Cloud computer abnormal behavior detection model training method and cloud computer abnormal behavior detection method

By determining the abnormal coefficients and properties of abnormal behavior in cloud computers, collecting and labeling data, and training multimodal large models, the problem of low detection accuracy of existing cloud computer detection tools is solved, and more efficient abnormal behavior detection is achieved.

CN118916745APending Publication Date: 2024-11-08CHINA TELECOM CORP LTD TECHNOLOGY INNOVATION CENTER +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411077262.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing cloud computer detection tools have limited detection range, and there are missed or false alarms, resulting in low detection accuracy of abnormal behavior of cloud computers.

Method used

By determining the anomaly coefficients and properties of each abnormal behavior in each application scenario, collecting and annotating abnormal behavior data, building a training set, training the initial multimodal large model, obtaining a cloud computer abnormal behavior detection model to improve detection accuracy.

Benefits of technology

It improves the accuracy of abnormal behavior detection of cloud computers, is suitable for a variety of application scenarios, reduces false alarms and missed reports, and enhances the effectiveness of operation and maintenance management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916745A_ABST
    Figure CN118916745A_ABST
Patent Text Reader

Abstract

The invention relates to a training method of a cloud computer abnormal behavior detection model, a cloud computer abnormal behavior detection method and device, computer equipment and a storage medium. The method comprises the following steps: determining an abnormal coefficient of each abnormal behavior in each application scene; for any abnormal behavior, determining the abnormal property of the abnormal behavior in each application scene based on the abnormal coefficient of the abnormal behavior in each application scene; for any application scene, abnormal behavior data collection and labeling are carried out according to the abnormal behavior with the abnormal property being abnormal in the application scene, and sample data in the application scene are obtained; and constructing a training set according to the sample data in each application scene, and training the initial multi-modal large model based on the training set to obtain a cloud computer abnormal behavior detection model. By adopting the method, the accuracy of abnormal behavior detection of the cloud computer can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of network security technology, and in particular to a training method for a cloud computer abnormal behavior detection model and a cloud computer abnormal behavior detection method, device, computer equipment and storage medium. Background Art

[0002] Cloud computing is a virtual computer service based on cloud computing technology. It migrates the computing, storage and application services of traditional computers to the cloud, allowing users to access and use these services at any time, anywhere and on any device through a network connection. With its elasticity, scalability and ease of use, cloud computing is widely used in mobile office, education and teaching and other fields.

[0003] Cloud computers optimize users' collaborative work and usage experience, but while users enjoy the convenience brought by cloud computers, they also have improper operations. The detection and processing of these behaviors puts higher requirements on the operation and maintenance management of cloud computers.

[0004] Currently, the operation and maintenance management of cloud computers mainly focuses on monitoring and alarming, including using detection tools to analyze the operating status, deploy software and update patches. After receiving abnormal alarms, users can decide the subsequent processing procedures by themselves. However, the detection range of traditional cloud computer detection tools is limited, and there are cases of missed reports or false reports, which means that the current detection accuracy of abnormal behavior of cloud computers is low. Summary of the invention

[0005] The embodiments of the present application provide a method for training a cloud computer abnormal behavior detection model and a method, device, computer equipment and storage medium for detecting abnormal behavior of a cloud computer, which can improve the accuracy of abnormal behavior detection of a cloud computer.

[0006] A method for training a cloud computer abnormal behavior detection model, the method comprising:

[0007] Determine the abnormal coefficient of each abnormal behavior in each application scenario;

[0008] For any of the abnormal behaviors, determining the abnormal nature of the abnormal behavior in each of the application scenarios based on the abnormal coefficient of the abnormal behavior in each of the application scenarios;

[0009] For any of the application scenarios, abnormal behavior data is collected and labeled according to the abnormal behavior with abnormal properties in the application scenario to obtain sample data in the application scenario;

[0010] A training set is constructed according to the sample data in each of the application scenarios, and the initial multimodal large model is trained based on the training set to obtain a cloud computer abnormal behavior detection model.

[0011] In one embodiment, determining the abnormal coefficient of each abnormal behavior in each application scenario includes:

[0012] In the target application scenario, for any abnormal category, the frequency of each abnormal behavior corresponding to the abnormal category occurring within a preset time period is counted respectively;

[0013] Determining a target frequency from the frequencies of occurrence of each of the abnormal behaviors in the abnormal category;

[0014] According to the frequency of occurrence of each abnormal behavior and the target frequency, an abnormal coefficient of each abnormal behavior in the target application scenario is determined, and the target application scenario is any application scenario among the application scenarios.

[0015] In one embodiment, determining the abnormal coefficient of each abnormal behavior in the target application scenario according to the frequency of occurrence of each abnormal behavior and the target frequency respectively includes:

[0016] For any of the abnormal behaviors, a ratio of the abnormal behavior to the target frequency is used as an abnormal coefficient of the abnormal behavior in the target application scenario.

[0017] In one embodiment, determining the abnormal nature of the abnormal behavior in each of the application scenarios based on the abnormal coefficient of the abnormal behavior in each of the application scenarios includes:

[0018] For a target application scenario, when the abnormal coefficient of the abnormal behavior in the target application scenario is greater than or equal to the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is abnormal; or,

[0019] When the abnormal coefficient of the abnormal behavior in the target application scenario is less than the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is normal;

[0020] The target application scenario is any one of the application scenarios.

[0021] In one embodiment, the abnormal behavior data collection and labeling according to the abnormal behavior with abnormal properties in the application scenario to obtain sample data in the application scenario includes:

[0022] For a target application scenario, obtaining a target abnormal behavior with abnormal properties in the target application scenario;

[0023] Based on each of the target abnormal behaviors, application behavior data is collected in a target application scenario, and each of the collected application behavior data is annotated based on the target abnormal behavior to obtain sample data in the target application scenario;

[0024] The target application scenario is any one of the application scenarios.

[0025] In one embodiment, the collected application behavior data are labeled based on the target abnormal behavior to obtain sample data in the target application scenario, including:

[0026] For any of the application behavior data, extract multimodal features of the application behavior data to obtain image features, audio features and text features, and fuse the image features, audio features and text features to obtain multimodal features corresponding to the application behavior data;

[0027] According to the target abnormal behavior corresponding to the application behavior data, the multimodal features corresponding to the application behavior data, and the abnormality resolution strategy, sample data under the target application scenario is constructed.

[0028] In one embodiment, the training of the initial multimodal large model based on the training set to obtain a cloud computer abnormal behavior detection model includes:

[0029] Performing instruction fine-tuning on the initial multimodal large model based on the training set to obtain the fine-tuned initial multimodal large model;

[0030] The fine-tuned initial multimodal large model is enhanced with training based on the training set to obtain a cloud computer abnormal behavior detection model.

