A method, device, electronic device and storage medium for detecting cheats in games

Abstract

The present application provides a method, device, electronic device, and storage medium for detecting cheats in a game, which obtain screen touch information of a player to be detected during the game operation process; the screen touch information includes the touch type of each touch operation completed in the game operation process and at least one touch position for completing the touch operation; according to the order in which each touch operation occurs in the game operation process, a touch type sequence and a touch position sequence are constructed, and the touch type sequence and the touch position sequence are input into a pre-trained cheat detection model, so that after feature extraction and feature fusion are performed based on the touch type sequence and the touch position sequence, the cheat suspicion of the touch behavior pattern of the player to be detected during the game operation process is output; if the regularity is greater than a preset threshold, the player to be detected is determined to be a cheating player. This helps to improve the accuracy of cheating player detection.

Description

Technical Field

[0001] The present application relates to the technical field of game cheat detection, and in particular to a method, device, electronic device and storage medium for detecting cheats in games. Background Art

[0002] In recent years, with the widespread adoption of mobile internet and smart devices, mobile gaming has developed rapidly, gradually becoming the largest and fastest-growing market segment within the gaming industry. However, with the development of mobile gaming, fairness issues have become increasingly prominent, with cheating being the most serious problem, harming the interests of both players and game developers. Simulated click cheats are the most common cheating method in mobile games. Using these tools, players can engage in behaviors such as bulk idle, automated tasks, automated resource collection, and automated reward claiming, significantly negatively impacting game fairness and the economic system.

[0003] When playing mobile games, players usually control their virtual characters by touching the screen of their mobile phone. Therefore, screen touch data is data that is easily collected on mobile devices such as mobile phones and tablets, and represents the player's finger operations on the device screen to control the virtual character when playing mobile games on mobile devices such as mobile phones. At present, the detection of game plug-ins is generally carried out through physical methods such as device information analysis, game environment detection, and abnormal game process detection for click-type plug-ins. Even the processing of screen touch data requires the detection personnel to identify the plug-in through the characteristics of the screen touch data, and the processing characteristics need to be compared one by one. Both of the above methods require the processing of large amounts of data. In addition, due to changes in the physical conditions themselves and the update of the characteristics of the screen touch data, the above comparison method will also have the problem of false detection and missed detection, resulting in low accuracy in the detection of plug-in players. Summary of the Invention

[0004] In view of this, the purpose of the present application is to provide a method, device, electronic device and storage medium for detecting cheats in games, by acquiring screen touch information and extracting a touch type sequence and a touch position sequence from the screen touch information, inputting the touch type sequence and the touch position sequence into a trained cheat detection model, performing feature extraction and feature fusion according to the touch type sequence and the touch position sequence respectively, analyzing the touch behavior pattern of the player to be detected, so that the cheat detection model can accurately output the cheat suspicion degree of the touch behavior pattern during the game determined according to the touch type sequence and the touch position sequence, and detecting whether the player is a cheat player by the cheat suspicion degree, which helps to improve the accuracy of cheat player detection.

[0005] The present application provides a method for detecting cheats in a game, the method comprising:

[0006] Acquiring screen touch information of the player to be detected during the game operation process; the screen touch information includes the touch type of each touch operation completed in the game operation process and at least one touch position for completing the touch operation;

[0007] Extracting the touch type in each touch operation, and constructing a touch type sequence according to the order in which the touch operations occur; and extracting at least one touch position in each touch operation, and constructing a touch position sequence according to the order in which the touch operations occur;

[0008] Inputting the touch type sequence and the touch position sequence into a pre-trained cheat detection model, so that the cheat detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, respectively, and outputs a cheat suspicion level of the touch behavior pattern of the player to be detected during the game operation;

[0009] If the cheat suspicion degree is greater than a preset threshold, the player to be detected is determined to be a cheating player.

