Game cheat detection method, device, server and storage medium

By combining the transaction graph and structural vector algorithm to identify the game plug-in node cluster, the problem of low accuracy of plug-in detection in the existing technology is solved, and more efficient plug-in recognition and defense is achieved.

CN112604297BActive Publication Date: 2024-11-08NETEASE (HANGZHOU) NETWORK CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202011588884.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-29
Publication Date
2024-11-08
Estimated Expiration
2040-12-29

AI Technical Summary

Technical Problem

In the prior art, the accuracy of game plug-in detection is low, and it is impossible to effectively identify and prevent plug-in behavior, which affects the game ecosystem.

Method used

By constructing a transaction map of game transaction logs, using structural vector algorithms for feature representation and clustering, identifying plug-in node clusters, thereby improving detection accuracy and coverage.

Benefits of technology

It improves the accuracy and coverage of game plug-in detection, reduces labor costs, and does not affect players' game performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112604297B_ABST
    Figure CN112604297B_ABST
Patent Text Reader

Abstract

The present application provides a method, device, server and storage medium for detecting game plug-ins, wherein the method includes: constructing a transaction graph based on a game transaction log, the game transaction log includes: log data of virtual asset transactions conducted by multiple game characters in the game, using a preset structural vector algorithm for feature representation based on the transaction graph to obtain a feature vector of each node, clustering multiple nodes based on the feature vectors of multiple nodes to obtain at least one node cluster, and performing plug-in detection on the nodes in each node cluster based on the characteristics of the nodes in each node cluster. Since plug-in nodes have certain regularity in asset transactions, the detection of game plug-ins based on the transaction graph can improve the accuracy of game plug-in detection, and compared with the prior art, plug-in detection has a wider coverage, a higher recall rate, and a lower labor cost, and does not affect the player's game performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to a method, device, server and storage medium for detecting game plug-ins. Background Art

[0002] Game plug-ins refer to programs designed for a certain game that modify some of the programs in the game. There are many types of game plug-ins. For Massive Multiplayer Online Role-Playing Game (MMORPG), automatic plug-ins use some program scripts to automatically complete tasks, kill monsters, upgrade, and make money, and transfer and collect in-game assets, which seriously damages the game ecosystem.

[0003] In the prior art, the game client records various features of the plug-in process and the feature code generated by the memory code fragment through process inspection, and bans the process when a process with the same features is found during the running of the game process.

[0004] However, as the plug-in program is updated, the same features mentioned above will also change, and the use of the above solution will result in low accuracy in plug-in detection. Summary of the invention

[0005] The purpose of this application is to provide a game plug-in detection method, device, server and storage medium to address the deficiencies in the above-mentioned prior art, so as to solve the problem of low accuracy of plug-in detection in the prior art.

[0006] To achieve the above purpose, the technical solution adopted in the embodiment of the present application is as follows:

[0007] In a first aspect, an embodiment of the present application provides a method for detecting game cheats, the method comprising:

[0008] Constructing a transaction graph according to the game transaction log, wherein the game transaction log includes: log data of virtual asset transactions conducted by multiple game characters in the game, and the transaction graph includes: the multiple nodes, and transaction relationships between the nodes, wherein each node corresponds to a game character;

[0009] According to the transaction graph, a preset structural vector algorithm is used for feature representation to obtain a feature vector of each node;

[0010] Clustering the multiple nodes according to the feature vectors of the multiple nodes to obtain at least one node cluster;

[0011] According to the characteristics of the nodes in each node cluster, the nodes in each node cluster are subjected to plug-in detection.

[0012] Optionally, the preset structural vector algorithm includes: a structural distance algorithm, a network construction algorithm, and a walk-based vector representation algorithm;

[0013] According to the transaction graph, a preset structural vector algorithm is used to perform feature representation to obtain a feature vector of each node, including:

[0014] According to the degree sequence of every two nodes in the transaction graph, the structural distance algorithm is used to calculate the k-order structural distance between every two nodes, and the k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes;

[0015] According to the k-order structural distance between each two nodes, the network construction algorithm is used to construct the k-th layer network to obtain the K-layer fully connected network with K-layer networks, wherein the K-layer fully connected network includes: the edge weights of each two nodes in each layer of the network, and the edge weights of the same node in adjacent layers of the network; k is 0 or any positive integer less than K;

[0016] According to the K-layer fully connected network, the walk-based vector characterization algorithm is adopted to obtain the feature vector of each node in the K-th layer network.

[0017] Optionally, the calculating the k-order structural distance between each two nodes using the structural distance algorithm according to the degree sequence of each two nodes in the transaction graph includes:

[0018] Obtain a first degree sequence of a first node and a second degree sequence of a second node in each of the two nodes;

[0019] calculating a sequence distance between the first degree sequence and the second degree sequence;

[0020] The k-order structural distance between the first node and the second node is calculated according to the sequence distance and the structural distance of the previous-order neighbor node, wherein the previous-order neighbor node includes: the previous-order neighbor node of the first node and the previous-order neighbor node of the second node in the transaction graph.

[0021] Optionally, obtaining the first degree sequence of the first node and the second degree sequence of the second node in every two nodes includes:

[0022] Arrange the degrees of the k-stage neighbor nodes of the first node in ascending order to obtain the first degree sequence;

[0023] The degrees of the k-stage neighbor nodes of the second node are arranged in ascending order to obtain the second degree sequence.

[0024] Optionally, calculating the sequence distance between the first degree sequence and the second degree sequence includes:

[0025] The sequence distance is calculated according to the first degree sequence and the second degree sequence using a dynamic time normalization algorithm.

[0026] Optionally, constructing the k-th layer network by using the network construction algorithm according to the k-order structural distance between every two nodes to obtain the K-layer fully connected network having K-layer networks includes:

[0027] Calculating the k-order edge weights of each two nodes in the k-th layer network according to the k-order structural distance;

[0028] Calculate the k+1-order edge weight of the same node in the k-th network and the k+1-th network according to the number of nodes whose edge weights pointing to each node in the k-th network satisfy the preset weight condition;

[0029] The K-layer fully connected network is constructed according to the k-order edge weight, the k+1-order edge weight and the k-1-order edge weight, wherein the k-1-order edge weight is a preset weight.

[0030] Optionally, satisfying a preset weight condition includes: being greater than or equal to an average edge weight of the k-th layer network.

[0031] Optionally, clustering the multiple nodes according to the feature vectors of the multiple nodes to obtain at least one node cluster includes:

[0032] Calculating the node distance between every two nodes according to the feature vectors of the multiple nodes;

[0033] Clustering the multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster;

[0034] At least one node cluster is obtained from the at least one initial cluster according to the number of nodes in each initial cluster.

[0035] Optionally, clustering the multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster includes:

[0036] If there is at least one pair of target nodes between every two nodes, cluster the nodes within a preset neighborhood range where each pair of target nodes are located to obtain at least one initial cluster, and the node distance between each pair of target nodes is less than or equal to a preset neighborhood value.

[0037] Optionally, acquiring at least one node cluster from the at least one initial cluster according to the number of nodes in each initial cluster includes:

[0038] If there is at least one cluster in the at least one initial cluster whose number of nodes is greater than or equal to a preset number, the at least one cluster is used as the at least one node cluster.

[0039] Optionally, performing cheat detection on the nodes in each node cluster according to the characteristics of the nodes in each node cluster includes:

[0040] If there are pre-marked plug-in nodes in the node cluster, obtaining the proportion of the plug-in nodes in the node cluster;

[0041] If the proportion of the external nodes is less than the preset proportion, obtaining the first feature of the external nodes and the second features of other nodes in the initial cluster except the external nodes;

[0042] Performing similarity analysis based on the first feature and the second feature to obtain similarity between the plug-in node and the other nodes;

[0043] If the similarity is greater than or equal to a preset similarity, the node cluster is determined to be an external cluster.

