Attack path prediction method and related device
By converting the log data to be predicted into a target map and using the prediction model to determine the attack path, the low accuracy problem caused by relying on manual rules in the prior art is solved, and more efficient attack path prediction is achieved.
Patent Information
- Application Number
- CN202411291786.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-12-03
AI Technical Summary
The existing attack path prediction methods rely on manual rules and are difficult to consider all attack situations, resulting in low prediction accuracy.
By mapping the log data to be predicted into a target graph, converting it into nodes and edges, and using the prediction model (including feature conversion layer and classifier) to determine the attack path based on the node feature vector, the trained model can automatically learn the latest attack path.
The prediction accuracy of attack paths is improved, it is higher than that of methods that rely on manual rules, and the node relationship is more accurately handled through the graph attention mechanism.
Smart Images

Figure CN119071077A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present application relate to the field of neural networks, and in particular to an attack path prediction method and related devices. Background Art
[0002] With the continuous development of Internet technology, people's living and production standards are also constantly improving. With malicious intentions, cyber criminals will use Internet technology to launch cyber attacks on personal or corporate computer terminals. Cyber criminals will analyze the target computer terminal to find loopholes and weaknesses, and attack accordingly to achieve the purpose of stealing sensitive information, destroying the target terminal function or tampering with related data. In order to avoid the above problems, network security personnel have designed corresponding attack path prediction methods. As an important part of security defense, attack path prediction aims to restore the attack path through the logs on the terminal and traffic equipment, so as to strengthen the system defense accordingly and reduce the risk of attack and intrusion.
[0003] However, the existing solutions rely on manual rules to predict attack paths, and it is difficult for manually formulated rules to take into account all attack situations. Therefore, the current prediction method that relies on manual rules has a low prediction accuracy. Summary of the invention
[0004] The embodiments of the present application provide an attack path prediction method and related devices for improving the prediction accuracy of the attack path.
[0005] A first aspect of an embodiment of the present application provides an attack path prediction method, comprising:
[0006] Receiving log data to be predicted, and mapping the log data to be predicted into a target graph, wherein the target graph includes a plurality of nodes and edges, wherein the nodes are used to represent computer entities in the log data to be predicted, and the edges are used to represent logical relationships between computer entities;
[0007] Converting the text attribute data of each computer entity in the to-be-predicted log data into a first node feature vector of a corresponding node in the target graph;
[0008] The first node feature vector is input into a prediction model to determine an attack path in the log data to be predicted. The prediction model is used to determine an edge in the target graph that meets a preset condition as an attack path based on the first node feature vector.
[0009] Optionally, the prediction model includes a feature conversion layer and a classifier, and the step of inputting the first node feature vector into the prediction model to determine the attack path in the log data to be predicted includes:
[0010] Inputting the first node feature vector into the feature conversion layer to obtain a first edge feature vector of each edge in the target graph;
[0011] Inputting the first edge feature vector into the classifier to obtain a corresponding probability value of each edge;
[0012] The edge whose probability value is greater than or equal to a preset threshold is determined as the attack path in the log data to be predicted.
[0013] Optionally, inputting the first node feature vector to the feature conversion layer to obtain a first edge feature vector of each edge in the target graph includes:
[0014] Inputting the first node feature vector to the feature conversion layer to concatenate the first node feature vectors of every two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph;
[0015] or,
[0016] Inputting the first node feature vector to the feature conversion layer to add the first node feature vectors of every two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph;
[0017] or,
[0018] The first node feature vector is input into the feature conversion layer to weightedly fuse the first node feature vectors of every two nodes that are neighboring nodes in the target graph and the similarity of the two vectors to obtain the first edge feature vector of each edge in the target graph.
[0019] Optionally, the prediction model further includes a graph attention layer. After converting the text attribute data of each computer entity in the to-be-predicted log data into a first node feature vector of a corresponding node in the target graph, the method further includes:
[0020] Inputting the first node feature vector into a graph attention layer of the prediction model to determine the similarity between the first node feature vector of each node in the target graph and each first node feature vector of the remaining nodes in the target graph;
[0021] Performing weighted fusion on the similarity and the first node feature vector based on the graph attention layer to obtain a second node feature vector for each node in the target graph;
[0022] The step of inputting the first node feature vector to the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge in the target graph includes:
[0023] The two second node feature vectors of every two nodes that are neighboring nodes in the target graph are input into the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge between every two nodes that are neighboring nodes in the target graph.
[0024] Optionally, before receiving the log data to be predicted, the method further includes:
[0025] Acquire first training log data from a historical database;
[0026] Mapping the first training log data into a training graph, wherein the training graph includes nodes and edges, the nodes are used to represent computer entities in the training log data, and the edges are used to represent logical relationships between computer entities;
[0027] Converting the text attribute data of each computer entity in the training log data into a third node feature vector of a corresponding node in the training graph;
[0028] Inputting a plurality of the third node feature vectors into the initialized prediction model to obtain a predicted probability value of each edge between every two nodes output by the initialized prediction model, wherein the predicted probability value is used to characterize the predicted probability that each edge between every two nodes is an attack path;
[0029] Obtaining a training probability value input for each edge in the historical database, wherein the training probability value is used to characterize a true probability that each edge is an attack path;
[0030] The initialized prediction model is trained using the prediction probability value and the training probability value until the initialized prediction model converges to obtain the prediction model.
[0031] Optionally, after acquiring the first training log data from the historical database, the method further includes:
[0032] Determine multiple groups of repeated log data that are continuously triggered and have the same content except for timestamps in the first training log data;
[0033] Deleting all log data except the latest time point in the multiple sets of repeated log data to obtain second training log data;
[0034] Mapping the first training log data into a training graph includes:
[0035] The second training log data is mapped into the training graph.