[0031] A method for detecting abnormal behavior of a cloud computer, the method comprising:

[0032] In response to an abnormality detection operation on a target cloud computer, obtaining data to be detected of the target cloud computer;

[0033] Calling the cloud computer abnormal behavior detection model through the target interface, performing abnormal behavior detection on the data to be detected of the target cloud computer, and obtaining an abnormal detection result;

[0034] The cloud computer abnormal behavior detection model is trained using any of the aforementioned cloud computer abnormal behavior detection model training methods.

[0035] In one of the embodiments, the abnormal detection result includes a first detection result or a second detection result, the first detection result indicates that there is no abnormal behavior, and the second detection result includes the abnormal behavior and the abnormal handling strategy.

[0036] In one embodiment, the process of the cloud computer abnormal behavior detection model performing abnormal behavior detection on the data to be detected of the target cloud computer includes:

[0037] Performing multimodal feature extraction on the data to be detected to obtain image features, audio features and text features, and fusing the image features, audio features and text features to obtain multimodal features corresponding to the data to be detected;

[0038] The multimodal features corresponding to the data to be detected are input into the cloud computer abnormal behavior detection model to perform abnormal behavior detection and obtain an abnormal detection result.

[0039] A training device for a cloud computer abnormal behavior detection model, the device comprising:

[0040] A first determination unit, used to determine an abnormality coefficient of each abnormal behavior in each application scenario;

[0041] A second determination unit is configured to determine, for any of the abnormal behaviors, the abnormal nature of the abnormal behavior in each of the application scenarios based on the abnormal coefficient of the abnormal behavior in each of the application scenarios;

[0042] A sample processing unit, for collecting and marking abnormal behavior data according to the abnormal behavior with abnormal nature in any of the application scenarios, to obtain sample data in the application scenario;

[0043] The training unit is used to construct a training set according to the sample data in each of the application scenarios, and train the initial multimodal large model based on the training set to obtain a cloud computer abnormal behavior detection model.

[0044] In one embodiment, the first determining unit is further configured to:

[0045] In the target application scenario, for any abnormal category, the frequency of each abnormal behavior corresponding to the abnormal category occurring within a preset time period is counted respectively;

[0046] Determining a target frequency from the frequencies of occurrence of each of the abnormal behaviors in the abnormal category;

[0047] According to the frequency of occurrence of each abnormal behavior and the target frequency, an abnormal coefficient of each abnormal behavior in the target application scenario is determined, and the target application scenario is any application scenario among the application scenarios.

[0048] In one embodiment, the first determining unit is further configured to:

[0049] For any of the abnormal behaviors, a ratio of the abnormal behavior to the target frequency is used as an abnormal coefficient of the abnormal behavior in the target application scenario.

[0050] In one embodiment, the second determining unit is further configured to:

[0051] For a target application scenario, when the abnormal coefficient of the abnormal behavior in the target application scenario is greater than or equal to the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is abnormal; or,

[0052] When the abnormal coefficient of the abnormal behavior in the target application scenario is less than the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is normal;

[0053] The target application scenario is any one of the application scenarios.

[0054] In one embodiment, the sample processing unit is further used for:

[0055] For a target application scenario, obtaining a target abnormal behavior with abnormal properties in the target application scenario;

[0056] Based on each of the target abnormal behaviors, application behavior data is collected in a target application scenario, and each of the collected application behavior data is annotated based on the target abnormal behavior to obtain sample data in the target application scenario;

[0057] The target application scenario is any one of the application scenarios.

[0058] In one embodiment, the sample processing unit is further used for:

[0059] For any of the application behavior data, extract multimodal features of the application behavior data to obtain image features, audio features and text features, and fuse the image features, audio features and text features to obtain multimodal features corresponding to the application behavior data;

[0060] According to the target abnormal behavior corresponding to the application behavior data, the multimodal features corresponding to the application behavior data, and the abnormality resolution strategy, sample data under the target application scenario is constructed.

[0061] In one embodiment, the training unit is further used for:

[0062] Performing instruction fine-tuning on the initial multimodal large model based on the training set to obtain the fine-tuned initial multimodal large model;

[0063] The fine-tuned initial multimodal large model is enhanced with training based on the training set to obtain a cloud computer abnormal behavior detection model.

[0064] A cloud computer abnormal behavior detection device, the device comprising:

[0065] an acquisition unit, configured to acquire the to-be-detected data of the target cloud computer in response to an abnormality detection operation on the target cloud computer;

[0066] A detection unit, configured to call a cloud computer abnormal behavior detection model through a target interface, perform abnormal behavior detection on the data to be detected of the target cloud computer, and obtain an abnormal detection result;

[0067] The cloud computer abnormal behavior detection model is trained using any of the cloud computer abnormal behavior detection model training methods described above.

[0068] In one of the embodiments, the abnormal detection result includes a first detection result or a second detection result, the first detection result indicates that there is no abnormal behavior, and the second detection result includes the abnormal behavior and the abnormal handling strategy.

[0069] In one embodiment, the process of the cloud computer abnormal behavior detection model performing abnormal behavior detection on the data to be detected of the target cloud computer includes:

[0070] Performing multimodal feature extraction on the data to be detected to obtain image features, audio features and text features, and fusing the image features, audio features and text features to obtain multimodal features corresponding to the data to be detected;

[0071] The multimodal features corresponding to the data to be detected are input into the cloud computer abnormal behavior detection model to perform abnormal behavior detection and obtain an abnormal detection result.

[0072] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, any one of the above-mentioned methods for training a cloud computer abnormal behavior detection model or methods for detecting abnormal behavior of a cloud computer is implemented.

[0073] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for training a cloud computer abnormal behavior detection model or methods for detecting abnormal behavior of a cloud computer.

[0074] A computer program product includes a computer program, which, when executed by a processor, implements any of the above-mentioned methods for training a cloud computer abnormal behavior detection model or methods for detecting abnormal behavior of a cloud computer.