[0044] Optionally, the method further comprises:

[0045] If the proportion of the external nodes is greater than or equal to a preset proportion, the node cluster is determined to be an external cluster.

[0046] In a second aspect, another embodiment of the present application provides a game cheat detection device, the device comprising:

[0047] A processing module, configured to construct a transaction graph according to a game transaction log, wherein the game transaction log includes: log data of virtual asset transactions conducted by multiple game characters in the game, and the transaction graph includes: the multiple nodes, and transaction relationships between the nodes, wherein each node corresponds to a game character;

[0048] An acquisition module is used to perform feature representation according to the transaction graph using a preset structural vector algorithm to obtain a feature vector of each node; cluster the multiple nodes according to the feature vectors of the multiple nodes to obtain at least one node cluster;

[0049] The detection module is used to perform plug-in detection on the nodes in each node cluster according to the characteristics of the nodes in each node cluster.

[0050] Optionally, the preset structural vector algorithm includes: a structural distance algorithm, a network construction algorithm, and a walk-based vector representation algorithm;

[0051] The acquisition module is specifically used for:

[0052] According to the degree sequence of every two nodes in the transaction graph, the structural distance algorithm is used to calculate the k-order structural distance between every two nodes, and the k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes;

[0053] According to the k-order structural distance between each two nodes, the network construction algorithm is used to construct the k-th layer network to obtain the K-layer fully connected network with K-layer networks, wherein the K-layer fully connected network includes: the edge weights of each two nodes in each layer of the network, and the edge weights of the same node in adjacent layers of the network; k is 0 or any positive integer less than K;

[0054] According to the K-layer fully connected network, the walk-based vector characterization algorithm is adopted to obtain the feature vector of each node in the K-th layer network.

[0055] Optionally, the acquisition module is specifically used to:

[0056] Obtain a first degree sequence of a first node and a second degree sequence of a second node in each of the two nodes;

[0057] Calculating a sequence distance between the first degree sequence and the second degree sequence;

[0058] The k-order structural distance between the first node and the second node is calculated according to the sequence distance and the structural distance of the previous-order neighbor node, wherein the previous-order neighbor node includes: the previous-order neighbor node of the first node and the previous-order neighbor node of the second node in the transaction graph.

[0059] Optionally, the acquisition module is specifically used to:

[0060] Arrange the degrees of the k-stage neighbor nodes of the first node in ascending order to obtain the first degree sequence;

[0061] The degrees of the k-stage neighbor nodes of the second node are arranged in ascending order to obtain the second degree sequence.

[0062] Optionally, the acquisition module is specifically used to:

[0063] The sequence distance is calculated according to the first degree sequence and the second degree sequence using a dynamic time normalization algorithm.

[0064] Optionally, the acquisition module is specifically used to:

[0065] Calculating the k-order edge weights of each two nodes in the k-th layer network according to the k-order structural distance;

[0066] Calculate the k+1-order edge weight of the same node in the k-th network and the k+1-th network according to the number of nodes whose edge weights pointing to each node in the k-th network satisfy the preset weight condition;

[0067] The K-layer fully connected network is constructed according to the k-order edge weight, the k+1-order edge weight and the k-1-order edge weight, wherein the k-1-order edge weight is a preset weight.

[0068] Optionally, satisfying a preset weight condition includes: being greater than or equal to an average edge weight of the k-th layer network.

[0069] Optionally, the acquisition module is specifically used to:

[0070] Calculating the node distance between every two nodes according to the feature vectors of the multiple nodes;

[0071] Clustering the multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster;

[0072] At least one node cluster is obtained from the at least one initial cluster according to the number of nodes in each initial cluster.

[0073] Optionally, the acquisition module is specifically used to:

[0074] If there is at least one pair of target nodes between every two nodes, cluster the nodes within a preset neighborhood range where each pair of target nodes are located to obtain at least one initial cluster, and the node distance between each pair of target nodes is less than or equal to a preset neighborhood value.

[0075] Optionally, the acquisition module is specifically used to:

[0076] If there is at least one cluster in the at least one initial cluster whose number of nodes is greater than or equal to a preset number, the at least one cluster is used as the at least one node cluster.

[0077] Optionally, the detection module is specifically used to:

[0078] If there are pre-marked plug-in nodes in the node cluster, obtaining the proportion of the plug-in nodes in the node cluster;

[0079] If the proportion of the external nodes is less than the preset proportion, obtaining the first feature of the external nodes and the second features of other nodes in the initial cluster except the external nodes;

[0080] Performing similarity analysis based on the first feature and the second feature to obtain similarity between the plug-in node and the other nodes;

[0081] If the similarity is greater than or equal to a preset similarity, the node cluster is determined to be an external cluster.

[0082] Optionally, the detection module is further used to:

[0083] If the proportion of the external nodes is greater than or equal to a preset proportion, the node cluster is determined to be an external cluster.

[0084] In the third aspect, another embodiment of the present application provides a server, comprising: a processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the server is running, the processor communicates with the storage medium through the bus, and the processor executes the machine-readable instructions to perform any method described in the first aspect above.

[0085] In a fourth aspect, another embodiment of the present application provides a storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in any one of the first aspects above is executed.

[0086] The present application provides a method, device, server and storage medium for detecting game plug-ins, wherein the method comprises: constructing a transaction map according to a game transaction log, the game transaction log comprises: log data of virtual asset transactions conducted by multiple game characters in the game, the transaction map comprises: multiple nodes, and transaction relationships between nodes, wherein each node corresponds to a game character, and according to the transaction map, a preset structural vector algorithm is used for feature representation to obtain a feature vector of each node, and multiple nodes are clustered according to the feature vectors of multiple nodes to obtain at least one node cluster, and the nodes in each node cluster are detected for plug-ins according to the features of the nodes in each node cluster. Since plug-in nodes have certain regularity in asset transactions, the detection of game plug-ins based on the transaction map can improve the accuracy of game plug-in detection, and compared with the prior art, plug-in detection has a wider coverage, a higher recall rate, and a lower labor cost, and does not affect the game performance of the player. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0088] Figure 1 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 1 ;

[0089] Figure 2 A schematic diagram of a transaction graph provided by an embodiment of the present application is shown;

[0090] Figure 3 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 2 ;

[0091] Figure 4 A schematic diagram of a K-layer fully connected network provided in an embodiment of the present application is shown;

[0092] Figure 5 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 3 ;

[0093] Figure 6 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 4 ;

[0094] Figure 7 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 5 ;

[0095] Figure 8 A schematic diagram of a node cluster provided in an embodiment of the present application is shown;

[0096] Fig. 9 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 6 ;

[0097] Fig.10 A schematic diagram of the structure of a game cheat detection device provided in an embodiment of the present application is shown;

[0098] Fig.11 A schematic diagram of the structure of a server provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0099] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0100] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0101] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0102] Before accepting the technical solution of this application, the professional terms involved in this application are first explained:

[0103] MMORPG: The abbreviation of Massive Multiplayer Online Role-Playing Game, which refers to a massively multiplayer online role-playing game, is a type of computer network game.

[0104] WinAPI: The abbreviation of Windows API, which means Windows operating system application program interface, is the interface for computer programs to operate Windows kernel applications.

[0105] DLL: The abbreviation of Dynamic Link Library, which means dynamic link library, contains some functions required by the application. DLL injection is a common means of plug-in.

[0106] HOOK: Chinese name: hook, it is a platform for Windows message processing. When the program calls WinAPI, each hook will be executed in sequence according to the order of the hook list. Using HOOK is a common method in plug-in production.

[0107] Struc2vec: An embedding algorithm for modeling the local structure of nodes in a graph, which can map the nodes in the graph into fixed-length vectors in a high-dimensional space.