[0036] Optionally, after deleting all log data except the latest time point in the multiple sets of duplicate log data to obtain the second training log data, the method further includes:
[0037] Adjusting the scale represented by the designated field in the second training log data based on the designated scale input by the user to obtain third training log data including the designated field represented by the designated scale;
[0038] Mapping the second training log data to the training graph includes:
[0039] The third training log data is mapped into the training graph.
[0040] Optionally, converting the text attribute data of each computer entity in the to-be-predicted log data into a first node feature vector of a corresponding node in the target graph includes:
[0041] Dividing the text attribute data of each node in the target graph into a plurality of text segments;
[0042] Determine a segment vector for each text segment based on a preset correspondence between the text segment and the segment vector;
[0043] All the fragment vectors of each node are concatenated or added together to obtain the first node feature vector of each node.
[0044] A second aspect of an embodiment of the present application provides an attack path prediction device, including:
[0045] A receiving unit, configured to receive the log data to be predicted, and map the log data to be predicted into a target graph, wherein the target graph includes a plurality of nodes and edges, wherein the nodes are used to represent computer entities in the log data to be predicted, and the edges are used to represent logical relationships between computer entities;
[0046] A conversion unit, used for converting the text attribute data of each computer entity in the log data to be predicted into a first node feature vector of a corresponding node in the target graph;
[0047] A determination unit is used to input the first node feature vector into a prediction model to determine the attack path in the log data to be predicted, and the prediction model is used to determine the edge in the target graph that meets the preset conditions as the attack path based on the first node feature vector.
[0048] A third aspect of the embodiments of the present application provides an attack path prediction device, including:
[0049] CPU, memory and input / output interface;
[0050] The memory is a short-term storage memory or a persistent storage memory;
[0051] The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the aforementioned method.
[0052] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enable the computer to execute the aforementioned method.
[0053] A fifth aspect of the embodiments of the present application provides a computer program product. When the computer program product is run on a computer, the computer executes the aforementioned method.
[0054] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:
[0055] First, the log data to be predicted is received, and the log data to be predicted is mapped into a target graph, then the text attribute data of each computer entity in the log data to be predicted is converted into the first node feature vector of the corresponding node in the target graph, and then the first node feature vector is input into the prediction model to determine the attack path in the log data to be predicted. Because the embodiment of the present application realizes the prediction of the attack path through the prediction model, and the prediction model in the embodiment of the present application determines the edge that meets the preset conditions in the target graph as the attack path according to the first node feature vector, and because the trained prediction model can automatically learn the latest attack path, the method of relying on the prediction model to predict the attack path is higher in the prediction accuracy of the attack path than the method of relying on artificial rules to predict the attack path in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 A schematic diagram of an embodiment of an attack path prediction method disclosed in the present application;
[0057] Figure 2 A schematic diagram of another embodiment of an attack path prediction method disclosed in the present application;
[0058] Figure 3 A schematic diagram of another embodiment of an attack path prediction method disclosed in the present application;
[0059] Figure 4 A schematic diagram of an embodiment of an attack path prediction device disclosed in the present application;
[0060] Figure 5 A schematic diagram of another embodiment of an attack path prediction device disclosed in the present application;
[0061] Figure 6A schematic diagram of another embodiment of an attack path prediction device disclosed in the present application. DETAILED DESCRIPTION
[0062] The present application is further described in detail below in conjunction with the accompanying drawings.
[0063] The embodiments of the present application provide an attack path prediction method and related devices for improving the prediction accuracy of the attack path.
[0064] With the continuous development of Internet technology, the attack techniques of cyber criminals are also constantly evolving. Cyber criminals can design attack paths according to their own illegal purposes to attack accurately and efficiently. The attack path is a series of steps taken by cyber criminals to invade or attack the target system. The attack path can be linear or branched, depending on the purpose of the cyber criminals. In order to prevent, network security personnel will predict the attack path, aiming to restore the network attack path through the log data on the terminal and traffic equipment, so as to establish a targeted defense mechanism to reduce the risk. However, the existing attack path prediction scheme relies on manual rules to achieve defense. The formulation of manual rules requires a lot of time and effort, and it is difficult to take into account all attack situations. The prediction accuracy is low, and the detection rate (that is, the ratio of the number of correctly predicted attack steps to the number of steps contained in the actual attack path) and the false alarm rate (that is, the ratio of the number of false alarm attack steps detected to the number of all detected attack steps) are both low. In order to solve the above problems, the present application provides an attack path prediction method and an attack path prediction device, and the trained prediction model can automatically learn the latest attack path. Therefore, the method of relying on the prediction model to predict the attack path has a higher prediction accuracy for the attack path than the method of relying on manual rules to predict the attack path in the prior art.
[0065] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only 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 should fall within the scope of protection of the present invention.
[0066] The terms "first", "second", "third", "fourth", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0067] The following describes an attack path prediction method of the present application. Figure 1 A prediction model involved in an attack path prediction method of an embodiment of the present application includes a feature conversion layer and a classifier, and the method includes:
[0068] 101. Receive log data to be predicted, and map the log data to be predicted into a target graph;
[0069] Receive the log data to be predicted, and map the log data to be predicted into a target graph. The target graph includes multiple nodes and edges, the nodes are used to represent the computer entities in the log data to be predicted, and the edges are used to represent the logical relationships between the computer entities. The log data to be predicted can be captured from a specific server or input by the user, which is not limited here. The log data to be predicted records the relevant computer entities and the logical relationships between them and other computer entities. The computer entities can be terminals, servers, processes, etc. The logical relationships between computer entities include reading, creating, modifying or accessing, etc. Find the computer entities involved in the log data to be predicted, map the computer entities into nodes, and then map the corresponding logical relationships between them into edges, thereby forming a target graph corresponding to the log data to be predicted.