[0075] The training method of the cloud computer abnormal behavior detection model and the detection method, device, computer equipment, storage medium and computer program product of the cloud computer abnormal behavior can determine the abnormal coefficient of each abnormal behavior in each application scenario, and determine the abnormal nature of the abnormal behavior in each application scenario based on the abnormal coefficient of each abnormal behavior in each application scenario. For any application scenario, abnormal behavior data is collected and labeled according to the abnormal behavior with abnormal nature in the application scenario, sample data in the application scenario is obtained, and a training set is constructed according to the sample data in each application scenario, and the initial multimodal large model is trained based on the training set to obtain a cloud computer abnormal behavior detection model. Based on the training method of the cloud computer abnormal behavior detection model provided by the embodiment of the present disclosure and the detection method, device, computer equipment, storage medium and computer program product of the cloud computer abnormal behavior, the cloud computer abnormal behavior detection model can be trained to detect abnormal behavior of the cloud computer, and the abnormal behavior in each application scenario is defined by the abnormal coefficient during training, so that a cloud computer abnormal behavior detection model suitable for a variety of application scenarios can be trained, which can improve the accuracy of abnormal behavior detection of cloud computers. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A flowchart of a method for training a cloud computer abnormal behavior detection model in one embodiment;

[0077] Figure 2 is a flow chart of step 102 in one embodiment;

[0078] Figure 3 is a flow chart of step 106 in one embodiment;

[0079] Figure 4 is a flow chart of a method for detecting abnormal behavior of a cloud computer in another embodiment;

[0080] Figure 5 A schematic diagram of a method for detecting abnormal behavior of a cloud computer in another embodiment;

[0081] Figure 6 A schematic diagram of a method for detecting abnormal behavior of a cloud computer in another embodiment;

[0082] Figure 7 It is a structural block diagram of a training device for a cloud computer abnormal behavior detection model in one embodiment;

[0083] Figure 8is a structural block diagram of a device for detecting abnormal behavior of a cloud computer in another embodiment;

[0084] Fig. 9 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0086] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0087] In one embodiment, Figure 1 As shown, a training method for a cloud computer abnormal behavior detection model is provided. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. The following steps 102 to 108 are included, wherein:

[0088] Step 102: determine the abnormal coefficient of each abnormal behavior in each application scenario.

[0089] In the disclosed embodiments, each abnormal behavior in each application scenario can be counted in advance. Since the cloud computer can serve a variety of application scenarios, the abnormal behavior can be counted in each application scenario. Exemplarily, based on the historical service records of the cloud computer service platform, the detection tool operation log, the system warning, and the professional information of each application field, 7 abnormal categories can be statistically determined, and each category corresponds to at least one abnormal behavior. Exemplarily, the abnormal behaviors obtained by statistics are shown in Table 1 below.

[0090] Table 1

[0091]

[0092] However, the definition scope of each abnormal behavior is different in each application scenario, that is, the same abnormal behavior may have different abnormal properties in different application scenarios. For example, taking abnormal behavior A as an example, it is abnormal in application scenario 1, but not abnormal in application scenario 2. The abnormal coefficient is a parameter used to define the abnormal nature of the abnormal behavior. Exemplarily, the frequency of occurrence of each abnormal behavior in different application scenarios is different. Therefore, the abnormal coefficient corresponding to the abnormal behavior in each business scenario can be calculated according to the frequency of occurrence of each abnormal behavior in each business scenario.

[0093] In an exemplary embodiment, referring to Figure 2 As shown, in step 102, determining the abnormal coefficient of each abnormal behavior in each application scenario includes the following steps 202 to 206, wherein:

[0094] Step 202, in the target application scenario, for any abnormal category, count the frequency of each abnormal behavior corresponding to the abnormal category occurring within a preset time period;

[0095] Step 204, determining a target frequency from the frequency of occurrence of each abnormal behavior in the abnormal category;

[0096] Step 206 , determining an abnormal coefficient of each abnormal behavior in a target application scenario according to the frequency of occurrence of each abnormal behavior and the target frequency, wherein the target application scenario is any application scenario among the application scenarios.

[0097] In the embodiments of the present disclosure, any application scenario is taken as an example of the target application scenario to illustrate the embodiments of the present disclosure. Abnormal behaviors are divided into categories, and there are multiple abnormal categories. For any abnormal category, the frequency of each abnormal behavior corresponding to the abnormal category in the target application scenario within a preset time period can be counted. For example: for the abnormal category of abnormal network traffic, the frequency of launching DDos (Distributed denial of service attack) attacks in the target application scenario within a preset time period, and the frequency of launching SQL (Structured Query Language) injection attacks within a preset time period can be determined respectively, and the target frequency is determined from the two frequencies. Exemplarily, the average of the frequencies of occurrence of each abnormal behavior can be used as the target frequency, or the maximum value of the frequencies of occurrence of each abnormal behavior can be used as the target frequency.

[0098] After determining the target frequency, the abnormal coefficient of the abnormal behavior in the target application scenario can be determined based on the frequency of occurrence of the abnormal behavior and the target frequency. In an exemplary embodiment, determining the abnormal coefficient of each abnormal behavior in the target application scenario based on the frequency of occurrence of each abnormal behavior and the target frequency can include: for any abnormal behavior, taking the ratio of the abnormal behavior to the target frequency as the abnormal coefficient of the abnormal behavior in the target application scenario.

[0099] In this way, the abnormal coefficient of the abnormal behavior in the target application scenario is obtained, and by analogy, the abnormal coefficients of all abnormal behaviors in the target application scenario can be obtained; and by analogy, the abnormal coefficients corresponding to all abnormal behaviors in each application scenario can be obtained.

[0100] Step 104 , for any abnormal behavior, determine the abnormal nature of the abnormal behavior in each application scenario based on the abnormal coefficient of the abnormal behavior in each application scenario.

[0101] In the embodiment of the present disclosure, an abnormal behavior is taken as an example to introduce the abnormal property definition process. The abnormal coefficient of the abnormal behavior in each abnormal scenario can be obtained, and the abnormal property corresponding to each application scenario can be determined based on each abnormal coefficient.

[0102] In an exemplary embodiment, determining the abnormal nature of the abnormal behavior in each application scenario based on the abnormal coefficient of the abnormal behavior in each application scenario includes:

[0103] For the target application scenario, when the abnormality coefficient of the abnormal behavior in the target application scenario is greater than or equal to the abnormality threshold of the target application scenario, the abnormal nature of the abnormal behavior in the target application scenario is determined to be abnormal; or, when the abnormality coefficient of the abnormal behavior in the target application scenario is less than the abnormality threshold of the target application scenario, the abnormal nature of the abnormal behavior in the target application scenario is determined to be normal; wherein the target application scenario is any application scenario among the application scenarios.

[0104] In the embodiment of the present disclosure, each application scenario is respectively pre-set with a corresponding abnormal threshold. Taking one of the application scenarios as a target application scenario as an example, the abnormal threshold of each target application scenario can be obtained. When the abnormal coefficient of the abnormal behavior in the target application scenario is greater than or equal to the abnormal threshold of the target application scenario, it can be determined that the abnormal behavior is indeed regarded as abnormal behavior in the target application scenario, and the abnormal nature of the abnormal behavior in the target application scenario can be determined to be abnormal; conversely, when the abnormal coefficient of the abnormal behavior in the target application scenario is less than the abnormal threshold of the target application scenario, it can be determined that the abnormal behavior is not regarded as abnormal behavior in the target application scenario, and the abnormal nature of the abnormal behavior in the target application scenario can be determined to be normal.