[0108] Hierarchical Softmax: A highly efficient model proposed to solve the problem of poor computational performance of the Softmax model when the amount of data is too large.

[0109] DTW: abbreviation of Dynamic Time Warpingy, dynamic time warping algorithm, an algorithm for calculating sequence similarity based on dynamic programming.

[0110] SkipGram: An algorithm for completing vector representations of words in text, which can map words with similar contexts into adjacent spaces.

[0111] Node2vec: A graph embedding algorithm based on random walk and SkipGram that can map nodes in a graph into fixed-length vectors in a high-dimensional space.

[0112] DBSCAN: A density-based clustering algorithm that can divide areas with sufficient density into clusters and ensure density transitivity within the clusters. DBSCAN has the following advantages. 1) There is no need to pre-specify the number of clusters. In the case of plug-in detection, developers do not need to pre-limit the types and prevalence of plug-ins. 2) DBSCAN is a density-based clustering algorithm that can find high-density and transitive blocks of any shape, so it is not limited by the meaning of character features. 3) The algorithm can effectively distinguish noise (outliers), thereby ensuring the effectiveness of plug-in group clustering.

[0113] T-SNE: abbreviation of T-distributed Stochastic Neighbor Embedding, a data dimensionality reduction algorithm based on T distribution and random neighbor embedding.

[0114] k-order neighbors: Given a node v in a graph, the set of all nodes whose shortest path length from node v is k is called the k-order neighbors of node v.

[0115] Graph: English name: graph, refers to a data structure consisting of a series of nodes and directed or undirected connections between nodes. A graph is also often called a network.

[0116] Degree: The number of nodes directly connected to a node v in a graph is called the degree of node v. In a directed graph, it can be further divided into in-degree and out-degree, which respectively represent the number of nodes pointing to the node and the number of nodes pointed to by the node.

[0117] Complete graph: A graph in which all edges between any nodes exist is called a complete graph, also often called a fully connected graph. For example, an undirected complete graph contains n nodes and n(n-1) / 2 edges; a directed complete graph contains n nodes and n(n-1) edges.

[0118] Then, several game cheat detection methods provided by the prior art are described:

[0119] First, the game client records various features of the plug-in process and the feature code generated by the memory code fragment through process inspection. When a process with the same features is found during the running of the game process, it will be banned. However, as the plug-in program is updated, the above-mentioned same features will also change. The use of this solution makes the accuracy of plug-in detection low.

[0120] The second type is based on client file inspection. The game client continuously scans game files and system files to see if these files have been modified. When an unknown program file is found, it is uploaded to the server to analyze whether there are dangerous behaviors and decide whether to ban it. However, since plug-in developers can be independent of the game client and only implement automatic plug-in through automatic clicking and other methods that simulate human operations, they can circumvent inspections. In addition, protection measures based on client files can easily cause a decline in game performance.

[0121] The third type is protection based on the Windows operating system application program interface (Windows API, WinAPI). The game client actively uses the hook mechanism to intervene in key WinAPIs and enable anti-debugging to prevent plug-ins from hooking these WinAPIs to implement related functions. However, the Windows system version is complicated, and the active HOOK solution is prone to system crashes and blue screens during system updates, affecting the user experience.

[0122] The fourth method is to intervene in operations for manual verification, use empirical features for statistical analysis, and check players based on these empirical features. However, this solution requires high manpower costs and relies on people's understanding of plug-ins and games.

[0123] The fifth method is anomaly detection based on graph analysis. It builds a relationship network for the game data logged in by users and uses traditional graph segmentation techniques to identify abnormal individual groups. However, graph segmentation algorithms generally have high time complexity and are greatly affected by the connectivity within the graph. For individuals that are similar but not directly related in the graph, there is a lack of migration recognition ability, which means that the accuracy of game cheat detection is not high.

[0124] The sixth method is based on modeling the player's in-game behavior sequence. By collecting and analyzing the character behavior logs, a representation vector of the behavior pattern is constructed to complete the identification of plug-ins. However, this solution only uses individual behavior data and lacks the ability to analyze and model the relationship between characters, resulting in low accuracy in game plug-in detection.

[0125] Based on the above problems, the applicant found that in MMORPG games, automatic plug-ins use some program scripts to automatically perform tasks, kill monsters, upgrade, and make money, and transfer and aggregate in-game property, and aggregate game property to a few "sponsor accounts" in a hierarchical manner, and then trade with the outside world for profit. Based on this aggregation model, it can be seen that plug-ins have certain regularity in property transfer, that is, in different plug-in groups, "plug-in accounts" or "sponsor accounts" of the same level have a higher structural similarity. Therefore, this application realizes the detection of game plug-ins based on a transaction graph, wherein the transaction graph includes: multiple nodes, and transaction relationships between nodes, wherein each node corresponds to a game character, thereby improving the accuracy of game plug-in detection and reducing labor costs compared to the above-mentioned existing solutions.

[0126] The following describes in detail the game plug-in detection method provided in the embodiment of the present application in combination with the above-mentioned several specific embodiments.

[0127] Figure 1 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 1 , the execution subject of this embodiment may be a game server. Figure 1 As shown, the method includes:

[0128] S101. Construct a transaction graph based on a game transaction log. The game transaction log includes: log data of virtual asset transactions conducted by multiple game characters in the game. The transaction graph includes: multiple nodes and transaction relationships between nodes, wherein each node corresponds to a game character.

[0129] Among them, the game client can generate a game log when running the game application and send the game log to the game server. The game server can store the game log in a preset log library. It should be noted that the game client can send the game log to the server in real time or at a fixed time, and this embodiment does not specifically limit this.

[0130] The game application may be any massively multiplayer online role-playing game (MMORPG). In order to perform plug-in detection on the game application, the game server may obtain the game log sent by the game client from a preset log library, and then obtain the game transaction log from the game log. The game transaction log includes: log data of virtual asset transactions performed by multiple game characters in the game, that is, the game log includes game transaction logs of virtual asset transactions performed by game characters sent by multiple game clients, wherein the virtual asset may be, for example, virtual currency paid by game character A to game character B when purchasing virtual items from game character B.

[0131] In this embodiment, a transaction graph is constructed based on the game transaction log. The transaction graph includes multiple nodes and transaction relationships between nodes. Each node corresponds to a game character. The edges between the nodes represent the transaction relationships between the nodes. The edges between the nodes can be directed edges. For example, if game character A pays virtual currency to game character B, the arrow of the edge between the node corresponding to game character A and the node corresponding to game character B points to game character B.

[0132] It should be noted that the log data of virtual asset transactions between multiple game characters in the game includes transaction data and transaction item data. Table 1 is a schematic diagram of the transaction database, and Table 2 is a schematic diagram of the transaction items.

[0133]

[0134]

[0135] Table 1

[0136] Fields explain ItemID Item ID ItemValue Item Value ItemType Item Type

[0137] Table 2

[0138] It should be noted that the value of an item can be estimated manually by game operation experts, or it can be estimated by the transaction price of the item in the game's virtual mall within a certain period of time. A transaction map can be constructed based on the above transaction data and transaction item data.

[0139] Since the edges in the transaction graph are used to represent the transaction relationship between nodes, in order to more effectively explore the property aggregation phenomenon, the transaction graph can also be edge filtered according to the preset transaction relationship.

[0140] Among them, the preset transaction relationship can be that the number of transactions is less than or equal to the preset number of transactions, and / or the actual value difference is less than or equal to the preset value difference. That is to say, the edges in the transaction graph that satisfy the preset transaction relationship are filtered out, and only the edges whose number of transactions is greater than the preset number of transactions and whose actual value difference exceeds the preset value difference are retained.