[0070] 102. Convert the text attribute data of each computer entity in the log data to be predicted into the first node feature vector of the corresponding node in the target graph;
[0071] The text attribute data of each computer entity in the log data to be predicted is converted into the first node feature vector of the corresponding node in the target graph. The text attribute data of each computer entity represents part or all of the attributes of the corresponding computer entity, such as the text attribute data here can be the identifier of a certain computer entity, the time point of the event generation of the computer entity, and the specific description of the event.
[0072] After obtaining the text attribute data of each computer entity, the text attribute data is further divided into multiple text segments, and then based on the preset relationship between the text segments and the segments, the first node feature vector of the corresponding node can be calculated using the BERT model or the TD-IDF algorithm. In the embodiment of the present application, one computer entity corresponds to one node, and the text attribute data of one computer entity corresponds to the first node feature vector of a certain node.
[0073] 103. Input the first node feature vector into the prediction model to determine the attack path in the log data to be predicted.
[0074] The first node feature vector is input into the prediction model to determine the attack path in the log data to be predicted, wherein the prediction model is used to determine the edge in the target graph that meets the preset conditions as the attack path according to the first node feature vector.
[0075] Specifically, the prediction model is a pre-trained model, and the prediction model specifically includes a feature conversion layer and a classifier, wherein the input of the feature conversion layer is the first node feature vector, and the output is the first edge feature vector of all edges in the target graph, that is, the feature conversion layer can obtain the first edge feature vector of each edge by performing mathematical processing such as weighted fusion or splicing on the first node feature vector, and then use the classifier to judge the first edge feature vector based on preset conditions. When the first edge feature vector of an edge does not meet the preset conditions, it means that the possibility of this edge being an attack path is small. When the first edge feature vector of an edge meets the preset conditions, it means that the possibility of this edge being an attack path is large. Then, the prediction model predicts the edge whose probability of the attack path is greater than the preset probability as the attack path of the log data to be predicted.
[0076] In the embodiment of the present application, log data to be predicted is received, and the log data to be predicted is mapped to a target graph, and then the text attribute data of each computer entity in the log data to be predicted is converted into the first node feature vector of the corresponding node in the target graph, and finally the first node feature vector is input into the prediction model to determine the attack path in the log data to be predicted. Because the embodiment of the present application realizes the prediction of the attack path through the prediction model, and the trained prediction model in the embodiment of the present application can automatically learn the latest attack path, the method of relying on the prediction model to predict the attack path is more accurate in predicting the attack path than the method of relying on artificial rules to predict the attack path in the prior art.
[0077] In order to accurately characterize the representation of nodes in the target graph and better capture the complex relationship between nodes, the prediction model in the embodiment of the present application may also include a graph attention mechanism. Among them, the graph attention mechanism can classify, cluster or predict the nodes in the graph, adaptively learn the relationship between every two nodes, and can also process graphs of different sizes and structures, with strong generalization ability. Please refer to Figure 2 Another embodiment of an attack path prediction method of the present application includes:
[0078] Among them, the prediction model in this embodiment includes a feature conversion layer, a graph attention layer and a classifier;
[0079] 201. Receive log data to be predicted, and map the log data to be predicted into a target graph;
[0080] Receive the log data to be predicted, and map the log data to be predicted into a target graph. The target graph includes multiple nodes and edges, the nodes are used to represent the computer entities in the log data to be predicted, and the edges are used to represent the logical relationships between the computer entities. The log data to be predicted can be captured from a specific server or input by the user, which is not limited here. The log data to be predicted records the relevant computer entities and the logical relationships between them and other computer entities. The computer entities can be terminals, servers, processes, etc. The logical relationships between computer entities include reading, creating, modifying or accessing, etc. Find the computer entities involved in the log data to be predicted, map the computer entities into nodes, and then map the corresponding logical relationships between them into edges, thereby forming a target graph corresponding to the log data to be predicted.
[0081] 202. Convert the text attribute data of each computer entity in the log data to be predicted into the first node feature vector of the corresponding node in the target graph;
[0082] The text attribute data of each computer entity in the log data to be predicted is converted into the first node feature vector of the corresponding node in the target graph. The text attribute data here is used to represent the relevant attributes of the corresponding computer entity, such as the text attribute data can be the identification of a certain computer entity, the time point of the event generation of the computer entity and the specific description of the event, etc. The embodiment of the present application specifically uses the text attribute data of each computer entity to characterize the corresponding node.
[0083] Specifically, the text attribute data of each node in the target graph can be divided into multiple text segments, and then the segment vector of each text segment is determined based on the preset correspondence between the text segment and the segment vector, and finally all the segment vectors of each node are spliced or added to obtain the first node feature vector of each node. Among them, the correspondence between the text segment and the segment vector can be preset and exists in the form of a vocabulary. If the text attribute data is "AB C", which consists of three text segments, if there are only 4 text segments in the vocabulary, namely A-(1, 0, 0, 0), B-(0, 1, 0, 0), C-(0, 0, 1, 0) and D-(0, 0, 0, 1), then the segment vectors of the node are (1, 0, 0, 0), (0, 1, 0, 0) and (0, 0, 1, 0), and then spliced or added to obtain the first node feature vector.
[0084] It is understandable that the text attribute data may also be converted into the corresponding first node feature vector using a BERT model or a TD-IDF algorithm, etc., which is not specifically limited here.
[0085] 203. Input the first node feature vector to the graph attention layer of the prediction model to determine the similarity between the first node feature vector of each node in the target graph and each first node feature vector of the remaining nodes in the target graph;
[0086] The first node feature vector is input into the graph attention layer of the prediction model to determine the similarity between the first node feature vector of each node in the target graph and the first node feature vectors of the remaining nodes in the target graph. The graph attention layer calculates the attention coefficient, i.e., the similarity, between the first node feature vector of each node and the first node feature vectors of other nodes in the target graph to serve as a weight. When calculating the similarity, Euclidean similarity or cosine similarity can be calculated, which is not limited here.
