be assigned a label of =1.0. At the same time, values ​​in the range of =[0.0, =0.5)= are considered to be high confidentiality, while values ​​in the range of =[0.5, =1.0]= correspond to low confidentiality. =i represents the integrity label. High integrity data in the range of [0.5, =1.0] is defined as benign, and low integrity data in the range of [0.0, =0.5] is defined as suspicious. With integrity and confidentiality, the current data label s.dtag=〈c,i〉 can be obtained.

[0082] In addition, it should be noted that some event types in code operations have two first preset rules corresponding to them, while data operations have only one. This is because for code operations, the system not only needs to update the subject tag (s.stag) to reflect changes in the subject's credibility, but also needs to update the data tag (s.dtag) to reflect the relationship between the subject and the data it processes. For data operations, the system only needs to update the data tag.

[0083] In addition, this step can track information flows and operational behaviors in real time by assigning dynamically updated confidentiality and integrity labels to data and entities. Compared with traditional static monitoring methods (such as signature-based malware detection after storage in a signature library), labels have significant advantages in dealing with complex and multi-stage attacks. Traditional methods rely solely on known threat signatures and rules, making it difficult to capture unknown zero-day attacks, and have high false alarm rates and lack of context awareness. This method combines labels with known signature rules and large language models to achieve more accurate threat detection and response. It can not only capture known threats, but also discover unknown threats through behavioral analysis and semantic understanding, and provide more targeted suggestions.

[0084] Step S103: performing attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value, and obtaining a corresponding detection strategy from a preset database according to the updated label value;

[0085] When the traceability graph nodes interact with system objects in large numbers, a large number of benign events may be mistaken for attacks, resulting in an increase in false positives. By controlling the propagation of labels, the integrity and confidentiality of the system can be effectively protected, false positives can be reduced, and detection accuracy can be improved. Based on this, this step introduces a label decay mechanism to ensure that the label values ​​of benign subjects and suspicious environmental subjects are gradually adjusted over time to reflect the changes in their credibility and confidentiality over time.

[0086] In some embodiments, it is necessary to define label decay rules according to different subject types (such as benign and suspicious environments). For all events whose event subject type is a benign subject, label decay is performed according to the following formula:

[0087] s″.dtag=max(s.dtag,d b ·s.dtag+(1-d b )·T qb )

[0088] For all events whose event subject type is suspicious environment subject, label decay is performed according to the following formula:

[0089] s″.dtag=max(s.dtag,d e ·s.dtag+(1-d e )·T qe )

[0090] Among them, s″.dtag represents the updated tag value, d b ,d e They represent the attenuation coefficients under benign subjects and suspicious environment subjects, respectively, and their values ​​are between 0 and 1. qb and T qe They represent the expected label values ​​under benign subjects and suspicious environment subjects respectively.

[0091] In some embodiments, the integrity label value of the entity or the executed object is obtained based on the updated label value, and the integrity label value that meets the suspicious level is screened out from all the integrity label values;

[0092] The corresponding detection strategy information is obtained from a preset database according to the integrity tag value of the suspicious level, wherein the detection strategy information includes name information, description information and operation information. The suspicious level means that the integrity tag value is less than or equal to 0.5.

[0093] For example, please refer to the following Table 3, which shows a preset database. According to Table 3, by searching the preset database using the data integrity condition and other conditions, relevant detection strategy information can be obtained:

[0094] Table 3

[0095]

[0096] Step S104: acquiring attack behavior information corresponding to each updated label value according to the detection strategy, and marking the initial traceability graph according to the attack behavior information to obtain a final traceability graph;

[0097] In some embodiments, in combination with the attack detection strategy of Table 3, known attack behaviors in log events are identified by using IoC (Indicator Threat Intelligence) definitions and signature rule-based detection methods. Maintaining rule sets and signatures may require a lot of work, so Sigma is used, which is an open, shareable, community-driven and universal rule format for log threat detection. These Sigma rules are translated into SPARQL query expressions, and these rules are executed on the traceability graph to detect potential attack threats in the threat and alarm modules, and then marked in the traceability graph. Exemplarily, for example, FileExec is a detection strategy in Table 3 that detects file-based malware execution. Operation: exec(s,o) or load(s,o), indicating that the subject s executes or loads the object o, data integrity condition: o.itag<0.5, that is, the integrity tag itag of the file is less than 0.5, indicating that the file may have been tampered with or untrusted. Other conditions: s.stag is benign, that is, the stag of the subject s must be benign. Then, combined with Sigma rules, the Sigma rules are converted into SPARQL query expressions to perform detection in the traceability graph and mark suspicious events.