[0141] The actual value difference refers to the actual transfer value of each transaction in Table 1 calculated using the transaction item data in Table 2. For example, if ID1 purchased an item worth 200 copper coins from ID2, but only gave 100 copper coins, the actual value difference of the transaction is 100 copper coins.

[0142] Figure 2 A schematic diagram of a transaction map provided by an embodiment of the present application is shown, such as Figure 2As shown, each node (represented by a dot) in the transaction graph represents a game character, and the edge between the nodes indicates that the two nodes have a transaction relationship. The edge between the nodes is a directed edge. For example, if game character A pays virtual currency to game character B, the direction of the arrow can be from game character A to game character B.

[0143] S102. According to the transaction graph, a preset structural vector algorithm is used for feature representation to obtain a feature vector for each node.

[0144] S103 . Clustering multiple nodes according to feature vectors of the multiple nodes to obtain at least one node cluster.

[0145] The preset structural vector algorithm is used to map the nodes in the graph into fixed-length vectors in a high-dimensional space. According to the transaction graph, the preset structural vector algorithm can be used to represent the features of each node in the transaction graph to obtain the feature vector of each node.

[0146] Then, the feature vectors of multiple nodes in the transaction graph are clustered to obtain at least one node cluster. That is to say, nodes with a certain similarity can be aggregated into a node cluster according to the feature vectors of the nodes. Since in MMORPG games, "plug-in accounts" of the same level have a certain regularity, therefore, the node clusters obtained by clustering the multiple nodes according to the feature vectors of the multiple nodes can be preliminarily considered to be plug-in clusters, that is, the nodes in the node cluster are all plug-in nodes.

[0147] It should be noted that the preset structure vector algorithm can be improved based on the Struc2vec algorithm.

[0148] S104: Perform plug-in detection on the nodes in each node cluster according to the characteristics of the nodes in each node cluster.

[0149] In this embodiment, after obtaining at least one node cluster, in order to further identify whether the nodes in each node cluster are plug-in nodes, plug-in detection can also be performed on the nodes in each node cluster according to the characteristics of the nodes in each node cluster to identify plug-in nodes from each node cluster.

[0150] Among them, the characteristics of the node can be the portrait characteristics of the game character corresponding to the node. The portrait characteristics may include: basic information and statistical information of the character. The basic information may include account information, character name, online time, recharge amount, etc. The statistical information may include gameplay, average daily participation times, average daily chat times, etc.

[0151] In one possible implementation, the server pre-saves the features of the plug-in node, and then calculates the similarity between the features of the node in each node cluster and the features of the plug-in node. If the similarity is greater than or equal to the preset similarity, the node is considered to be a plug-in node. If the similarity is less than the preset similarity, the node is considered not to be a plug-in node.

[0152] The game plug-in detection method provided in this embodiment includes: constructing a transaction graph according to the game transaction log, the game transaction log includes: log data of virtual asset transactions conducted by multiple game characters in the game, the transaction graph includes: multiple nodes and transaction relationships between nodes, each node corresponds to a game character, according to the transaction graph, using a preset structural vector algorithm for feature representation to obtain a feature vector of each node, clustering multiple nodes according to the feature vectors of multiple nodes to obtain at least one node cluster, and performing plug-in detection on the nodes in each node cluster according to the features of the nodes in each node cluster. Since plug-in nodes have certain regularity in asset transactions, the detection of game plug-ins based on the transaction graph can improve the accuracy of game plug-in detection and reduce labor costs compared with the prior art.

[0153] Optionally, the preset structural vector algorithm includes: structural distance algorithm, network construction algorithm, and walk-based vector representation algorithm. Figure 2 The embodiment describes how to obtain a feature vector of each node by performing feature representation based on a transaction graph using a preset structural vector algorithm.

[0154] Figure 3 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 2 ,like Figure 3 According to the transaction graph, a preset structural vector algorithm is used for feature representation to obtain a feature vector of each node, including:

[0155] S201. According to the degree sequence of every two nodes in the transaction graph, a structural distance algorithm is used to calculate the k-order structural distance between every two nodes. The k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes.

[0156] The structural distance algorithm is used to calculate the structural distance between two nodes. Generally speaking, the larger the structural distance, the smaller the structural similarity, and the smaller the structural distance, the greater the structural similarity.

[0157] The degree sequence of a node can be a sequence of the degrees of the k-order neighbor nodes of the node in the transaction graph. Generally, if the degrees of two nodes are the same, the node structures of the two nodes are similar. If the degrees of the neighbors of the two nodes are also the same, the node structure of the two nodes is higher than the former. Therefore, in this embodiment, the degree sequence of the k-order neighbor nodes is taken into consideration. According to the degree sequence of every two nodes in the transaction graph, the k-order structural distance of every two nodes can be calculated by using the structural distance algorithm. The k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes. In other words, the k-order structural distance is used to reflect the structural similarity between the k-order neighbor nodes of the two nodes.

[0158] S202. According to the k-order structural distance between every two nodes, a network construction algorithm is used to construct a k-th layer network, and a K-layer fully connected network with a K-layer network is obtained. The K-layer fully connected network includes: the edge weights of every two nodes in each layer of the network, and the edge weights of the same node in adjacent layer networks.

[0159] Among them, a fully connected network can also be called a complete graph, which refers to a graph in which all edges between any nodes exist. A K-layer fully connected network means that each node is replicated K times, that is, the nodes in each layer of the fully connected network are the same. It should be noted that the nodes in each layer of the fully connected network are all the nodes in the transaction graph.

[0160] Figure 4 A schematic diagram of a K-layer fully connected network provided in an embodiment of the present application is shown, Figure 4 As shown, each layer of the fully connected network includes four nodes, with a total of K layers.

[0161] The K-layer fully connected network can be a K-layer fully connected graph or a K-layer weighted network. That is, the K-layer fully connected network includes: the edge weights of every two nodes in each layer of the network, and the edge weights of the same node in the adjacent layer of the network. The edge weights of every two nodes in each layer of the network are as follows: Figure 4 The solid lines shown in Figure 2 show the edges of the same node in adjacent networks. Figure 4 Dotted line shown.

[0162] In this embodiment, the network construction algorithm is used to construct a k-th layer fully connected network using the k-order structural distance between every two nodes. That is to say, the k-order structural distance between every two nodes has a certain corresponding relationship with the k-th layer network in the K-layer fully connected network, and then based on the constructed k-th layer network, a K-layer fully connected network with K-layer networks can be obtained, where k is 0 or any positive integer less than K.

[0163] S203. According to the K-layer fully connected network, a walk-based vector representation algorithm is used to obtain a feature vector of each node in the K-th layer network.

[0164] Among them, the walk-based vector representation algorithm includes a random walk algorithm and a training language model. For each node, the random walk algorithm is used to obtain the walk sequence of the node in the K-layer fully connected network. The walk sequence includes the nodes that the node passes through when randomly walking in the K-layer fully connected network, thereby obtaining multiple walk sequences, and then the obtained multiple walk sequences are used as input, and the training language model is used to obtain the vector of each node in each layer of the network, and the vector of each node in the K-th layer of the network is used as the feature vector of the node.

[0165] Among them, the K-layer fully connected network includes: the edge weights of every two nodes in each layer of the network, and the edge weights of the same node in the adjacent layer of the network. Then, when performing a random walk, if there are two walk directions, the node corresponding to the edge with the larger edge weight is used as the next walk node, for example Figure 3 Node A in the figure can move to node B and node C, but since the edge weight between node A and node B is 0.9, and the edge weight between node A and node C is 0.1, the next node to move to is node B.

[0166] It should be noted that the training language model may be a language model based on Hierarchical Softmax and SkipGram, and the dimension of the output vector may be pre-set, for example, it may be 128 dimensions. For details, please refer to the relevant description in the prior art.