[0010] In a possible implementation, the cheat suspicion level of the touch behavior pattern indicates whether the changes in the touch position and / or touch type of the player to be detected during the completion of the game operation conform to a regular change pattern;

[0011] The change in the touch position of the to-be-detected player during the game operation process conforms to a regular change pattern if the change in the touch position of the to-be-detected player during the game operation process satisfies at least one of the following conditions:

[0012] The distance between different touch positions of the player to be detected during the game operation is less than a preset distance threshold; among the multiple tracks formed by the touch position changes, the track direction and / or track length of each track are consistent with the other tracks; the tracks formed by the touch position changes form a regular polygon;

[0013] The change in the touch type of the to-be-detected player during the completion of the game operation conforms to a regular change pattern if the change in the touch type of the to-be-detected player during the completion of the game operation satisfies at least one of the following conditions:

[0014] During the game operation, the number of times the same touch type appears is greater than a preset threshold; the time interval between the appearances of the same touch type is less than a preset time threshold.

[0015] In a possible implementation, the cheat detection model includes a touch type processing sub-model, a touch position processing sub-model, and a cheat prediction sub-model;

[0016] Inputting the touch type sequence and the touch position sequence into a pre-trained cheat detection model, so that the cheat detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, and outputs a cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation, including:

[0017] Inputting the touch type sequence into the touch type processing sub-model, so that the touch type processing sub-model outputs type change information representing the touch type change;

[0018] Inputting the touch position sequence into the touch position processing sub-model, so that the touch position processing sub-model outputs position change information representing touch position changes;

[0019] The type change information and the position change information are input into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the cheat suspicion of the touch behavior pattern of the player to be detected during the process of completing the game operation.

[0020] In a possible implementation, the touch type processing sub-model includes an embedded feature representation layer, a temporal feature extraction layer, and a first feature fusion layer;

[0021] Inputting the touch type sequence into the touch type processing sub-model so that the touch type processing sub-model outputs type change information representing the touch type change includes:

[0022] Inputting the touch type sequence into the embedded feature representation layer, so that the embedded feature representation layer converts the touch type of each touch operation into a corresponding first type feature vector;

[0023] Inputting the obtained at least one first-type feature vector into the temporal feature extraction layer, so that the temporal feature extraction layer sequentially extracts features from the at least one first-type feature vector and outputs at least one second-type feature vector;

[0024] The at least one second-type feature vector is input into the first feature fusion layer, so that the first feature fusion layer outputs the type change information according to the at least one second-type feature vector.

[0025] In a possible implementation, the touch position processing sub-model includes a first position feature extraction layer and a second position feature extraction layer;

[0026] Inputting the touch position sequence into the touch position processing sub-model so that the touch position processing sub-model outputs position change information representing touch position changes includes:

[0027] Inputting the touch position sequence into the first position feature extraction layer, so that the first position feature extraction layer converts at least one touch position into a feature vector, performs weighted fusion, and outputs at least one first position representation vector;

[0028] The at least one first position representation vector is input into the second position feature extraction layer, so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient, and outputs the position change information.

[0029] In a possible implementation, the first position feature extraction layer includes a first temporal sub-extraction layer and a first feature sub-fusion layer;

[0030] Inputting the touch position sequence into the first position feature extraction layer so that the first position feature extraction layer converts the touch position sequence into a feature vector according to at least one touch position, performs weighted fusion, and outputs at least one first position representation vector, including:

[0031] Inputting the touch position sequence into the first sequential sub-extraction layer, so that the first sequential sub-extraction layer classifies at least one touch position according to different touch operations, performs feature extraction, and outputs at least one position feature vector;

[0032] According to the order in which each touch operation occurs during the game operation, at least one position feature vector is input into the first feature sub-fusion layer, so that the first feature sub-fusion layer performs weighted fusion on the at least one position feature vector according to the weight coefficients of different dimensions in the preset vector, and outputs at least one first position representation vector.

[0033] In a possible implementation, the second position feature extraction layer includes a second temporal sub-extraction layer and a second feature sub-fusion layer;

[0034] Inputting the at least one first position representation vector into the second position feature extraction layer so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient and outputs the position change information, includes:

[0035] Inputting at least one first position representation vector into the second sequential sub-extraction layer according to the order in which each touch operation occurs during the game operation, so that the second sequential sub-extraction layer classifies the at least one first position representation vector according to different touch operations, performs feature extraction, and outputs at least one touch type feature vector;

[0036] According to the order in which each touch operation occurs during the game operation, at least one touch type feature vector is input into the second feature sub-fusion layer, so that the second feature sub-fusion layer performs weighted fusion on the weight coefficients of different dimensions in at least one touch type feature vector and outputs at least one position change information.