[0098] Step S105: construct a network security knowledge graph, and input the network security knowledge graph and the final traceability graph into the initial threat detection and response model for training to obtain the final threat detection and response model.

[0099] It should be pointed out that regarding the process of building a network security knowledge graph: first, network security data needs to be collected, such as from sources such as MITRE ATT&CK, CAPEC, and CWE, and the network security data includes information such as attack patterns, vulnerability types, and attack strategies; and then a network security knowledge graph containing the relationships between entities is constructed based on the network security data. Specifically, the collected network security data is used to build a knowledge graph of entities (threat actors, attack techniques, vulnerabilities, etc.) and relationships between entities ("open", "execute", "belong to", etc.) using the open source tool Neo4j.

[0100] The Neo4j database is connected through a Python script to extract the knowledge graph data and automatically pass it to the large language model (LLM) for analysis and answering, ultimately achieving real-time monitoring and intelligent response to network security threats through an automated process. In this step, the initial threat detection and response model uses the large prediction model.

[0101] Based on the results of the above threat detection engine, the threat alert type and the attack scene sub-graph are input into the initial threat detection and response model, and the graph of the cyberspace security domain knowledge base (constructed by ATT&CK, etc.) is used to enhance the response strategy generation effect of the initial threat detection and response model.

[0102] For example, please refer to the following Table 4 and Table 5 for comparison. Table 4 is the response strategy generation when LLM is not enhanced, and Table 5 is the response strategy generation generated after enhancement. The technical solution of the present invention greatly enhances the response strategy generation effect:

[0103] Table 4

[0104]

[0105] Table 5

[0106]

[0107]

[0108] In summary, the above-mentioned method for constructing a threat detection and response model based on a tracing graph has the following advantages:

[0109] 1. This invention combines the label-based provenance graph technology, the large language model LLM and the network security knowledge graph KG for the first time, that is, using these three as an overall framework to solve the problem of network security threat detection and response, especially in real-time automatic generation of mitigation suggestions to combat complex attack scenarios, thereby improving the work efficiency of network security practitioners.

[0110] 2. The present invention breaks the limitations of static rules by applying dynamic and accurate label-based traceability graph technology, and greatly improves the accuracy of identifying attack paths and key attack nodes, which makes it possible to more accurately locate the source and propagation path of attack threats, providing a strong basis and foundation for formulating effective defense strategies.

[0111] 3. The present invention uses a dynamically integrated network security knowledge graph, which not only provides LLM with rich network security domain knowledge, but also enables the model to use the latest threat intelligence to provide targeted and timely mitigation suggestions and response strategies. This strategy not only improves the efficiency of detection and response, but also ensures that the measures taken can effectively target the actual threats in the current specific network environment, significantly improving the work efficiency of security personnel in network security threat detection and response.

[0112] 4. The present invention proposes a knowledge graph KG in the field of cyberspace security to enhance LLM, and integrates the dynamic knowledge graph in the field of network security into the large language model LLM in real time to provide instant and updated security knowledge, so that LLM can utilize the latest security intelligence when providing mitigation recommendations.