[0105] Step 106, for any application scenario, abnormal behavior data is collected and labeled according to abnormal behaviors with abnormal properties in the application scenario to obtain sample data in the application scenario.

[0106] In the embodiments of the present disclosure, after determining the abnormal nature of each abnormal behavior in each application scenario, abnormal behavior data can be collected and labeled according to the abnormal behavior with abnormal nature in each application scenario to obtain sample data in the application scenario. Exemplarily, taking any application scenario as a target application scenario, target abnormal behavior with abnormal nature in the target application scenario can be obtained, and abnormal behavior data corresponding to each target abnormal behavior can be collected in the target application scenario. After labeling the collected abnormal behavior data based on the target abnormal behavior, sample data in the target application scenario can be obtained, and sample data in each application scenario can be obtained by analogy.

[0107] In an exemplary embodiment, referring to Figure 3 As shown, in step 106, abnormal behavior data collection and labeling are performed according to abnormal behaviors with abnormal properties in the application scenario to obtain sample data in the application scenario, which may include the following steps 302 to 304, wherein:

[0108] Step 302, for the target application scenario, obtaining target abnormal behavior with abnormal abnormality in the target application scenario;

[0109] Step 304, based on each target abnormal behavior, application behavior data is collected in the target application scenario, and each collected application behavior data is labeled based on the target abnormal behavior to obtain sample data under the target application scenario; wherein the target application scenario is any application scenario among the application scenarios.

[0110] For example, taking the abnormal behaviors including abnormal behaviors a, abnormal behaviors b, abnormal behaviors c, and abnormal behaviors d as an example, wherein the abnormal properties corresponding to abnormal behaviors a, abnormal behaviors b, and abnormal behaviors c in application scenario 1 are abnormal, and the abnormal properties corresponding to abnormal behaviors a, abnormal behaviors b, abnormal behaviors c, and abnormal behaviors d in application scenario 2 are abnormal, then in application scenario 1, abnormal behavior data collection can be performed for abnormal behaviors a, abnormal behaviors b, and abnormal behaviors c respectively, and the collected abnormal behavior data can be labeled with the corresponding abnormal behaviors respectively, so as to obtain sample data under application scenario 1, and the sample data can be used to train a model to detect abnormal behaviors in application scenario 1; in application scenario 2, abnormal behavior data collection can be performed for abnormal behaviors a, abnormal behaviors b, abnormal behaviors c, and abnormal behaviors d respectively, and the collected abnormal behavior data can be labeled with the corresponding abnormal behaviors respectively, so as to obtain sample data under target application scenario 2, and the sample data can be used to train a model to detect abnormal behaviors in application scenario 2.

[0111] In an exemplary embodiment, the collected application behavior data is annotated based on the target abnormal behavior to obtain sample data in the target application scenario, which may include:

[0112] For any application behavior data, multimodal feature extraction is performed on the application behavior data to obtain image features, audio features and text features, and the image features, audio features and text features are fused to obtain the multimodal features corresponding to the application behavior data; based on the target abnormal behavior corresponding to the application behavior data, the multimodal features corresponding to the application behavior data and the abnormality resolution strategy, sample data under the target application scenario is constructed.

[0113] In the embodiments of the present disclosure, for any application scenario, in the application scenario, the cloud computer operation log or the video recorded during the operation of the cloud computer can be used as the application behavior data. And the operation information can be extracted from the application behavior data such as the cloud computer operation log or the video recorded during the operation of the cloud computer, including at least user login account information, login time information, resource package interaction information, operation frequency and other information, and the sequence of user behavior changes over time can be recorded.

[0114] Exemplarily, the collected application behavior data can be desensitized to remove privacy information that is strongly related to the user, followed by data cleaning, denoising, and data format conversion, including: adjusting image data to the input size required by the model and regularizing it, adjusting the sampling rate of audio data to the standard required by the model and converting it to a feature representation, performing word segmentation on text data, and finally performing data annotation to label each type of data.

[0115] In the feature conversion process, it can include multimodal feature extraction of application behavior data to obtain image features, audio features and text features, and fuse the image features, audio features and text features to obtain multimodal features corresponding to the application behavior data. The multimodal features may include login time, operation duration, keyboard and mouse movement trajectory, page layout, access frequency, performance indicators, audio and other key features to characterize the target abnormal behavior. Feature mapping is then performed to form a training data triple (sample data), expressed as (target abnormal behavior, multimodal features, abnormal solution strategy).

[0116] Step 108, construct a training set according to the sample data in each application scenario, and train the initial multimodal large model based on the training set to obtain a cloud computer abnormal behavior detection model.

[0117] In the disclosed embodiment, after obtaining sample data in each application scenario, a knowledge base can be constructed based on the sample data in each application scenario, and the knowledge base can be divided into a training set and a verification set. The initial multimodal large model is trained based on the training set, and after obtaining the cloud computer abnormal behavior detection model, the model performance of the cloud computer abnormal behavior detection model is verified based on the verification set.

[0118] For example, after training the cloud computer abnormal behavior detection model based on the sample data of application scenario 1 and application scenario 2, in application scenario 1, abnormal behaviors a, abnormal behaviors b, and abnormal behaviors c in application scenario 1 can be detected, and in application scenario 2, abnormal behaviors a, abnormal behaviors b, abnormal behaviors c, and abnormal behaviors d in application scenario 2 can be detected, that is, it is suitable for accurate abnormal behavior detection in different application scenarios.

[0119] In an exemplary embodiment, training the initial multimodal large model based on the training set to obtain a cloud computer abnormal behavior detection model may include:

[0120] Fine-tune the initial multimodal large model based on the training set to obtain the fine-tuned initial multimodal large model;

[0121] Based on the training set, the fine-tuned initial multimodal large model is enhanced and trained to obtain a cloud computer abnormal behavior detection model.

[0122] In the disclosed embodiment, the cloud computer abnormal behavior detection model adopts a multimodal large model. The training process of the cloud computer abnormal behavior detection model includes two rounds of training: instruction fine-tuning training and enhanced training. Among them, instruction fine-tuning is the first round of training. For various types of abnormal behaviors, the multimodal instruction fine-tuning design includes the joint input of image-text and audio-text instructions, so that the cloud computer abnormal behavior detection model can learn the information association and fusion between different modalities. The customized multimodal fine-tuning instructions include but are not limited to the following instructions: "Briefly describe the image", "Provide a brief description of the given image", "Please give a concise explanation of the picture", "Carefully examine the image and share its details", "Describe the given image in detail", "Can you decompose the elements of the image in a detailed way?", "Describe the audio you hear in a few words", "What information do you hear in this audio?", etc.