[0037] In a possible implementation, the number of the first position feature extraction layers and the number of the second position feature extraction layers are respectively consistent with the number of position coordinates included in each touch position.

[0038] In one possible implementation, inputting the type change information and the position change information into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the degree of regularity of the touch position changes of the to-be-detected player during the completion of the game operation, includes:

[0039] The type change information and the position change information are input into the plug-in prediction sub-model, so that the plug-in prediction sub-model performs feature extraction after splicing the type change information and the position change information, determines the touch type change law and the touch position change law of the touch operation, and outputs the plug-in suspicion degree of the touch behavior pattern according to the type change law and the touch position change law of the touch operation.

[0040] In a possible embodiment, the detection method further includes:

[0041] Within a preset time period, if the number of times the player to be detected is determined to be a cheating player is greater than a preset threshold, a warning message for the player to be detected is sent to other players to remind other users that the player to be detected is in an abnormal state.

[0042] The present application also provides a cheat detection device for a game, the cheat detection device comprising:

[0043] An information acquisition module is used to acquire screen touch information of the player to be detected during the game operation process; the screen touch information includes the touch type of each touch operation completed during the game operation process and at least one touch position where the touch operation is completed;

[0044] A sequence construction module is used to extract the touch type in each touch operation, construct a touch type sequence according to the order in which the touch operations occur, and extract at least one touch position in each touch operation, and construct a touch position sequence according to the order in which the touch operations occur;

[0045] a cheating suspicion output module, configured to input the touch type sequence and the touch position sequence into a pre-trained cheating detection model, so that the cheating detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, and outputs a cheating suspicion level of the touch behavior pattern of the player to be detected during the game operation;

[0046] The cheat detection module is used to determine that the player to be detected is a cheat player if the cheat suspicion degree is greater than a preset threshold.

[0047] In a possible implementation, the cheat suspicion level of the touch behavior pattern indicates whether the changes in the touch position and / or touch type of the player to be detected during the completion of the game operation conform to a regular change pattern;

[0048] The change in the touch position of the to-be-detected player during the game operation process conforms to a regular change pattern if the change in the touch position of the to-be-detected player during the game operation process satisfies at least one of the following conditions:

[0049] The distance between different touch positions of the player to be detected during the game operation is less than a preset distance threshold; among the multiple tracks formed by the touch position changes, the track direction and / or track length of each track are consistent with the other tracks; the tracks formed by the touch position changes form a regular polygon;

[0050] The change in the touch type of the to-be-detected player during the completion of the game operation conforms to a regular change pattern if the change in the touch type of the to-be-detected player during the completion of the game operation satisfies at least one of the following conditions:

[0051] During the game operation, the number of times the same touch type appears is greater than a preset threshold; the time interval between the appearances of the same touch type is less than a preset time threshold.

[0052] In a possible implementation, the cheat detection model includes a touch type processing sub-model, a touch position processing sub-model, and a cheat prediction sub-model;

[0053] The cheating suspicion output module is configured to input the touch type sequence and the touch position sequence into a pre-trained cheating detection model, so that the cheating detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, and outputs the cheating suspicion of the touch behavior pattern of the player to be detected during the game operation. The cheating suspicion output module is configured to:

[0054] Inputting the touch type sequence into the touch type processing sub-model, so that the touch type processing sub-model outputs type change information representing the touch type change;

[0055] Inputting the touch position sequence into the touch position processing sub-model, so that the touch position processing sub-model outputs position change information representing touch position changes;

[0056] The type change information and the position change information are input into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the cheat suspicion of the touch behavior pattern of the player to be detected during the process of completing the game operation.

[0057] In a possible implementation, the touch type processing sub-model includes an embedded feature representation layer, a temporal feature extraction layer, and a first feature fusion layer;

[0058] When the cheating suspicion output module is used to input the touch type sequence into the touch type processing sub-model so that the touch type processing sub-model outputs type change information representing the touch type change, the regularity degree output module is used to:

[0059] Inputting the touch type sequence into the embedded feature representation layer, so that the embedded feature representation layer converts the touch type of each touch operation into a corresponding first type feature vector;

[0060] Inputting the obtained at least one first-type feature vector into the temporal feature extraction layer, so that the temporal feature extraction layer sequentially extracts features from the at least one first-type feature vector and outputs at least one second-type feature vector;

[0061] The at least one second-type feature vector is input into the first feature fusion layer, so that the first feature fusion layer outputs the type change information according to the at least one second-type feature vector.