[0087] 204. Perform weighted fusion on the similarity and the first node feature vector based on the graph attention layer to obtain a second node feature vector for each node in the target graph;
[0088] Based on the graph attention layer, the similarity and the first node feature vector are weightedly fused to obtain the second node feature vector of each node in the target graph. For a certain node, the similarity between the first node feature vector of the node and the first node feature vector of the neighboring nodes of the node is first calculated. After obtaining multiple similarities related to the first feature vector of the node, the first node feature vector of each node of other nodes is matched with their respective similarities, and weighted fusion is performed. The first node feature vector of each node in the neighboring nodes of the node is matched with their respective similarities, and weighted fusion is performed to finally obtain the new second node feature vector of the node.
[0089] For example, assuming that there are only three nodes in the target graph, namely A, B and C, and the corresponding first node feature vectors are x, y and z, first find the second node feature vector of node A. Specifically, we can first find the similarity Wab between the first node feature vector x of node A and the first node feature vector y of node B, and the similarity Wac between the first node feature vector z of node A and node C, and then perform weighted fusion. Finally, we get the second node feature vector of node A as (y*Wab+z*Wac) / (Wab+Wac). We can find the second node feature vectors of nodes B and C based on the above idea. By analogy, we can find the new second node feature vectors of all nodes in the target graph.
[0090] 205. Input two second node feature vectors of every two nodes that are neighboring nodes in the target graph into a feature conversion layer of the prediction model to obtain a first edge feature vector of each edge between every two nodes that are neighboring nodes in the target graph;
[0091] The two second node feature vectors of every two nodes that are neighboring nodes in the target graph are input into the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge between every two nodes that are neighboring nodes in the target graph. Among them, neighboring nodes are nodes that have a direct logical relationship with each other (there is an edge between the two nodes), that is, the two second node feature vectors of every two nodes that have a direct logical relationship with each other in the target graph are input into the feature conversion layer. In this way, all edges have corresponding first edge feature vectors.
[0092] Specifically, the second node feature vector can be input into the feature conversion layer to concatenate the second node feature vectors of each two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph, or the second node feature vector can be input into the feature conversion layer to add the second node feature vectors of each two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph, or the second node feature vector can be input into the feature conversion layer to perform weighted fusion on the second node feature vectors of each two nodes that are neighboring nodes in the target graph and the similarity of the two vectors to obtain the first edge feature vector of each edge in the target graph. The two second node feature vectors of each two nodes that are neighboring nodes are calculated to obtain the first edge feature vector of the edge between the two nodes. In addition to concatenation, addition or weighted fusion, there can be other types of calculation methods, which are not limited here.
[0093] For ease of understanding, an example is given below: Assume that there are only two pairs of nodes that are neighboring nodes in the target graph, namely nodes A, B and C, and A is connected to B and C respectively (i.e., there is an edge), and there is no edge between node B and node C. Specifically, assuming that the second node feature vectors of nodes A, B and C are a, b and c respectively, a and b are subjected to mathematical processing such as concatenation, addition or weighted fusion to obtain the first edge feature vector of the edge between node A and node B, and a and c are subjected to mathematical processing such as concatenation, addition or weighted fusion to obtain the first edge feature vector of the edge between node A and node C.
[0094] 206. Input the first edge feature vector into the classifier of the prediction model to obtain a corresponding probability value of each edge;
[0095] The first edge feature vector is input into the classifier of the prediction model to obtain the corresponding probability value of each edge. The input of the classifier is the first edge feature vector, and the output is the corresponding probability value of the edge. Before applying the classifier, supervised learning is performed on the classifier, and the weight of the classifier is trained by using the historical database and labeling to calculate a more accurate probability value, thereby accurately giving the probability value of the edge corresponding to each first edge feature vector.
[0096] 207. Determine the edge whose probability value is greater than or equal to the preset threshold as the attack path in the log data to be predicted.
[0097] The edges with probability values greater than or equal to the preset threshold are determined as attack paths in the log data to be predicted. To determine the qualified attack paths from all the edges in the target graph, the probability values of all edges need to be compared with the preset threshold, where the preset threshold can be set according to actual needs and is not limited here. When the probability value of an edge is less than the preset threshold, it means that the possibility of this edge being an attack path is small, and the edge is predicted as a normal path. When the probability value of an edge is greater than or equal to the preset threshold, it means that the possibility of this edge being an attack path is large, and the edge is predicted as the attack path of the log data to be predicted.
[0098] In this embodiment, the log data to be predicted is first received and mapped into a target graph. Then, the text attribute data of each computer entity in the log data to be predicted is converted into the first node feature vector of the corresponding node in the target graph. After the first node feature vector is converted into a second node feature vector, the second node feature vector is input into the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge in the target graph. Then, the first edge feature vector is input into the classifier of the prediction model to obtain the corresponding probability value of each edge. Finally, the edge with a probability value greater than or equal to a preset threshold is determined as the attack path of the log data to be predicted.
[0099] Because the trained prediction model in the embodiment of the present application can automatically learn the latest attack path, the method of relying on the prediction model to predict the attack path has a higher prediction accuracy for the attack path than the method of relying on artificial rules to predict the attack path in the prior art. In addition, since the prediction module further introduces a graph attention mechanism to learn and obtain the relationship between each node, the graph attention mechanism in the prediction model can more accurately calculate the edge feature vector of the corresponding edge based on the relationship between each node, that is, the edge feature vector of each edge in the target graph output by the graph attention mechanism has a higher accuracy, so the prediction model with the graph attention mechanism has a higher prediction accuracy for the attack path.