[0123] During the instruction fine-tuning process, multiple initial multimodal large models can be trained at the same time. For example, BILP-2, MiniGPT-4, VisualGLM-6B and other initial multimodal large models can be trained. After the training is fine-tuned, the sample data in the validation set is used to evaluate the accuracy and reliability of each initial multimodal large model. The multimodal large model with better performance is compared based on the accuracy, precision, recall, and F1 value (a comprehensive indicator of precision and recall). The initial multimodal large model after fine-tuning is selected for the second round of training.

[0124] Enhanced training is used as the second round of training. The enhanced training process may include data enhancement, including: data enhancement from image data, audio data, text data, and multimodal data, so as to improve the generalization ability and robustness of the model. The enhancement method can be referred to as shown in Table 2 below:

[0125] Table 2

[0126]

[0127] After data enhancement is performed on the sample data in the training set, the fine-tuned initial multimodal large model is further trained based on the enhanced sample data to obtain the final cloud computer abnormal behavior detection model, which is used to detect abnormal behavior of cloud computers.

[0128] After obtaining the cloud computer abnormal behavior detection model, it can be deployed on the cloud server to provide abnormal behavior detection services to cloud computer users through a unified interface.

[0129] The training method of the above-mentioned cloud computer abnormal behavior detection model can determine the abnormal coefficient of each abnormal behavior in each application scenario, and determine the abnormal nature of the abnormal behavior in each application scenario based on the abnormal coefficient of each abnormal behavior in each application scenario. For any application scenario, abnormal behavior data is collected and labeled according to the abnormal behavior with abnormal nature in the application scenario to obtain sample data in the application scenario, and a training set is constructed based on the sample data in each application scenario. The initial multimodal large model is trained based on the training set to obtain a cloud computer abnormal behavior detection model. Based on the training method of the cloud computer abnormal behavior detection model provided by the embodiment of the present disclosure, the cloud computer abnormal behavior detection model can be trained to detect abnormal behavior of the cloud computer, and the abnormal behavior in each application scenario is defined by the abnormal coefficient during training, so that a cloud computer abnormal behavior detection model suitable for a variety of application scenarios can be trained, which can improve the accuracy of abnormal behavior detection of cloud computers.

[0130] In one embodiment, Figure 4 As shown, a method for detecting abnormal behavior of a cloud computer is provided. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. The following steps 402 to 404 are included, wherein:

[0131] Step 402, in response to the abnormality detection operation on the target cloud computer, obtaining the data to be detected of the target cloud computer;

[0132] Step 404, calling the cloud computer abnormal behavior detection model through the target interface, performing abnormal behavior detection on the data to be detected of the target cloud computer, and obtaining an abnormal detection result;

[0133] The cloud computer abnormal behavior detection model is trained using any of the aforementioned cloud computer abnormal behavior detection model training methods.

[0134] In the disclosed embodiment, the user end can call the cloud computer abnormal behavior detection model deployed in the cloud server through the target interface to perform abnormal behavior detection on the target cloud computer. Exemplarily, the target cloud computer can periodically collect user behavior data. When receiving the abnormal detection instruction, the collected user behavior data can be used as the data to be detected, and the cloud computer abnormal behavior detection model deployed in the cloud server can be called through the target interface to perform abnormal behavior detection on the data to be detected to obtain the abnormal detection result. The training process of the cloud computer abnormal behavior detection model can refer to the relevant description of the aforementioned embodiment, and will not be repeated in the disclosed embodiment.

[0135] In an exemplary embodiment, the abnormal detection result includes a first detection result or a second detection result. The first detection result indicates that there is no abnormal behavior, and the second detection result includes abnormal behavior and an abnormal handling strategy. In this way, when a user receives an abnormal alarm, he or she can handle the exception based on the abnormal handling strategy, thereby avoiding the problem of subjective operational errors of users due to individual differences in cognition.

[0136] For example, when detecting abnormal behavior of a target cloud computer, video data of the behavior to be detected can be obtained. The video data is usually collected when the cloud computer management platform detects that the CPU, memory and other resource usage rates of a cloud computer are high or the resource usage time is abnormal and issues an alert, or it is captured regularly with the user's consent.

[0137] In one embodiment, the process of the cloud computer abnormal behavior detection model performing abnormal behavior detection on the data to be detected of the target cloud computer may include:

[0138] Multimodal features are extracted from the data to be detected to obtain image features, audio features and text features, and the image features, audio features and text features are fused to obtain multimodal features corresponding to the data to be detected; the multimodal features corresponding to the data to be detected are input into the cloud computer abnormal behavior detection model to perform abnormal behavior detection and obtain abnormal detection results.

[0139] In the disclosed embodiment, the video is segmented into a multi-frame image sequence through a frame extraction module, and key frames are captured by motion detection-based frame extraction. For the key frame images, an image encoder is used to extract visual features; the sound stream of the video is segmented into multiple audio segments through an audio processing module, and an audio encoder is used to extract audio features; the text description of the corresponding image and audio is obtained through a text processing module, and a text encoder is used to extract text features related to abnormal behavior.

[0140] The time alignment of different modal features is achieved based on events, and then the different modal features are dynamically weighted based on the cross-modal attention mechanism, including: mapping the features of each modality to a Query-Key-Value vector, using the similarity between the Query and the Key of each modality to calculate the attention score, and then normalizing the attention score to a probability distribution through a function, and using the probability distribution to perform weighted summation on the Value of each modality to fuse multimodal features. This allows the model to automatically adjust the importance of each modality according to task requirements and input characteristics, thereby achieving multimodal feature fusion.

[0141] The cloud computer abnormal behavior detection model is called through the target interface to perform abnormal analysis on the multimodal features to obtain the abnormal detection result. If the abnormal detection result is the second detection result, a warning message is issued based on the abnormal behavior in the second detection result, and a corresponding abnormal handling strategy is provided to the cloud computer user and the management personnel, so that the cloud computer user and the management personnel can handle the abnormality based on the abnormal handling strategy; if the abnormal detection result is the first detection result, that is, it is determined that there is no abnormal behavior at present, then no reminder is required, and the cloud computer user behavior detection task is completed.

[0142] The above-mentioned cloud computer abnormal behavior detection method can detect cloud computer abnormal behavior based on the cloud computer abnormal behavior detection model. Since the cloud computer abnormal behavior detection model is trained by defining abnormal behaviors in various application scenarios through abnormal coefficients, a cloud computer abnormal behavior detection model suitable for various application scenarios can be trained, which can improve the accuracy of cloud computer abnormal behavior detection.

[0143] In order to enable those skilled in the art to better understand the embodiments of the present disclosure, the embodiments of the present disclosure are described below through specific examples.