[0113] See also Figure 2 , which shows a system for constructing a threat detection and response model based on a traceability graph in an embodiment of the present invention, and the system includes:

[0114] The data acquisition module 10 is used to acquire the original log events and construct an initial traceability graph according to the original log events;

[0115] The label propagation execution module 20 is used to define the data structure of each event in the initial traceability graph, and retrieve the first preset rule corresponding to each event from the preset label propagation table according to the data structure, and obtain the label value of each event according to the first preset rule;

[0116] The label attenuation execution module 30 is used to perform attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value, and obtain a corresponding detection strategy from a preset database according to the updated label value;

[0117] A labeling module 40 is used to obtain attack behavior information corresponding to each updated label value according to the detection strategy, and label the initial traceability graph according to the attack behavior information to obtain a final traceability graph;

[0118] The model training module 50 is used to construct a network security knowledge graph, and input the network security knowledge graph and the final traceability graph into the initial threat detection and response model for training to obtain the final threat detection and response model.

[0119] On the other hand, the present invention further proposes a readable storage medium on which one or more programs are stored, and when the program is executed by a processor, the above-mentioned method for constructing a threat detection and response model based on a tracing graph is implemented.

[0120] On the other hand, the present invention also proposes a computer device, including a memory and a processor, wherein the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so as to implement the above-mentioned method for constructing a threat detection and response model based on the traceability graph.

[0121] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0122] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0123] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0124] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A method for constructing a threat detection and response model based on a traceability graph, characterized in that: The method comprises: Obtaining original log events, and constructing an initial traceability graph based on the original log events; Define the data structure of each event in the initial traceability graph, retrieve the first preset rule corresponding to each event from the preset label propagation table according to the data structure, and obtain the label value of each event according to the first preset rule; Performing attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value, and acquiring a corresponding detection strategy from a preset database according to the updated label value; Acquire attack behavior information corresponding to each updated label value according to the detection strategy, and annotate the initial traceability graph according to the attack behavior information to obtain a final traceability graph; A network security knowledge graph is constructed, and the network security knowledge graph and the final traceability graph are input into an initial threat detection and response model for training to obtain a final threat detection and response model.

2. The method for constructing a threat detection and response model based on a traceability graph according to claim 1, characterized in that: The step of obtaining the original log event and constructing an initial traceability graph according to the original log event includes: The original log event includes at least one event, and information is extracted from the original log event to obtain at least one system entity corresponding to each event and the relationship between the system entities; An initial traceability graph is constructed with the system entities as edges and the relationships between the system entities as nodes.

3. The method for constructing a threat detection and response model based on a traceability graph according to claim 2, characterized in that: The step of defining the data structure of each event in the initial traceability graph, retrieving the first preset rule corresponding to each event from the preset label propagation table according to the data structure, and obtaining the label value of each event according to the first preset rule includes: The data structure includes an event type and an event subject type corresponding to the event type, the event type includes creation, reading, writing, loading, execution, and injection, and the event subject type includes benign subject, suspicious subject, and suspicious environment subject; If the event type is creation, the first preset rule under all event subject types is the same, specifically: x.dtag = s.dtag If the event type is read, the first preset rule under all event subject types is the same, specifically: s.dtag = min(s.dtag, x.dtag) If the event type is write, when the event subject type is benign subject, the first preset rule is: x.dtag=min(s.dtag+a b ,x.dtag); when the event subject type is a suspicious subject, the first preset rule is: x.dtag = min(s.dtag,x.dtag); when the event subject type is a suspicious environment subject, the first preset rule is x.dtag = min(s.dtag + a e ,x.dtag); If the event type is loading, the first preset rule under all event subject types is the same, specifically: s.stag=min(s.stag,x.itag), s.dtag=min(s.dtag,x.dtag); If the event type is execution, when the event subject type is a benign subject, the first preset rule is: s.stag=x.itag, s.dtag=<1.0,1.0>; when the event subject type is a suspicious subject, the first preset rule is: s.stag=min(x.itag,susp_env), s.dtag=min(s.dtag,x.dtag); when the event subject type is a suspicious environment subject, the first preset rule is: s.dtag==x.itag, s.dtag=min(s.dtag,x.dtag); If the event type is injection, the first preset rule under all event subject types is the same, specifically: s′.stag=min(s′.stag,s.itag), s′.dtag=min(s.dtag,s′.dtag); Among them, x.dtag represents the data tag value of the executed object under the corresponding event, s.dtag represents the data tag value of the entity under the corresponding event, s.stag represents the subject tag value of the entity under the corresponding event, s′.stag represents the data tag value injected into another entity under the corresponding event, and a b 、a e Both represent correction values, and x.itag represents the integrity label value of the executed object under the corresponding event.