[0100] It is easy to understand that before using the prediction model, it is necessary to use the historical database to train the prediction model, such as using a large amount of normal user behavior data and attack data in the historical database for end-to-end training to adaptively learn the attack path. Figure 3 Another embodiment of the attack path prediction method in the embodiment of the present application includes:
[0101] 301. Obtain first training log data from a historical database;
[0102] The first training log data is obtained from the historical database. The historical database stores multiple log data, including both unattacked log data, i.e., the log data has no attack path, and attacked log data, i.e., the log data has an attack path. The multiple log data randomly obtained from the historical database for training are the first training log data, including both normal paths and attack paths.
[0103] 302. Determine multiple groups of repeated log data in the first training log data that are continuously triggered and have the same content except for the timestamp;
[0104] Determine multiple sets of repeated log data in the first training log data that are continuously triggered and have the same content except for the timestamp. Since there is a high probability of duplication in the log data, it is necessary to perform a step to eliminate redundant data. Find multiple sets of repeated log data that are basically consistent in content except for the timestamp. Specifically, select a certain log data from the first training log data, and query it according to the computer entities involved in the log data, the logical relationship between the computer entities, and the attributes, and the timestamp does not need to be queried.
[0105] 303. Delete all log data except the latest time point in the multiple sets of duplicate log data to obtain second training log data;
[0106] Delete all log data except the latest time point in the multiple sets of duplicate log data to obtain the second training log data. Since the log data is repeated and only one is needed, the redundant log data needs to be deleted, and the log data of the latest time point in the multiple sets of duplicate log data is saved, and the other duplicate log data is deleted, and so on, and the remaining log data in the first training log data is also de-redundantly eliminated to obtain the second training log data without duplication.
[0107] 304. Adjust the scale represented by the designated field in the second training log data based on the designated scale input by the user to obtain third training log data including the designated field represented by the designated scale;
[0108] The scale represented by the specified field in the second training log data is adjusted based on the specified scale input by the user to obtain the third training log data containing the specified field represented by the specified scale. Since the log data specifications are different, the second training log data needs to be standardized, that is, the logs with the same field meaning but different scales are standardized. For example, the specified scale of the field "threat level" is 0-10, that is, the highest threat is 10 and the lowest threat is 0. Now there are two log data, one with a scale of 0-5 for the "threat level" field and the other with a scale of 0-100. These two training log data are unified as 0-10, that is, the third training log data.
[0109] 305. Map the third training log data into a training graph;
[0110] The third training log data is mapped into a training graph. The training graph includes nodes and edges, the nodes are used to represent computer entities in the training log data, and the edges are used to represent logical relationships between computer entities. Specifically, the computer entities in the third training log data are first queried, and the computer entities are mapped into nodes, and the logical relationships corresponding to the computer entities are mapped into edges to form a training graph.
[0111] 306. Convert the text attribute data of each computer entity in the training log data into a third node feature vector of a corresponding node in the training graph;
[0112] The text attribute data of each computer entity in the training log data is converted into the third node feature vector of the corresponding node in the training graph. The text attribute data represents part or all of the attributes of the corresponding computer entity, and the first node feature vector of the corresponding node can be calculated based on the relationship between text fragments, or using the BERT model, or using the TD-IDF algorithm.
[0113] 307. Input a plurality of third node feature vectors into the initialized prediction model to obtain a predicted probability value of each edge between every two nodes output by the initialized prediction model;
[0114] Multiple third node feature vectors are input into the initialized prediction model to obtain the predicted probability value of each edge between every two nodes output by the initialized prediction model, wherein the predicted probability value is used to characterize the predicted probability that each edge between every two nodes is an attack path.
[0115] As an optional embodiment, the prediction model in the embodiment of the present application includes a graph attention layer, a feature conversion layer and a classifier. The third node feature vector is first input into the graph attention layer of the initialized prediction model to determine the training similarity between the third node feature vector of each node in the training graph and the third node feature vectors of the remaining nodes in the training graph, wherein the training similarity is the training weight, i.e., the attention coefficient. Then, the training similarity and the third node feature vector are weightedly fused based on the graph attention layer to obtain the fourth node feature vector of each node in the training graph. In short, for a certain node, the training similarity between the third node feature vector of the node and the third node feature vector of the neighboring nodes of the node is first calculated. After obtaining multiple training similarities related to the third feature vector of the node, the third node feature vector of each node in the neighboring nodes of the node is matched with the respective training similarities, and weighted fusion is performed to finally obtain the new fourth node feature vector of the node. For example, assuming that there are only three nodes in the target graph, namely A, B and C, and the corresponding third node feature vectors are x, y and z, first calculate the third node feature vector of node A. Specifically, calculate the similarity Wab between the third node feature vector x of node A and the third node feature vector y of node B, as well as the similarity Wac between the third node feature vectors z of node A and node C, and then perform weighted fusion, that is, the fourth node feature vector of node A is (y*Wab+z*Wac) / (Wab+Wac), and the fourth node feature vectors of node B and node C are calculated based on the above idea.
[0116] Similarly, the fourth node feature vectors of all nodes in the target graph can be calculated. Then, the two fourth node feature vectors of every two nodes that are neighboring nodes in the training graph are input into the feature conversion layer of the prediction model to obtain the second side feature vector of each edge between every two nodes that are neighboring nodes in the training graph. Specifically, the two fourth node feature vectors of every two nodes that are neighboring nodes can be concatenated, added, or weightedly fused to obtain the second side feature vector of the edge between the two nodes. In addition to the above three calculation methods, there can be other methods, which are not limited here. For ease of understanding, the following example is given:
[0117] Assume that there are only two pairs of nodes that are neighboring nodes in the target graph, namely nodes A, B and C, and A is connected to B and C respectively (i.e., there is an edge), and there is no edge between node B and node C. Specifically, assuming that the second node feature vectors of nodes A, B and C are a, b and c respectively, a and b are subjected to mathematical processing such as splicing, addition or weighted fusion to obtain the second edge feature vector of the edge between node A and node B, and a and c are subjected to mathematical processing such as splicing, addition or weighted fusion to obtain the second edge feature vector of the edge between node A and node C.