[0144] Reference Figure 5 and Figure 6 As shown, a schematic diagram of the abnormal behavior detection process for cloud computers in the embodiment of the present disclosure is shown. It includes a model training process and an abnormality detection process. The model training process includes the abnormal behavior evaluation standard construction process and the multi-modal large model training process, specifically:

[0145] In the process of constructing abnormal behavior evaluation standards, abnormal categories can be pre-screened, including: determining abnormal behavior classification based on historical service records of cloud computer service providers, detection tool operation logs, system warnings and professional information of various application fields, a total of 7 major categories; enumerating users' abnormal behaviors, including: corresponding various detailed abnormal behaviors to relevant abnormal classifications; defining the abnormal coefficient of abnormal behavior of cloud computer users, including: calculating the abnormal coefficient of abnormal behavior for each application scenario, and obtaining an evaluation table for corresponding to the abnormal behavior differences between users in each scenario.

[0146] In the process of training a large multimodal model, data is first collected and labeled based on the evaluation table. This includes: obtaining the user's login IP, login time, resource package purchase status, operation frequency and other information from the operation log, recording / screenshotting the user's current page on the cloud computer at a fixed time with the user's permission, and recording the sequence of user behavior changes over time. Desensitize the data to remove identity information that is strongly related to the user, then clean and denoise the data, convert the data format, adjust the image data to the input size required by the model and regularize it, adjust the sampling rate of the audio data to the standard required by the model and convert it to feature representation, divide the text data into words, and finally label the data. According to business understanding, based on the behavior data, key features such as login time, operation duration, keyboard and mouse movement trajectory, page layout, access frequency, performance indicators, audio, etc. are extracted to form multimodal features that characterize the behavior, and then feature mapping is performed to form a training data triple, expressed as (abnormal behavior, multimodal features, solutions). A knowledge base is constructed based on each training data triplet, the knowledge base is divided into a training set and a validation set, and multimodal large model training is performed based on the training set.

[0147] During the training of the multimodal large model, the multimodal large model is fine-tuned through customized multimodal instructions, and the multimodal large model with better performance is selected for enhanced training based on accuracy, precision, recall, and F1 score comparison. During the enhanced training process, data enhancement is performed from the aspects of image data, audio data, text data, and multimodal data to improve the generalization ability and robustness of the model. After the cloud computer abnormal behavior detection model is trained, the cloud computer abnormal behavior detection model can be deployed in the cloud server to provide services to each cloud computer user through a unified interface.

[0148] During the anomaly detection process, the cloud computer abnormal behavior detection model can be called through a unified interface to perform anomaly detection, including: segmenting the behavior video to be detected into image sequences and audio segments, followed by corresponding text descriptions; using corresponding encoders to extract visual features, audio features, and text features for key frame images, audio, and text; aligning the features of each modality over time, and then dynamically weighting the features of different modalities based on the cross-modal attention mechanism to obtain multi-modal features, and judging whether there is abnormal behavior based on the fine-tuned cloud computer abnormal behavior detection model, and providing anomaly handling strategies.

[0149] The following is a further explanation of the present disclosure using two scenario examples:

[0150] Scenario 1: In the field of education and teaching, cloud computers are public resources, and the user mobility is high. It is easy to have abnormal behaviors such as illegal account sharing, non-compliant access, and script abuse. First, determine the abnormal coefficient of each type of behavior based on the business scenario characteristics in this field, and confirm that each cloud computer can use the cloud computer abnormal behavior detection model normally through a unified interface. Through the early warning information of the management platform, the cloud computer manager monitors that cloud computer A logs in during abnormal teaching time and uses too much network traffic. The cloud computer system obtains multiple videos of behaviors to be detected for further confirmation of abnormality. For the behavior video a1 to be detected, the frame extraction module uses OpenCV to divide the video into multiple frame image sequences. It is found that the cloud computer frequently accesses and refreshes the same URL, and the data is captured. The key frame is captured by frame extraction based on motion detection. The key frame image is scaled, grayed, and normalized, and then the image encoder is used to extract visual features. The audio processing module uses ffmpeg to divide the sound stream of the video into multiple audio segments, and the audio encoder is used to extract audio features, including Mel frequency cepstral coefficients. The text processing module will obtain the text description of the corresponding image and audio, and use the text encoder to extract text features related to abnormal behavior, including script codes, etc.

[0151] The time alignment of different modal features is achieved based on events, and then the different modal features are dynamically weighted based on the cross-modal attention mechanism. Specifically, the features of each modality are mapped to a Query-Key-Value vector, the similarity between the Query and the Key of each modality is used to calculate the attention score, and then the simple attention score is normalized into a probability distribution through a function, and the Value of each modality is weighted and summed using the probability distribution. After such calculation, the model can automatically adjust the importance of each modality according to the task requirements and input characteristics, thereby realizing multi-modal feature fusion. The abnormality of a1 is judged based on the cloud computer abnormal behavior detection model. The above process is used for the remaining behavior videos a2...an to be detected. Combined with the results of multiple behavior judgments, the model determines that the behavior of the cloud computer user is to write scripts for web crawling, which is an abnormal behavior. The cloud computer management platform notifies the management personnel and pops up a window to warn the current user of cloud computer A. The multi-modal large model provides relevant solutions, including limiting the network speed of cloud computer A, using security group rules, and forcing offline.

[0152] Scenario 2: In the government and enterprise office scenario, cloud computers are basically used by a single person, but they are prone to abnormal behaviors such as non-compliant access and data leakage. First, determine the abnormal coefficient of each type of behavior based on the characteristics of the business scenario in this field, and confirm that each cloud computer can normally use the fine-tuned multimodal large model through a unified interface. Reach an agreement with the user before deploying the cloud computer and regularly obtain the user's cloud computer usage record video. The cloud computer system regularly obtains the video of the behavior to be detected for further confirmation of abnormality. For the behavior video b1 to be detected, the frame extraction module uses OpenCV to divide the video into a multi-frame image sequence, and the frame extraction based on motion detection is used to capture the key frame. The key frame image is scaled, grayed, and normalized, and then the image encoder is used to extract visual features. The audio processing module uses ffmpeg to divide the sound stream of the video into multiple audio segments to obtain the speech audio, and the audio encoder is used to extract audio features, including Mel frequency cepstral coefficients. The text processing module will obtain the text description of the corresponding image and audio, and the text encoder is used to extract text features related to abnormal behavior, including subtitles, document text, etc.