[0167] It should be noted that the degree may include the in-degree and the out-degree. Therefore, when calculating the sequence distance between the first node and the second node, two K-layer fully connected networks can be constructed according to the in-degree sequence and the out-degree sequence respectively according to steps S201-S202. Therefore, in step S203, two feature vectors of the same node can be obtained, and then the two feature vectors are concatenated to obtain the feature vector of the node.

[0168] The game cheat detection method provided in this embodiment includes: according to the degree sequence of every two nodes in the transaction graph, using the structural distance algorithm to calculate the k-order structural distance of every two nodes, the k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes, according to the k-order structural distance of every two nodes, using the network construction algorithm to construct the k-th layer network, and obtain a K-layer fully connected network with K-layer networks, the K-layer fully connected network includes: the edge weights of every two nodes in each layer of the network, and the edge weights of the same node in the adjacent layer network, according to the K-layer fully connected network, using the walk-based vector characterization algorithm to obtain the feature vector of each node in the K-th layer network. Developers do not need to have strong expert experience in games, and the feature vector of the node can be obtained by using an intelligent algorithm, which improves the accuracy of the feature vector.

[0169] Optionally, the following combination Figure 5The embodiment describes how to calculate the k-order structural distance between every two nodes according to the degree sequence of every two nodes in the transaction graph using a structural distance algorithm. Figure 5 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 3 ,like Figure 5 As shown, according to the degree sequence of every two nodes in the transaction graph, the structural distance algorithm is used to calculate the k-order structural distance between every two nodes, including:

[0170] S301 , obtaining a first degree sequence of a first node and a second degree sequence of a second node in every two nodes.

[0171] S302: Calculate the sequence distance between the first degree sequence and the second degree sequence.

[0172] S303. Calculate the k-order structural distance between the first node and the second node according to the sequence distance and the structural distance of the previous-order neighboring nodes, wherein the previous-order neighboring nodes include: the previous-order neighboring nodes of the first node and the previous-order neighboring nodes of the second node in the transaction graph.

[0173] Among them, every two nodes include a first node and a second node, the first degree sequence of the first node is a sequence composed of the degrees of the first node's k-order neighbor nodes, and the second degree sequence of the second node is a sequence composed of the degrees of the second node's k-order neighbor nodes.

[0174] In this embodiment, the first degree sequence of the first node and the second degree sequence of the second node of every two nodes are obtained, the sequence distance of the first degree sequence and the second degree sequence is calculated, and then the k-order structural distance of the first node and the second node is calculated based on the sequence distance and the structural distance of the previous order neighbor node. The k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes, wherein the previous order neighbor nodes include: the previous order neighbor node of the first node and the previous order neighbor node of the second node in the transaction graph.

[0175] That is to say, the k-order structural distance between the first node and the second node depends on the structural distance of the previous-order neighboring nodes.

[0176] In a possible implementation, the k-order structural distance between the first node and the second node may be calculated according to the following formula:

[0177] f k (u,v)=f k-1 (u,v)+g(s(R k (u)),s(R k (v)))

[0178] Among them, f k (u,v) is the k-order structural distance between node u and node v, fk-1 (u,v) is the structural distance between node u and node v’s previous-order neighbor nodes, s(R k (u)) is the degree sequence of node u, s(R k (v)) is the degree sequence of node v, g(s(R k (u)),s(R k (v))) is s(R k (u)) and s(R k (v)) sequence distance. Among them, |R k (u)|>0,|R k (v)|>0,f -1 =0.

[0179] Optionally, obtaining a first degree sequence of a first node and a second degree sequence of a second node in every two nodes includes:

[0180] Arrange the degrees of the k-stage neighbor nodes of the first node in ascending order to obtain the first degree sequence.

[0181] Arrange the degrees of the k-stage neighbor nodes of the second node in ascending order to obtain the second degree sequence.

[0182] In one possible implementation, the degrees of the k-stage neighbor nodes of the first node can be sorted in ascending order to obtain a first degree sequence, and the degrees of the k-stage neighbor nodes of the second node can be sorted in ascending order to obtain a second degree sequence. That is, the degrees of the nodes whose path length from the first node is k are sorted in ascending order to obtain the first degree sequence, and the degrees of the nodes whose path length from the second node is k are sorted in ascending order to obtain the second degree sequence.

[0183] It should be noted that the path length between adjacent nodes is 1, and the path length between two nodes separated by one node is 2, that is, one edge represents a path length of 1, and two edges represent a path length of 2.

[0184] Optionally, calculating the sequence distance between the first degree sequence and the second degree sequence includes:

[0185] According to the first-degree sequence and the second-degree sequence, the sequence distance is calculated using the dynamic time normalization algorithm.

[0186] Among them, the Dynamic Time Warping (DTW) algorithm is an algorithm for calculating sequence similarity based on dynamic programming, that is, taking the first-degree sequence and the second-degree sequence as input, the dynamic time warping algorithm is used to calculate the sequence distance between the first-degree sequence and the second-degree sequence.

[0187] It should be noted that the core of the DTW algorithm is to first calculate the element distance between every two elements in the two sequences, and then calculate the sequence distance of the two sequences based on the calculated multiple element distances. The element distance is calculated as follows:

[0188]

[0189] Where a is any element in the first degree sequence, b is any element in the second degree sequence, max(a, b) is the maximum value of a and b, min(a, b) is the minimum value of a and b, and d(a, b) is the element distance between a and b.

[0190] The game cheat detection method provided in this embodiment includes: obtaining the first degree sequence of the first node and the second degree sequence of the second node in every two nodes, calculating the sequence distance of the first degree sequence and the second degree sequence, and calculating the k-order structural distance of the first node and the second node according to the sequence distance and the structural distance of the previous order neighbor node, wherein the previous order neighbor node includes: the previous order neighbor node of the first node and the previous order neighbor node of the second node in the transaction graph. Thus, the structural distance of the nodes can be calculated, and then the structural similarity of the two nodes can be reflected, which is helpful for subsequent cheat detection based on the structural similarity of the nodes.

[0191] Combine the following Figure 6 The acquisition process of the K-layer fully connected network is described in the embodiment. Figure 6 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 4 ,like Figure 6 As shown, according to the k-order structural distance between every two nodes, a network construction algorithm is used to construct the k-th layer network, and a K-layer fully connected network with a K-layer network is obtained, including:

[0192] S401. Calculate the k-order edge weights between every two nodes in the k-th layer network according to the k-order structural distance.

[0193] There is a certain correspondence between the k-th layer network in the K-layer fully connected network and the k-th layer structure distance. Therefore, the k-th layer edge weight between every two nodes in the k-th layer network can be calculated based on the k-th layer structure distance. For example, the following formula can be used for calculation:

[0194]

[0195] Among them, w k (u,v) is the k-order edge weight between node u and node v, f k (u,v) is the k-order structural distance between node u and node v.

[0196] Among them, the K-layer fully connected network also includes: the edge weight of the same node in the adjacent layer network. It is also necessary to obtain the k+1-order edge weight of the same node in the k-th layer network and the k+1-th layer network, and the k-1-order edge weight of the same node in the k+1-th layer network and the k-th layer network, which is explained below with reference to steps S402-S403.

[0197] S402. Calculate the k+1-order edge weight of the same node in the k-th network and the k+1-th network according to the number of nodes whose edge weights pointing to each node in the k-th network satisfy a preset weight condition.

[0198] The preset weight condition may be greater than or equal to a preset weight value.

[0199] Alternatively, the k+1-order edge weight can be calculated by referring to the following formula:

[0200] w(u k ,u k+1 ) = log(Γ k (u)+e)

[0201] Among them, w(u k ,u k+1 ) is the k+1-order edge weight between node u in the k-th layer network and the k+1-th layer network, Γ k (u) is the number of nodes in the k-th layer network whose edge weights pointing to node u satisfy the preset weight conditions, and e is a constant.