4. The method for constructing a threat detection and response model based on a traceability graph according to claim 3 is characterized in that: The step of performing attenuation control on the label value according to the preset label attenuation rule to obtain an updated label value comprises: For all events whose event subject type is benign subject, label decay is performed according to the following formula: s″.dtag=max(s.dtag,d b ·s.dtag+(1-d b )·T qb ) For all events whose event subject type is suspicious environment subject, label decay is performed according to the following formula: s″.dtag=max(s.dtag,d e ·s.dtag+(1-d e )·T qe ) Among them, s″.dtag represents the updated tag value, d b ,d e They represent the attenuation coefficients under benign subjects and suspicious environment subjects, respectively, and their values ​​are between 0 and 1. qb and T qe They represent the expected label values ​​under benign subjects and suspicious environment subjects respectively.

5. The method for constructing a threat detection and response model based on a traceability graph according to claim 4 is characterized in that: The step of acquiring the corresponding detection strategy from the preset database according to the updated tag value includes: Obtain the integrity label value of the entity or the executed object according to the updated label value, and filter out the integrity label values ​​that meet the suspicious level from all integrity label values; The corresponding detection strategy information is obtained from a preset database according to the integrity tag value of the suspicious level, and the detection strategy information includes name information, description information and operation information.

6. The method for constructing a threat detection and response model based on a traceability graph according to claim 5, characterized in that: The steps of constructing a network security knowledge graph, and inputting the network security knowledge graph and the final traceability graph into an initial threat detection and response model for training to obtain a final threat detection and response model include: Collecting network security data, including attack patterns, vulnerability types, and attack strategies; A network security knowledge graph including relationships between entities is constructed based on the network security data.

7. A system for constructing a threat detection and response model based on a traceability graph, characterized in that: The system comprises: A data acquisition module is used to acquire original log events and construct an initial traceability graph based on the original log events; A label propagation execution module, used to define the data structure of each event in the initial traceability graph, and retrieve the first preset rule corresponding to each event from the preset label propagation table according to the data structure, and obtain the label value of each event according to the first preset rule; A label attenuation execution module, used to perform attenuation control on the label value according to a preset label attenuation rule to obtain an updated label value, and obtain a corresponding detection strategy from a preset database according to the updated label value; A labeling module, used to obtain attack behavior information corresponding to each updated label value according to the detection strategy, and label the initial traceability graph according to the attack behavior information to obtain a final traceability graph; The model training module is used to construct a network security knowledge graph, and input the network security knowledge graph and the final traceability graph into the initial threat detection and response model for training to obtain the final threat detection and response model.

8. The system for constructing a threat detection and response model based on a traceability graph according to claim 7, characterized in that: The data acquisition module is also used for: The original log event includes at least one event, and information is extracted from the original log event to obtain at least one system entity corresponding to each event and the relationship between the system entities; An initial traceability graph is constructed with the system entities as edges and the relationships between the system entities as nodes.

9. A readable storage medium, characterized in that: The readable storage medium stores one or more programs, which, when executed by a processor, implement the method for constructing a threat detection and response model based on a tracing graph as described in any one of claims 1 to 6.

10. A computer device comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor is used to execute a computer program stored in the memory, it implements the method for constructing a threat detection and response model based on a tracing graph as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Label influence degree calculation method and system based on exponential decay

    CN113806688A

  • Threat detection method and system based on traceability graph behavior semantic extraction

    CN116938587A

  • Attack detection method and system based on information flow tracking

    CN118368138A

  • Method and system for improving security threats detection in communication networks

    US20140223555A1

Cited By

  • Host intrusion detection confrontation purification method and device based on link confrontation recognition

    CN119402300A

  • Network security data tracing method and device, medium and program product

    CN119922021A