[0118] Finally, the second edge feature vector is used as the input of the classifier of the initialized prediction model to obtain the predicted probability value output by the classifier. At this time, the initialized prediction model is not trained, so it is necessary to train the initialized prediction model according to the predicted probability value finally output by the initialized prediction model and the training probability value of the following steps. As for the specific training process, it will be described in detail in the following steps and will not be repeated here.
[0119] 308. Receive a training probability value for each edge input by a user based on a historical database;
[0120] Receive the training probability value for each edge input by the user based on the historical database. The user will determine the training probability value of each edge based on the historical database and personal experience, and input it into the prediction model. For example, in 10 scenarios, 7 scenarios of an edge are attack paths and 3 scenarios are normal paths. At this time, the training probability value of this edge should be 70%. Combined with the user's experience and judgment of this edge, the actual value may be higher, which can be determined as 72% and input into the prediction model as a label.
[0121] 309. Train the initialized prediction model using the prediction probability value and the training probability value until the initialized prediction model converges to obtain a prediction model;
[0122] The initialized prediction model is trained using the prediction probability value and the training probability value until the initialized prediction model converges to obtain the prediction model.
[0123] Specifically, the training process is as follows:
[0124] First, use the preset loss function to calculate the loss value between the predicted probability value and the training probability value corresponding to the pre-input second edge feature vector, where the loss value represents the deviation between the predicted probability value and the training probability value. The larger the loss value, the greater the difference between the two. Subsequent adjustments are required to lower the loss value and then adjust the weight. Then use the loss value and the back propagation algorithm to train the prediction model until the prediction model converges. After obtaining the loss value, use the back propagation algorithm to adjust the weights involved in the prediction model, and use this new weight to calculate the predicted probability value of a certain edge, and compare this new predicted probability value with the training probability value. If the loss value is still large, then continue to adjust the weight. If the loss value is less than the preset threshold, it means that the prediction model has converged and the training is basically completed. The corresponding weight at this time is determined as the final weight of the prediction model, which can be applied later.
[0125] 310. Receive log data to be predicted, and map the log data to be predicted into a target graph;
[0126] 311. Convert the text attribute data of each computer entity in the log data to be predicted into the first node feature vector of the corresponding node in the target graph;
[0127] 312. Input the first node feature vector to the graph attention layer of the prediction model to determine the similarity between the first node feature vector of each node in the target graph and each first node feature vector of the remaining nodes in the target graph;
[0128] 313. Perform weighted fusion on the similarity and the first node feature vector based on the graph attention layer to obtain a second node feature vector for each node in the target graph;
[0129] 314. Input two second node feature vectors of every two nodes that are neighboring nodes in the target graph into a feature conversion layer of the prediction model to obtain a first edge feature vector of each edge between every two nodes that are neighboring nodes in the target graph;
[0130] 315. Input the first edge feature vector into the classifier of the prediction model to obtain the corresponding probability value of each edge;
[0131] 316. Determine the edge whose probability value is greater than or equal to the preset threshold as the attack path of the log data to be predicted.
[0132] Steps 310 to 316 in this embodiment are similar to those in the previous embodiment. Figure 2 Steps 201 to 207 in the illustrated embodiment are similar and will not be described in detail here.
[0133] In this embodiment, the prediction model is first trained, the log data to be predicted is received, and the log data to be predicted is mapped into a target graph, then the text attribute data of each computer entity in the log data to be predicted is converted into the first node feature vector of the corresponding node in the target graph, the first node feature vector is converted into a second node feature vector, and then the second node feature vector is input into the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge in the target graph, and then the first edge feature vector is input into the classifier of the prediction model to obtain the corresponding probability value of each edge, and finally the edge with a probability value greater than or equal to a preset threshold is determined as the attack path of the log data to be predicted.
[0134] Because the trained prediction model in the embodiment of the present application can automatically learn the latest attack path, the method of relying on the prediction model to predict the attack path has a higher prediction accuracy for the attack path than the method of relying on artificial rules to predict the attack path in the prior art. In addition, since the prediction module further introduces a graph attention mechanism to learn and obtain the relationship between each node, the graph attention mechanism in the prediction model can more accurately calculate the node feature vector of each node based on the relationship between each node, that is, the accuracy of the node feature vector of each node in the target graph output by the graph attention mechanism is higher, so the prediction model with the graph attention mechanism has a higher prediction accuracy for the attack path.
[0135] The attack path prediction method in the embodiment of the present application is described above, and the attack path prediction device in the embodiment of the present application is described below. Figure 4 , an attack path prediction device in an embodiment of the present application includes:
[0136] The receiving unit 401 is used to receive the log data to be predicted, and map the log data to be predicted into a target graph, wherein the target graph includes a plurality of nodes and edges, wherein the nodes are used to represent computer entities in the log data to be predicted, and the edges are used to represent logical relationships between computer entities;
[0137] A conversion unit 402, configured to convert text attribute data of each computer entity in the log data to be predicted into a first node feature vector of a corresponding node in the target graph;
[0138] The determination unit 403 is used to input the first node feature vector into the prediction model to determine the attack path in the log data to be predicted. The prediction model is used to determine the edge in the target graph that meets the preset conditions as the attack path based on the first node feature vector.
[0139] In the embodiment of the present application, the receiving unit 401 first receives the log data to be predicted, and maps the log data to be predicted into a target graph, then the conversion unit 402 converts the text attribute data of each computer entity in the log data to be predicted into the first node feature vector of the corresponding node in the target graph, and then the determination unit 403 inputs the first node feature vector into the prediction model to determine the attack path in the log data to be predicted. The attack path is predicted by the prediction model, and the trained prediction model has automatically learned the latest attack path, so the method of relying on the prediction model to predict the attack path is more accurate than the method of relying on artificial rules to predict the attack path in the prior art.