[0202] Optionally, satisfying a preset weight condition includes: being greater than or equal to an average edge weight of the k-th layer network.

[0203] Among them, the average edge weight is the average value of all edge weights in the k-th layer network. That is to say, based on the number of nodes in the k-th layer network whose edge weight is greater than or equal to the average edge weight of the k-th layer network and points to each node, the k+1-order edge weight of the same node in the k-th layer network and the k+1-th layer network can be calculated.

[0204] S403: construct a K-layer fully connected network according to the k-order edge weight, the k+1-order edge weight, and the k-1-order edge weight, wherein the k-1-order edge weight is a preset weight.

[0205] The k+1-order edge weight may be a preset weight, for example, 1, as shown in the following formula:

[0206] w(u k+1 ,u k )=1

[0207] Among them, w(u k+1 ,u k) is the k-1-order edge weight between the node u in the k-th network and the k+1-th network. That is to say, the edge weight between the k-th network and the same node in the k+1-th network and the edge weight between the k+1-th network and the k-th network are directional. The edge weight from the node u in the k-th network to the node u in the k+1-th network is calculated, while the edge weight from the node u in the k+1-th network to the node u in the k-th network is a known preset weight.

[0208] According to the k-order edge weights between every two nodes in each layer of the network, the k+1-order edge weights between the same node in the k-th layer network and the k+1-th layer network, and the k-1-order edge weights between the same node in the k+1-th layer network and the k-th layer network, the entire K-layer fully connected network can be constructed.

[0209] The game cheat detection method provided in this embodiment includes: calculating the k-order edge weight of every two nodes in the k-th layer network according to the k-order structural distance, calculating the k+1-order edge weight of the same node in the k-th layer network and the k+1-th layer network according to the number of nodes whose edge weights pointing to each node in the k-th layer network meet the preset weight conditions, and constructing a K-layer fully connected network according to the k-order edge weight, k+1-order edge weight and k-1-order edge weight, wherein the k-1-order edge weight is a preset weight. Thus, a K-layer fully connected network can be constructed according to the edge weights of each layer of the network and the edge weights of the adjacent networks, which is helpful for the subsequent acquisition of the feature vector of the node.

[0210] Optionally, the following combination Figure 7 The embodiment illustrates a process of acquiring at least one node cluster. Figure 7 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 5 ,like Figure 7 As shown, multiple nodes are clustered according to their feature vectors to obtain at least one node cluster, including:

[0211] S501. Calculate the node distance between every two nodes according to the feature vectors of multiple nodes.

[0212] S502: Cluster multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster.

[0213] The node distance between every two nodes can be calculated based on the feature vectors of multiple nodes. The method of calculating the node distance based on the feature vector is similar to the prior art and will not be described in detail here.

[0214] Then, according to the calculated node distance between every two nodes, multiple nodes are clustered. For example, nodes whose node distance is less than a preset distance may be clustered into one category, thereby obtaining at least one initial cluster.

[0215] Optionally, clustering the multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster includes:

[0216] If there is at least one pair of target nodes between every two nodes, cluster the nodes within the preset neighborhood range where each pair of target nodes are located to obtain at least one initial cluster, and the node distance of each pair of target nodes is less than or equal to the preset neighborhood value.

[0217] If there is at least one pair of target nodes between every two nodes, the nodes within the preset neighborhood range where each pair of target nodes are located are clustered, that is, a preset neighborhood value is set in advance, and if there is at least one pair of target nodes between every two nodes whose node distance is less than or equal to the preset neighborhood value, the nodes within the preset neighborhood range where each pair of target nodes are located are aggregated into one category, that is, only when the node distance is less than or equal to the preset neighborhood value, the nodes within the preset neighborhood range where the target nodes are located are clustered to obtain at least one initial cluster.

[0218] Optionally, the preset neighborhood value may be 1.7, and the preset neighborhood range may be the range of a circle with the midpoint of each pair of target nodes as the center and a radius equal to the preset neighborhood value.

[0219] S503 . Acquire at least one node cluster from at least one initial cluster according to the number of nodes in each initial cluster.

[0220] In order to further determine whether the nodes in each initial cluster are external nodes, it is also necessary to further determine whether each initial cluster is a cluster that meets a condition, wherein the condition may be, for example, that the number of nodes is greater than or equal to a preset number.

[0221] Optionally, obtaining at least one node cluster from at least one initial cluster according to the number of nodes in each initial cluster includes:

[0222] If there is at least one cluster in the at least one initial cluster where the number of nodes is greater than or equal to a preset number, the at least one cluster is used as at least one node cluster.

[0223] Even if an initial cluster is formed, it is still necessary to determine whether it forms a node cluster based on the number of nodes in the initial cluster. If there is at least one cluster in the initial cluster whose number of nodes is greater than or equal to the preset number, then the at least one cluster is regarded as at least one node cluster. In other words, the number of nodes in each node cluster is greater than or equal to the preset number.

[0224] Optionally, the preset number may be 30.

[0225] It should be noted that since the feature vectors of the nodes in each node cluster are high-dimensional, each node can also be visualized in terms of dimensions. Figure 8A schematic diagram of a node cluster provided by an embodiment of the present application is shown, Figure 8 As shown, the neighborhood size is set to 1.7, and the number of nodes in the node cluster can be set to 30. Figure 8 The node clusters identified in the figure are clearly distinguishable from other initial clusters (labeled as cluster 1, cluster 2, and cluster 3, respectively).

[0226] It should be noted that steps S601-S603 may be executed using the DBSCAN algorithm.

[0227] The game cheat detection method provided in this embodiment includes: calculating the node distance between every two nodes according to the feature vectors of multiple nodes, clustering the multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster, and obtaining at least one node cluster from the at least one initial cluster according to the number of nodes in each initial cluster. Thus, the node cluster can be obtained through the intelligent algorithm, which is helpful for the subsequent detection of cheat nodes according to the node cluster.

[0228] Optionally, for Figure 8 At least one node cluster in the embodiment may further perform a plug-in detection on the nodes in each node cluster. Fig. 9 The embodiments are described below. Fig. 9 The following is a schematic diagram showing the process of the game cheat detection method provided in the embodiment of the present application. Figure 6 ,like Fig. 9 As shown, according to the characteristics of the nodes in each node cluster, the nodes in each node cluster are detected for plug-in, including:

[0229] S601: If there are pre-marked plug-in nodes in the node cluster, obtain the proportion of the plug-in nodes in the node cluster.

[0230] Among them, the pre-marked plug-in node can be a pre-recorded plug-in node, for example, the account of the plug-in node detected before can be saved. When the node cluster is obtained, the account of the pre-saved plug-in node can be queried to determine whether there is a pre-marked plug-in node in the node cluster. If there is a pre-marked plug-in node in the node cluster, the proportion of plug-in nodes in the node cluster is obtained. The proportion of plug-in nodes is the proportion of plug-in nodes in the node cluster. If the node cluster includes 30 nodes and there are 15 plug-in nodes, it means that the proportion of plug-in nodes is 50%.

[0231] S602: If the proportion of the plug-in nodes is less than a preset proportion, obtain the first features of the plug-in nodes and the second features of other nodes in the initial cluster except the plug-in nodes.

[0232] S603: Perform similarity analysis based on the first feature and the second feature to obtain similarities between the plug-in node and other nodes.

[0233] S604: If the similarity is greater than or equal to a preset similarity, determine that the node cluster is an external cluster.

[0234] The first feature of the plug-in node is the portrait feature of the plug-in node, and the second feature of the other nodes in the initial cluster except the plug-in node is the portrait feature of the other nodes. The portrait feature may include: basic information and statistical information of the role, the basic information may include account information, role name, online time, recharge amount, etc., and the statistical information may include gameplay, average daily participation times, average daily chat times, etc.