[0153] The time alignment of different modal features is achieved based on events, and then the different modal features are dynamically weighted based on the cross-modal attention mechanism. Specifically, it includes mapping the features of each modality into a Query-Key-Value vector, using the similarity between the Query and the Key of each modality to calculate the attention score, and then normalizing the attention score into a probability distribution through a function, and using the probability distribution to perform weighted summation on the Value of each modality. After such calculations, the model can automatically adjust the importance of each modality according to task requirements and input characteristics, thereby achieving multimodal feature fusion. The abnormality of b1 is judged based on the fine-tuned multimodal large model. The above process is used for the remaining behavior videos b2...bn to be detected. Combined with the results of multiple behavior judgments, the model determines that the behavior of the cloud computer user is normal, and the cloud computer user behavior detection task is completed.

[0154] By adopting the cloud computer abnormal behavior detection method provided by the embodiment of the present disclosure, a cloud computer user abnormal behavior evaluation standard is established, and an abnormal coefficient is defined to describe the abnormal nature of the behavior, which is used to explain the differences in the definition of abnormal behavior in different business scenarios, so that the trained cloud computer abnormal behavior detection model can be applied to different business scenarios. And the model is trained by multimodal instruction fine-tuning and data enhancement methods, which can improve the model accuracy and cross-modal task execution ability, improve its generalization ability, accuracy and robustness; through the multimodal feature fusion method, the image, audio, and text data features are comprehensively considered to comprehensively determine the abnormality of cloud computer user behavior, breaking the singleness of the detection tool. And the model is deployed on the cloud server, and provides behavior detection and solution services to each cloud computer user through a unified interface, instead of installing detection tools in each cloud computer, so as to reduce the memory, CPU and other resources occupied by the cloud computer.

[0155] It should be understood that although Figure 1-6 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1-6 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0156] In one embodiment, Figure 7 As shown, a training device 700 for detecting abnormal behavior of a cloud computer is provided, comprising: a first determination unit 702, a second determination unit 704, a sample processing unit 706 and a training unit 708, wherein:

[0157] A first determining unit 702 is used to determine an abnormality coefficient of each abnormal behavior in each application scenario;

[0158] A second determining unit 704 is configured to determine, for any of the abnormal behaviors, the abnormal nature of the abnormal behavior in each of the application scenarios based on the abnormal coefficient of the abnormal behavior in each of the application scenarios;

[0159] The sample processing unit 706 is used for collecting and marking abnormal behavior data according to the abnormal behavior with abnormal nature in the application scenario for any of the application scenarios, so as to obtain sample data in the application scenario;

[0160] The training unit 708 is used to construct a training set according to the sample data in each of the application scenarios, and train the initial multimodal large model based on the training set to obtain a cloud computer abnormal behavior detection model.

[0161] In one embodiment, the first determining unit 702 is further configured to:

[0162] In the target application scenario, for any abnormal category, the frequency of each abnormal behavior corresponding to the abnormal category occurring within a preset time period is counted respectively;

[0163] Determining a target frequency from the frequencies of occurrence of each of the abnormal behaviors in the abnormal category;

[0164] According to the frequency of occurrence of each abnormal behavior and the target frequency, an abnormal coefficient of each abnormal behavior in the target application scenario is determined, and the target application scenario is any application scenario among the application scenarios.

[0165] In one embodiment, the first determining unit 702 is further configured to:

[0166] For any of the abnormal behaviors, a ratio of the abnormal behavior to the target frequency is used as an abnormal coefficient of the abnormal behavior in the target application scenario.

[0167] In one embodiment, the second determining unit 704 is further configured to:

[0168] For a target application scenario, when the abnormal coefficient of the abnormal behavior in the target application scenario is greater than or equal to the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is abnormal; or,

[0169] When the abnormal coefficient of the abnormal behavior in the target application scenario is less than the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is normal;

[0170] The target application scenario is any one of the application scenarios.

[0171] In one embodiment, the sample processing unit 706 is further configured to:

[0172] For a target application scenario, obtaining a target abnormal behavior with abnormal properties in the target application scenario;

[0173] Based on each of the target abnormal behaviors, application behavior data is collected in a target application scenario, and each of the collected application behavior data is annotated based on the target abnormal behavior to obtain sample data in the target application scenario;

[0174] The target application scenario is any one of the application scenarios.

[0175] In one embodiment, the sample processing unit 706 is further configured to:

[0176] For any of the application behavior data, extract multimodal features of the application behavior data to obtain image features, audio features and text features, and fuse the image features, audio features and text features to obtain multimodal features corresponding to the application behavior data;

[0177] According to the target abnormal behavior corresponding to the application behavior data, the multimodal features corresponding to the application behavior data, and the abnormality resolution strategy, sample data under the target application scenario is constructed.

[0178] In one embodiment, the training unit 708 is further configured to:

[0179] Performing instruction fine-tuning on the initial multimodal large model based on the training set to obtain the fine-tuned initial multimodal large model;

[0180] The fine-tuned initial multimodal large model is enhanced with training based on the training set to obtain a cloud computer abnormal behavior detection model.

[0181] Based on the training device of the cloud computer abnormal behavior detection model provided by the embodiment of the present disclosure, the cloud computer abnormal behavior detection model can be trained to perform cloud computer abnormal behavior detection, and the abnormal behavior in each application scenario is defined through the abnormal coefficient during training, so that a cloud computer abnormal behavior detection model suitable for a variety of application scenarios can be trained, which can improve the accuracy of cloud computer abnormal behavior detection.

[0182] In one embodiment, Figure 8 As shown, a cloud computer abnormal behavior detection 800 is provided, including: an acquisition unit 802 and a detection unit 804, wherein:

[0183] An acquisition unit 802 is used to acquire the to-be-detected data of the target cloud computer in response to an abnormality detection operation on the target cloud computer;

[0184] The detection unit 804 is used to call the cloud computer abnormal behavior detection model through the target interface, perform abnormal behavior detection on the data to be detected of the target cloud computer, and obtain an abnormal detection result;

[0185] The cloud computer abnormal behavior detection model is trained using any of the cloud computer abnormal behavior detection model training methods described above.

[0186] In one of the embodiments, the abnormal detection result includes a first detection result or a second detection result, the first detection result indicates that there is no abnormal behavior, and the second detection result includes the abnormal behavior and the abnormal handling strategy.

[0187] In one embodiment, the process of the cloud computer abnormal behavior detection model performing abnormal behavior detection on the data to be detected of the target cloud computer includes:

[0188] Performing multimodal feature extraction on the data to be detected to obtain image features, audio features and text features, and fusing the image features, audio features and text features to obtain multimodal features corresponding to the data to be detected;

[0189] The multimodal features corresponding to the data to be detected are input into the cloud computer abnormal behavior detection model to perform abnormal behavior detection and obtain an abnormal detection result.

[0190] The above-mentioned cloud computer abnormal behavior detection device can detect cloud computer abnormal behavior based on the cloud computer abnormal behavior detection model. Since the cloud computer abnormal behavior detection model is trained by defining abnormal behavior in each application scenario through abnormal coefficients, a cloud computer abnormal behavior detection model suitable for multiple application scenarios can be trained, which can improve the accuracy of cloud computer abnormal behavior detection.