[0140] The following is a detailed description of an attack path prediction device in an embodiment of the present application. Figure 5 Another embodiment of an attack path prediction device of the embodiment of the present application includes:
[0141] The receiving unit 501 is used to receive the log data to be predicted, and map the log data to be predicted into a target graph, wherein the target graph includes a plurality of nodes and edges, wherein the nodes are used to represent computer entities in the log data to be predicted, and the edges are used to represent logical relationships between computer entities;
[0142] A conversion unit 502, configured to convert text attribute data of each computer entity in the to-be-predicted log data into a first node feature vector of a corresponding node in the target graph;
[0143] The determination unit 503 is used to input the first node feature vector into the prediction model to determine the attack path in the log data to be predicted. The prediction model is used to determine the edge in the target graph that meets the preset conditions as the attack path based on the first node feature vector.
[0144] The determining unit 503 is specifically configured to:
[0145] Inputting the first node feature vector into the feature conversion layer to obtain a first edge feature vector of each edge in the target graph;
[0146] Inputting the first edge feature vector into the classifier to obtain a corresponding probability value of each edge;
[0147] The edge whose probability value is greater than or equal to a preset threshold is determined as the attack path in the log data to be predicted.
[0148] The determining unit 503 is specifically configured to:
[0149] Inputting the first node feature vector to the feature conversion layer to concatenate the first node feature vectors of every two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph;
[0150] or,
[0151] Inputting the first node feature vector to the feature conversion layer to add the first node feature vectors of every two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph;
[0152] or,
[0153] The first node feature vector is input into the feature conversion layer to weightedly fuse the first node feature vectors of every two nodes that are neighboring nodes in the target graph and the similarity of the two vectors to obtain the first edge feature vector of each edge in the target graph.
[0154] The determining unit 503 is further configured to:
[0155] Inputting the first node feature vector into a graph attention layer of the prediction model to determine the similarity between the first node feature vector of each node in the target graph and each first node feature vector of the remaining nodes in the target graph;
[0156] The attack path prediction device also includes:
[0157] A processing unit 504 is used to perform weighted fusion on the similarity and the first node feature vector based on the graph attention layer to obtain a second node feature vector for each node in the target graph;
[0158] The determining unit 503 is specifically configured to:
[0159] The two second node feature vectors of every two nodes that are neighboring nodes in the target graph are input into the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge between every two nodes that are neighboring nodes in the target graph.
[0160] The attack path prediction device further includes a training unit 505, which is specifically used to:
[0161] Acquire first training log data from a historical database;
[0162] Mapping the first training log data into a training graph, wherein the training graph includes nodes and edges, the nodes are used to represent computer entities in the training log data, and the edges are used to represent logical relationships between computer entities;
[0163] Converting the text attribute data of each computer entity in the training log data into a third node feature vector of a corresponding node in the training graph;
[0164] Inputting a plurality of the third node feature vectors into the initialized prediction model to obtain a predicted probability value of each edge between every two nodes output by the initialized prediction model, wherein the predicted probability value is used to characterize the predicted probability that each edge between every two nodes is an attack path;
[0165] Obtaining a training probability value input for each edge in the historical database, wherein the training probability value is used to characterize a true probability that each edge is an attack path;
[0166] The initialized prediction model is trained using the prediction probability value and the training probability value until the initialized prediction model converges to obtain the prediction model.
[0167] The training unit 505 is further configured to:
[0168] Determine multiple groups of repeated log data that are continuously triggered and have the same content except for timestamps in the first training log data;
[0169] Deleting all log data except the latest time point in the multiple sets of repeated log data to obtain second training log data;
[0170] The second training log data is mapped into the training graph.
[0171] The training unit 505 is further configured to:
[0172] Adjusting the scale represented by the designated field in the second training log data based on the designated scale input by the user to obtain third training log data including the designated field represented by the designated scale;
[0173] The third training log data is mapped into the training graph.
[0174] The conversion unit 502 is specifically used for:
[0175] Dividing the text attribute data of each node in the target graph into a plurality of text segments;
[0176] Determine a segment vector for each text segment based on a preset correspondence between the text segment and the segment vector;
[0177] All the fragment vectors of each node are concatenated or added together to obtain the first node feature vector of each node.
[0178] The functions and processes performed by each unit in the attack path prediction device of this embodiment are similar to those of the aforementioned Figures 1 to 3 The functions and processes performed by the attack path prediction device are similar and will not be repeated here.
[0179] Figure 6 It is a schematic diagram of the structure of an attack path prediction device provided in an embodiment of the present application. The attack path prediction device 600 may include one or more central processing units (CPU) 601 and a memory 605, and the memory 605 stores one or more application programs or data.
[0180] The memory 605 may be a volatile storage or a persistent storage. The program stored in the memory 605 may include one or more modules, each of which may include a series of instruction operations in the attack path prediction device 600. Furthermore, the central processor 601 may be configured to communicate with the memory 605 and execute a series of instruction operations in the memory 605 on the attack path prediction device 600.
[0181] The attack path prediction device 600 may also include one or more power supplies 602, one or more wired or wireless network interfaces 603, one or more input and output interfaces 604, and / or one or more operating systems, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.
[0182] The CPU 601 can execute the aforementioned Figures 1 to 3 The operations performed by the attack path prediction device in the illustrated embodiment will not be described in detail here.
[0183] An embodiment of the present application also provides a computer-readable storage medium, including instructions, which, when executed on a computer, enable the computer to execute the method in the aforementioned embodiment.
[0184] The embodiments of the present application also provide a computer program product. When the computer program product is run on a computer, the computer executes the method in the aforementioned embodiments.
[0185] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0186] It should be noted that, although the steps in the flowcharts involved in the embodiments are drawn in sequence according to the instructions of the arrows, unless otherwise clearly stated in this document, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0187] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.