[0235] If the proportion of plug-in nodes is less than the preset proportion, the first feature of the plug-in nodes and the second feature of other nodes are further obtained, and a similarity analysis is performed on the first feature and the second feature, that is, the similarity between the first feature and the second feature is calculated. If the similarity is greater than or equal to the preset similarity, the node cluster is considered to be a plug-in cluster, that is, all the nodes in the node cluster are plug-in nodes.

[0236] It should be noted that the number of the plug-in nodes is at least one, so when calculating the similarity, the average feature of the first feature of the plug-in node can be taken, and the similarity between the average feature and the second feature of each other node can be calculated.

[0237] Optionally, if the proportion of external nodes is greater than or equal to a preset proportion, the method further includes:

[0238] S605: If the proportion of external nodes is greater than or equal to a preset proportion, determine that the node cluster is an external cluster.

[0239] That is to say, if the proportion of plug-in nodes is greater than or equal to the preset proportion, all nodes in the node cluster are plug-in nodes.

[0240] Of course, the account information of the detected plug-in nodes can also be saved in the MySQL database. At the same time, all the portrait features and clustering results detected as plug-ins can be saved in the form of files and stored according to the File Transfer Protocol (FTP) to facilitate the subsequent verification of the results and batch bans by game operators.

[0241] It should be noted that the portrait features of the plug-in node can also be displayed through a visualization tool. This visualization tool can present the difference between the current node and other nodes in a visual way, providing evidentiary support for determining whether a node uses a plug-in. It can also help verify the accuracy of the output results, completing the closed-loop logic of plug-in detection.

[0242] The game cheat detection method provided in this embodiment includes: if there are pre-marked cheat nodes in the node cluster, then obtain the proportion of the cheat nodes in the node cluster; if the proportion of the cheat nodes is less than the preset proportion, then obtain the first feature of the cheat nodes and the second feature of other nodes in the initial cluster except the cheat nodes; perform similarity analysis based on the first feature and the second feature to obtain the similarity between the cheat nodes and other nodes; if the similarity is greater than or equal to the preset similarity, then determine that the node cluster is a cheat cluster; if the proportion of the cheat nodes is greater than or equal to the preset proportion, then determine that the node cluster is a cheat cluster. Thus, the cheat nodes in the node cluster can be detected based on the proportion of the cheat nodes and the characteristics of the nodes, thereby improving the accuracy of cheat detection.

[0243] Fig.10 The structure diagram of the game cheat detection device provided in the embodiment of the present application is shown as follows: Fig.10 As shown, the game cheat detection device 70 includes:

[0244] The processing module 701 is used to construct a transaction graph according to the game transaction log, wherein the game transaction log includes: log data of virtual asset transactions performed by multiple game characters in the game, and the transaction graph includes: the multiple nodes and transaction relationships between the nodes, wherein each node corresponds to a game character;

[0245] The acquisition module 702 is used to perform feature representation according to the transaction graph using a preset structural vector algorithm to obtain a feature vector of each node; cluster the multiple nodes according to the feature vectors of the multiple nodes to obtain at least one node cluster;

[0246] The detection module 703 is used to perform plug-in detection on the nodes in each node cluster according to the characteristics of the nodes in each node cluster.

[0247] Optionally, the preset structural vector algorithm includes: a structural distance algorithm, a network construction algorithm, and a walk-based vector representation algorithm;

[0248] The acquisition module 702 is specifically used for:

[0249] According to the degree sequence of every two nodes in the transaction graph, the structural distance algorithm is used to calculate the k-order structural distance between every two nodes, and the k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes;

[0250] According to the k-order structural distance between each two nodes, the network construction algorithm is used to construct the k-th layer network to obtain the K-layer fully connected network with K-layer networks, wherein the K-layer fully connected network includes: the edge weights of each two nodes in each layer of the network, and the edge weights of the same node in adjacent layers of the network; k is 0 or any positive integer less than K;

[0251] According to the K-layer fully connected network, the walk-based vector characterization algorithm is adopted to obtain the feature vector of each node in the K-th layer network.

[0252] Optionally, the acquisition module 702 is specifically configured to:

[0253] Obtain a first degree sequence of a first node and a second degree sequence of a second node in each of the two nodes;

[0254] calculating a sequence distance between the first degree sequence and the second degree sequence;

[0255] The k-order structural distance between the first node and the second node is calculated according to the sequence distance and the structural distance of the previous-order neighbor node, wherein the previous-order neighbor node includes: the previous-order neighbor node of the first node and the previous-order neighbor node of the second node in the transaction graph.

[0256] Optionally, the acquisition module 702 is specifically configured to:

[0257] Arrange the degrees of the k-stage neighbor nodes of the first node in ascending order to obtain the first degree sequence;

[0258] The degrees of the k-stage neighbor nodes of the second node are arranged in ascending order to obtain the second degree sequence.

[0259] Optionally, the acquisition module 702 is specifically configured to:

[0260] The sequence distance is calculated according to the first degree sequence and the second degree sequence using a dynamic time normalization algorithm.

[0261] Optionally, the acquisition module 702 is specifically configured to:

[0262] Calculating the k-order edge weights of each two nodes in the k-th layer network according to the k-order structural distance;

[0263] Calculate the k+1-order edge weight of the same node in the k-th network and the k+1-th network according to the number of nodes whose edge weights pointing to each node in the k-th network satisfy the preset weight condition;

[0264] The K-layer fully connected network is constructed according to the k-order edge weight, the k+1-order edge weight and the k-1-order edge weight, wherein the k-1-order edge weight is a preset weight.

[0265] Optionally, satisfying a preset weight condition includes: being greater than or equal to an average edge weight of the k-th layer network.

[0266] Optionally, the acquisition module 702 is specifically configured to:

[0267] Calculating the node distance between every two nodes according to the feature vectors of the multiple nodes;

[0268] Clustering the multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster;

[0269] At least one node cluster is obtained from the at least one initial cluster according to the number of nodes in each initial cluster.

[0270] Optionally, the acquisition module 702 is specifically configured to:

[0271] If there is at least one pair of target nodes between every two nodes, cluster the nodes within a preset neighborhood range where each pair of target nodes are located to obtain at least one initial cluster, and the node distance between each pair of target nodes is less than or equal to a preset neighborhood value.

[0272] Optionally, the acquisition module 702 is specifically configured to:

[0273] If there is at least one cluster in the at least one initial cluster whose number of nodes is greater than or equal to a preset number, the at least one cluster is used as the at least one node cluster.

[0274] Optionally, the detection module 703 is specifically configured to:

[0275] If there are pre-marked plug-in nodes in the node cluster, obtaining the proportion of the plug-in nodes in the node cluster;

[0276] If the proportion of the external nodes is less than the preset proportion, obtaining the first feature of the external nodes and the second features of other nodes in the initial cluster except the external nodes;

[0277] Performing similarity analysis based on the first feature and the second feature to obtain similarity between the plug-in node and the other nodes;

[0278] If the similarity is greater than or equal to a preset similarity, the node cluster is determined to be an external cluster.

[0279] Optionally, the detection module 703 is further configured to:

[0280] If the proportion of the external nodes is greater than or equal to a preset proportion, the node cluster is determined to be an external cluster.

[0281] For descriptions of the processing flow of each module in the device and the interaction flow between each module, reference may be made to the relevant descriptions in the above method embodiment, which will not be described in detail here.

[0282] Fig.11 The schematic diagram of the structure of the server provided in the embodiment of the present application is shown as follows: Fig.11 As shown, the server 80 includes: a processor 801, a storage medium 802 and a bus 803. The storage medium 802 stores machine-readable instructions executable by the processor 801. When the server 80 is running, the processor 801 communicates with the storage medium 802 through the bus 803, and the processor 801 executes the machine-readable instructions to execute the method of the above method embodiment.