[0062] In a possible implementation, the touch position processing sub-model includes a first position feature extraction layer and a second position feature extraction layer;

[0063] When the cheating suspicion output module is used to input the touch position sequence into the touch position processing sub-model so that the touch position processing sub-model outputs position change information representing touch position changes, the cheating suspicion output module is used to:

[0064] Inputting the touch position sequence into the first position feature extraction layer, so that the first position feature extraction layer converts at least one touch position into a feature vector, performs weighted fusion, and outputs at least one first position representation vector;

[0065] The at least one first position representation vector is input into the second position feature extraction layer, so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient, and outputs the position change information.

[0066] In a possible implementation, the first position feature extraction layer includes a first temporal sub-extraction layer and a first feature sub-fusion layer;

[0067] When the cheating suspicion output module is configured to input the touch position sequence into the first position feature extraction layer, so that the first position feature extraction layer converts the touch position sequence into a feature vector according to at least one touch position, performs weighted fusion, and outputs at least one first position representation vector, the cheating suspicion output module is configured to:

[0068] Inputting the touch position sequence into the first sequential sub-extraction layer, so that the first sequential sub-extraction layer classifies at least one touch position according to different touch operations, performs feature extraction, and outputs at least one position feature vector;

[0069] According to the order in which each touch operation occurs during the game operation, at least one position feature vector is input into the first feature sub-fusion layer, so that the first feature sub-fusion layer performs weighted fusion on the at least one position feature vector according to the weight coefficients of different dimensions in the preset vector, and outputs at least one first position representation vector.

[0070] In a possible implementation, the second position feature extraction layer includes a second temporal sub-extraction layer and a second feature sub-fusion layer;

[0071] When the cheating suspicion output module is used to input the at least one first position representation vector into the second position feature extraction layer so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient and outputs the position change information, the cheating suspicion output module is used to:

[0072] Inputting at least one first position representation vector into the second sequential sub-extraction layer according to the order in which each touch operation occurs during the game operation, so that the second sequential sub-extraction layer classifies the at least one first position representation vector according to different touch operations, performs feature extraction, and outputs at least one touch type feature vector;

[0073] According to the order in which each touch operation occurs during the game operation, at least one touch type feature vector is input into the second feature sub-fusion layer, so that the second feature sub-fusion layer performs weighted fusion on the weight coefficients of different dimensions in at least one touch type feature vector and outputs at least one position change information.

[0074] In a possible implementation, the number of the first position feature extraction layers and the number of the second position feature extraction layers are respectively consistent with the number of position coordinates included in each touch position.

[0075] In one possible implementation, when the cheating suspicion output module is configured to input the type change information and the position change information into the cheating prediction sub-model, so that the cheating prediction sub-model combines the type change information and the position change information to output the degree of regularity of the touch position changes of the to-be-detected player during the completion of the game operation, the cheating suspicion output module is configured to:

[0076] The type change information and the position change information are input into the plug-in prediction sub-model, so that the plug-in prediction sub-model performs feature extraction after splicing the type change information and the position change information, determines the touch type change law and the touch position change law of the touch operation, and outputs the plug-in suspicion degree of the touch behavior pattern according to the type change law and the touch position change law of the touch operation.

[0077] In a possible implementation, the cheat detection device further includes an early warning module, which is configured to:

[0078] Within a preset time period, if the number of times the player to be detected is determined to be a cheating player is greater than a preset threshold, a warning message for the player to be detected is sent to other players to remind other users that the player to be detected is in an abnormal state.

[0079] An embodiment of the present application also provides an electronic device, including: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the cheat detection method in the game as described above are performed.

[0080] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the cheat detection method in the game as described above are executed.