[0188] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0189] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0190] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.
Claims
1. A method for predicting an attack path, characterized in that: include: Receiving log data to be predicted, and mapping the log data to be predicted into a target graph, wherein the target graph includes a plurality of nodes and edges, wherein the nodes are used to represent computer entities in the log data to be predicted, and the edges are used to represent logical relationships between computer entities; Converting the text attribute data of each computer entity in the to-be-predicted log data into a first node feature vector of a corresponding node in the target graph; The first node feature vector is input into a prediction model to determine an attack path in the log data to be predicted. The prediction model is used to determine an edge in the target graph that meets a preset condition as an attack path based on the first node feature vector.
2. The attack path prediction method according to claim 1, characterized in that: The prediction model includes a feature conversion layer and a classifier, and the step of inputting the first node feature vector into the prediction model to determine the attack path in the log data to be predicted includes: Inputting the first node feature vector into the feature conversion layer to obtain a first edge feature vector of each edge in the target graph; Inputting the first edge feature vector into the classifier to obtain a corresponding predicted probability value of each edge; The edge whose predicted probability value is greater than or equal to a preset threshold is determined as the attack path in the log data to be predicted.
3. The attack path prediction method according to claim 2, characterized in that: The step of inputting the first node feature vector to the feature conversion layer to obtain a first edge feature vector of each edge in the target graph includes: Inputting the first node feature vector to the feature conversion layer to concatenate the first node feature vectors of every two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph; or, Inputting the first node feature vector to the feature conversion layer to add the first node feature vectors of every two nodes that are neighboring nodes in the target graph to obtain the first edge feature vector of each edge in the target graph; or, The first node feature vector is input into the feature conversion layer to weightedly fuse the first node feature vectors of every two nodes that are neighboring nodes in the target graph and the similarity of the two vectors to obtain the first edge feature vector of each edge in the target graph.
4. The attack path prediction method according to claim 2, characterized in that: The prediction model further includes a graph attention layer. After converting the text attribute data of each computer entity in the to-be-predicted log data into a first node feature vector of a corresponding node in the target graph, the method further includes: Inputting the first node feature vector into a graph attention layer of the prediction model to determine the similarity between the first node feature vector of each node in the target graph and each first node feature vector of the remaining nodes in the target graph; Performing weighted fusion on the similarity and the first node feature vector based on the graph attention layer to obtain a second node feature vector for each node in the target graph; The step of inputting the first node feature vector to the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge in the target graph includes: The two second node feature vectors of every two nodes that are neighboring nodes in the target graph are input into the feature conversion layer of the prediction model to obtain the first edge feature vector of each edge between every two nodes that are neighboring nodes in the target graph.
5. The attack path prediction method according to claim 4, characterized in that: Before receiving the log data to be predicted, the method further includes: Acquire first training log data from a historical database; Mapping the first training log data into a training graph, wherein the training graph includes nodes and edges, the nodes are used to represent computer entities in the training log data, and the edges are used to represent logical relationships between computer entities; Converting the text attribute data of each computer entity in the training log data into a third node feature vector of a corresponding node in the training graph; Inputting a plurality of the third node feature vectors into the initialized prediction model to obtain a predicted probability value of each edge between every two nodes output by the initialized prediction model, wherein the predicted probability value is used to characterize the predicted probability that each edge between every two nodes is an attack path; Obtaining a training probability value input for each edge in the historical database, wherein the training probability value is used to characterize a true probability that each edge is an attack path; The initialized prediction model is trained using the prediction probability value and the training probability value until the initialized prediction model converges to obtain the prediction model.
6. The attack path prediction method according to claim 5, characterized in that: After acquiring the first training log data from the historical database, the method further includes: Determine multiple groups of repeated log data that are continuously triggered and have the same content except for timestamps in the first training log data; Deleting all log data except the latest time point in the multiple sets of repeated log data to obtain second training log data; Mapping the first training log data into a training graph includes: The second training log data is mapped into the training graph.
7. The attack path prediction method according to claim 6, characterized in that: After deleting all log data except the latest time point in the multiple sets of duplicate log data to obtain the second training log data, the method further includes: Adjusting the scale represented by the designated field in the second training log data based on the designated scale input by the user to obtain third training log data including the designated field represented by the designated scale; Mapping the second training log data to the training graph includes: The third training log data is mapped into the training graph.
8. The attack path prediction method according to claim 1, characterized in that: The step of converting the text attribute data of each computer entity in the log data to be predicted into a first node feature vector of a corresponding node in the target graph includes: Dividing the text attribute data of each node in the target graph into a plurality of text segments; Determine a segment vector for each text segment based on a preset correspondence between the text segment and the segment vector; All the fragment vectors of each node are concatenated or added together to obtain the first node feature vector of each node.
9. An attack path prediction device, characterized in that: include: A receiving unit, configured to receive the log data to be predicted, and map the log data to be predicted into a target graph, wherein the target graph includes a plurality of nodes and edges, wherein the nodes are used to represent computer entities in the log data to be predicted, and the edges are used to represent logical relationships between computer entities; A conversion unit, used for converting the text attribute data of each computer entity in the log data to be predicted into a first node feature vector of a corresponding node in the target graph; A determination unit is used to input the first node feature vector into a prediction model to determine the attack path in the log data to be predicted, and the prediction model is used to determine the edge in the target graph that meets the preset conditions as the attack path based on the first node feature vector.
10. A computer device, characterized in that: include: CPU, memory and input / output interface; The memory is a short-term storage memory or a persistent storage memory; The central processing unit is configured to communicate with the memory and execute instructions in the memory to perform the method according to any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that: The method comprises instructions, which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 8.
12. A computer program product, characterized in that When the computer program product is run on a computer, the computer is caused to execute the method according to any one of claims 1 to 8.