[0283] An embodiment of the present application further provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method of the above method embodiment is executed.

[0284] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the system and device described above can refer to the corresponding process in the method embodiment, and will not be repeated in this application. In the several embodiments provided in this application, it should be understood that the disclosed system, device and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules 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 through some communication interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0285] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the function 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 invention is essentially or part of the technical solution that contributes to the prior art 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 several instructions for 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 invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, RandomAccess Memory), disk or optical disk and other media that can store program code.

[0286] The above are only specific implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for detecting game cheats, characterized in that: include: Constructing a transaction graph based on the game transaction log, wherein the game transaction log includes: log data of virtual asset transactions conducted by multiple game characters in the game, and the transaction graph includes: multiple nodes and transaction relationships between the nodes, wherein each node corresponds to a game character; According to the transaction graph, a preset structural vector algorithm is used for feature representation to obtain a feature vector of each node; Clustering the multiple nodes according to the feature vectors of the multiple nodes to obtain at least one node cluster; Performing plug-in detection on the nodes in each node cluster according to the characteristics of the nodes in the node cluster; The preset structural vector algorithms include: structural distance algorithm, network construction algorithm, and walk-based vector representation algorithm; The feature representation is performed using a preset structural vector algorithm according to the transaction graph to obtain the feature vector of each node, including: According to the degree sequence of every two nodes in the transaction graph, the structural distance algorithm is used to calculate the k-order structural distance between every two nodes, and the k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes; According to the k-order structural distance between each two nodes, the network construction algorithm is used to construct a k-th layer network to obtain a K-layer fully connected network with K-layer networks, wherein the K-layer fully connected network includes: edge weights of each two nodes in each layer of the network, and edge weights of the same node in adjacent layers of the network; k is 0 or any positive integer less than K; According to the K-layer fully connected network, the walk-based vector characterization algorithm is adopted to obtain the feature vector of each node in the K-th layer network.

2. The method according to claim 1, characterized in that The step of calculating the k-order structural distance between each two nodes using the structural distance algorithm according to the degree sequence of each two nodes in the transaction graph includes: Obtain a first degree sequence of a first node and a second degree sequence of a second node in every two nodes; calculating a sequence distance between the first degree sequence and the second degree sequence; The k-order structural distance between the first node and the second node is calculated according to the sequence distance and the structural distance of the previous-order neighbor node, wherein the previous-order neighbor node includes: the previous-order neighbor node of the first node and the previous-order neighbor node of the second node in the transaction graph.

3. The method according to claim 2, characterized in that The obtaining of the first degree sequence of the first node and the second degree sequence of the second node in each two nodes comprises: Arrange the degrees of the k-stage neighbor nodes of the first node in ascending order to obtain the first degree sequence; The degrees of the k-stage neighbor nodes of the second node are arranged in ascending order to obtain the second degree sequence.

4. The method according to claim 2, characterized in that: The calculating the sequence distance between the first degree sequence and the second degree sequence includes: The sequence distance is calculated according to the first degree sequence and the second degree sequence using a dynamic time normalization algorithm.

5. The method according to claim 1, characterized in that The method of constructing a k-th layer network by using the network construction algorithm according to the k-order structural distance between each two nodes to obtain the K-layer fully connected network having K-layer networks includes: Calculating the k-order edge weights of each two nodes in the k-th layer network according to the k-order structural distance; Calculate the k+1-order edge weight of the same node in the k-th network and the k+1-th network according to the number of nodes whose edge weights pointing to each node in the k-th network satisfy the preset weight condition; The K-layer fully connected network is constructed according to the k-order edge weight, the k+1-order edge weight and the k-1-order edge weight, wherein the k-1-order edge weight is a preset weight.

6. The method according to claim 5, characterized in that The preset weight condition is satisfied including: being greater than or equal to the average edge weight of the k-th layer network.

7. The method according to claim 1, characterized in that The clustering of the multiple nodes according to the feature vectors of the multiple nodes to obtain at least one node cluster includes: Calculating the node distance between every two nodes according to the feature vectors of the multiple nodes; Clustering the multiple nodes according to the node distance between every two nodes to obtain at least one initial cluster; At least one node cluster is obtained from the at least one initial cluster according to the number of nodes in each initial cluster.

8. The method according to claim 7, characterized in that The clustering of the plurality of nodes according to the node distance between every two nodes to obtain at least one initial cluster comprises: If there is at least one pair of target nodes between every two nodes, cluster the nodes within a preset neighborhood range where each pair of target nodes are located to obtain at least one initial cluster, and the node distance between each pair of target nodes is less than or equal to a preset neighborhood value.

9. The method according to claim 7 or 8, characterized in that: The acquiring at least one node cluster from the at least one initial cluster according to the number of nodes in each initial cluster comprises: If there is at least one cluster in the at least one initial cluster whose number of nodes is greater than or equal to a preset number, the at least one cluster is used as the at least one node cluster.

10. The method according to claim 7, characterized in that The performing cheat detection on the nodes in each node cluster according to the characteristics of the nodes in each node cluster includes: If there are pre-marked plug-in nodes in the node cluster, obtaining the proportion of the plug-in nodes in the node cluster; If the proportion of the external nodes is less than a preset proportion, obtaining a first feature of the external nodes and a second feature of other nodes in the initial cluster except the external nodes; Performing similarity analysis based on the first feature and the second feature to obtain similarity between the plug-in node and the other nodes; If the similarity is greater than or equal to a preset similarity, the node cluster is determined to be an external cluster.

11. The method according to claim 10, characterized in that The method further comprises: If the proportion of the external nodes is greater than or equal to a preset proportion, the node cluster is determined to be an external cluster.

12. A detection device for game cheats, characterized in that: include: A processing module, configured to construct a transaction graph according to a game transaction log, wherein the game transaction log includes: log data of virtual asset transactions conducted by multiple game characters in the game, and the transaction graph includes: multiple nodes and transaction relationships between the nodes, wherein each node corresponds to a game character; An acquisition module is used to perform feature representation according to the transaction graph using a preset structural vector algorithm to obtain a feature vector of each node; cluster the multiple nodes according to the feature vectors of the multiple nodes to obtain at least one node cluster; A detection module, used for performing plug-in detection on the nodes in each node cluster according to the characteristics of the nodes in each node cluster; The preset structural vector algorithms include: structural distance algorithm, network construction algorithm, and walk-based vector representation algorithm; The acquisition module is specifically used for: According to the degree sequence of every two nodes in the transaction graph, the structural distance algorithm is used to calculate the k-order structural distance between every two nodes, and the k-order structural distance is used to characterize the structural similarity of the k-order neighbor nodes of every two nodes; According to the k-order structural distance between each two nodes, the network construction algorithm is used to construct the k-th layer network to obtain the K-layer fully connected network with K-layer networks, wherein the K-layer fully connected network includes: the edge weights of each two nodes in each layer of the network, and the edge weights of the same node in adjacent layers of the network; k is 0 or any positive integer less than K; According to the K-layer fully connected network, the walk-based vector characterization algorithm is adopted to obtain the feature vector of each node in the K-th layer network.

13. A server, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the server is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to execute any of the methods described in claims 1-11.

14. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is executed.

Citation Information

Patent Citations

  • User mobile game behavior map-based game recommendation method

    CN106022877A

  • Knowledge graph-based case retrieval method, device and equipment, and storage medium

    CN111241241A

  • Network deep threat detection method

    CN111431865A

  • Network abnormal user detection method, device and equipment based on knowledge graph

    CN111949803A

  • Methods and apparatus for distributed gaming over a mobile device

    US20050288080A1