[0191] Regarding the specific limitations of the training device for the cloud computer abnormal behavior detection model and the cloud computer abnormal behavior detection device, please refer to the limitations of the training method for the cloud computer abnormal behavior detection model and the cloud computer abnormal behavior detection method above, which will not be repeated here. Each module in the above-mentioned training device for the cloud computer abnormal behavior detection model and the cloud computer abnormal behavior detection device can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0192] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig. 9As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a training method for a cloud computer abnormal behavior detection model or a cloud computer abnormal behavior detection method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0193] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0194] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0195] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0196] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0197] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0198] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0199] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0200] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A training method for a cloud computer abnormal behavior detection model, characterized in that: The method comprises: Determine the abnormal coefficient of each abnormal behavior in each application scenario; For any of the abnormal behaviors, determining the abnormal nature of the abnormal behavior in each of the application scenarios based on the abnormal coefficient of the abnormal behavior in each of the application scenarios; For any of the application scenarios, abnormal behavior data is collected and labeled according to the abnormal behavior with abnormal properties in the application scenario to obtain sample data in the application scenario; A training set is constructed according to the sample data in each of the application scenarios, and the initial multimodal large model is trained based on the training set to obtain a cloud computer abnormal behavior detection model.

2. The method according to claim 1, characterized in that Determining the abnormal coefficient of each abnormal behavior in each application scenario includes: In the target application scenario, for any abnormal category, the frequency of each abnormal behavior corresponding to the abnormal category occurring within a preset time period is counted respectively; Determining a target frequency from the frequencies of occurrence of each of the abnormal behaviors in the abnormal category; According to the frequency of occurrence of each abnormal behavior and the target frequency, an abnormal coefficient of each abnormal behavior in the target application scenario is determined, and the target application scenario is any application scenario among the application scenarios.

3. The method according to claim 2, characterized in that The determining, according to the frequency of occurrence of each abnormal behavior and the target frequency, the abnormal coefficient of each abnormal behavior in the target application scenario includes: For any of the abnormal behaviors, a ratio of the abnormal behavior to the target frequency is used as an abnormal coefficient of the abnormal behavior in the target application scenario.

4. The method according to any one of claims 1 to 3, characterized in that The determining the abnormal nature of the abnormal behavior in each of the application scenarios based on the abnormal coefficient of the abnormal behavior in each of the application scenarios respectively includes: For a target application scenario, when the abnormal coefficient of the abnormal behavior in the target application scenario is greater than or equal to the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is abnormal; or, When the abnormal coefficient of the abnormal behavior in the target application scenario is less than the abnormal threshold of the target application scenario, determining that the abnormal nature of the abnormal behavior in the target application scenario is normal; The target application scenario is any one of the application scenarios.

5. The method according to claim 1, characterized in that The abnormal behavior data collection and labeling according to the abnormal behavior with abnormal properties in the application scenario to obtain sample data in the application scenario includes: For a target application scenario, obtaining a target abnormal behavior with abnormal properties in the target application scenario; Based on each of the target abnormal behaviors, application behavior data is collected in a target application scenario, and each of the collected application behavior data is annotated based on the target abnormal behavior to obtain sample data in the target application scenario; The target application scenario is any one of the application scenarios.

6. The method according to claim 5, characterized in that The step of labeling the collected application behavior data based on the target abnormal behavior to obtain sample data in the target application scenario includes: For any of the application behavior data, extract multimodal features of the application behavior data to obtain image features, audio features and text features, and fuse the image features, audio features and text features to obtain multimodal features corresponding to the application behavior data; According to the target abnormal behavior corresponding to the application behavior data, the multimodal features corresponding to the application behavior data, and the abnormality resolution strategy, sample data under the target application scenario is constructed.

7. The method according to claim 1 or 6, characterized in that: The initial multimodal large model is trained based on the training set to obtain a cloud computer abnormal behavior detection model, including: Performing instruction fine-tuning on the initial multimodal large model based on the training set to obtain the fine-tuned initial multimodal large model; The fine-tuned initial multimodal large model is enhanced with training based on the training set to obtain a cloud computer abnormal behavior detection model.

8. A method for detecting abnormal behavior of a cloud computer, characterized in that: The method comprises: In response to an abnormality detection operation on a target cloud computer, obtaining data to be detected of the target cloud computer; Calling the cloud computer abnormal behavior detection model through the target interface, performing abnormal behavior detection on the data to be detected of the target cloud computer, and obtaining an abnormal detection result; The cloud computer abnormal behavior detection model is trained using the cloud computer abnormal behavior detection model training method described in any one of claims 1 to 7.

9. The method according to claim 8, characterized in that The abnormal detection result includes a first detection result or a second detection result, the first detection result indicates that there is no abnormal behavior, and the second detection result includes abnormal behavior and an abnormal handling strategy.

10. The method according to claim 8 or 9, characterized in that: The process of the cloud computer abnormal behavior detection model performing abnormal behavior detection on the data to be detected of the target cloud computer includes: Performing multimodal feature extraction on the data to be detected to obtain image features, audio features and text features, and fusing the image features, audio features and text features to obtain multimodal features corresponding to the data to be detected; The multimodal features corresponding to the data to be detected are input into the cloud computer abnormal behavior detection model to perform abnormal behavior detection and obtain an abnormal detection result.

11. A training device for a cloud computer abnormal behavior detection model, characterized in that: The device comprises: A first determination unit, used to determine an abnormality coefficient of each abnormal behavior in each application scenario; A second determination unit is configured to determine, for any of the abnormal behaviors, the abnormal nature of the abnormal behavior in each of the application scenarios based on the abnormal coefficient of the abnormal behavior in each of the application scenarios; A sample processing unit, for collecting and marking abnormal behavior data according to the abnormal behavior with abnormal nature in any of the application scenarios, to obtain sample data in the application scenario; The training unit is used to construct a training set according to the sample data in each of the application scenarios, and train the initial multimodal large model based on the training set to obtain a cloud computer abnormal behavior detection model.

12. A cloud computer abnormal behavior detection device, characterized in that: The device comprises: an acquisition unit, configured to acquire the to-be-detected data of the target cloud computer in response to an abnormality detection operation on the target cloud computer; A detection unit, configured to call a cloud computer abnormal behavior detection model through a target interface, perform abnormal behavior detection on the data to be detected of the target cloud computer, and obtain an abnormal detection result; The cloud computer abnormal behavior detection model is trained using the cloud computer abnormal behavior detection model training method described in any one of claims 1 to 7.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 or 8 to 10 are implemented.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 or 8 to 10 are implemented.

15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 or 8 to 10 are implemented.