[0081] The embodiments of the present application provide a method, device, electronic device and storage medium for detecting cheats in games, which obtain screen touch information of a player to be detected during a game operation; the screen touch information includes the touch type of each touch operation that completes the game operation process and at least one touch position for completing the touch operation; according to the order in which each touch operation occurs during the game operation process, the touch type and at least one touch position are input into a pre-trained cheat detection model, so that the cheat detection model outputs the cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation process according to the touch type and at least one touch position; if the cheat suspicion degree is greater than a preset threshold, the player to be detected is determined to be a cheating player. In this way, the touch type and touch position in the acquired screen touch information are used to input the constructed touch type sequence and touch position sequence into the trained plug-in detection model. By performing feature fusion on the constructed touch type sequence and touch position sequence, the plug-in suspicion of the touch behavior pattern of the player to be detected during the game operation is directly and accurately output to detect whether the player to be detected is a plug-in player, which helps to improve the accuracy of plug-in player detection.

[0082] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. 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 relevant drawings can be obtained based on these drawings without creative work.

[0084] Figure 1 A flowchart of a method for detecting cheats in a game provided by an embodiment of the present application;

[0085] Figure 2 A flowchart of another method for detecting cheats in games provided in an embodiment of the present application;

[0086] Figure 3 A schematic diagram of the structure of the cheat detection model provided in the embodiment of the present application;

[0087] Figure 4A flowchart of another method for detecting cheats in games provided in an embodiment of the present application;

[0088] Figure 5 A schematic diagram of the flow of data processing and cheat player prediction provided in an embodiment of the present application;

[0089] Figure 6 This is one of the structural diagrams of a cheat detection device in a game provided by an embodiment of the present application;

[0090] Figure 7 This is a second structural diagram of a cheat detection device in a game provided by an embodiment of the present application;

[0091] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0092] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions 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. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different 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 for which protection is claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work falls within the scope of protection of the present application.

[0093] Research has found that when players play mobile games, they usually control their virtual characters by touching the phone screen. Therefore, screen touch data is data that is easily collected on mobile devices such as mobile phones and tablets, and represents the player's finger operations on the device screen to control the virtual character when playing mobile games on mobile devices such as mobile phones. At present, the detection of game plug-ins is generally carried out through physical methods such as device information analysis, game environment detection, and abnormal game process detection for click-type plug-ins. Even the processing of screen touch data requires the detection personnel to identify the plug-in through the characteristics of the screen touch data, and the processing characteristics need to be compared one by one. Both of the above methods require the processing of large amounts of data. In addition, due to changes in physical conditions themselves and updates to the characteristics of screen touch data, the above comparison method will also have problems with false detection and missed detection, resulting in low accuracy in detecting plug-in players.

[0094] Based on this, an embodiment of the present application provides a method for detecting cheats in games. By obtaining the touch type and touch position in the screen touch information, the constructed touch type sequence and touch position sequence are input into the trained cheat detection model. By performing feature fusion on the constructed touch type sequence and touch position sequence, the cheat suspicion of the touch behavior pattern of the player to be detected during the game operation is directly and accurately output, which is used to detect whether the player to be detected is a cheat player, which helps to improve the accuracy of cheat player detection.

[0095] See also Figure 1 , Figure 1 This is a flow chart of a method for detecting cheats in a game provided by an embodiment of the present application. Figure 1 As shown in , the method for detecting cheats in games provided by the embodiments of the present application includes:

[0096] S101. Obtain screen touch information of a player to be detected during a game operation; the screen touch information includes a touch type of each touch operation completed during the game operation and at least one touch position for completing the touch operation.

[0097] S102: Extract the touch type from each touch operation and construct a touch type sequence based on the order in which each touch operation occurs. Also, extract at least one touch position from each touch operation and construct a touch position sequence based on the order in which each touch operation occurs. S103: Input the touch type sequence and the touch position sequence into a pre-trained cheat detection model, so that the cheat detection model extracts and fuses features based on the touch type sequence and the touch position sequence, respectively, and outputs a cheat suspicion level for the touch behavior pattern of the player to be detected during the game operation.

[0098] S104: If the cheating suspicion level is greater than a preset threshold, determine that the player to be detected is a cheating player.

[0099] An embodiment of the present application provides a method for detecting cheats in a game. By obtaining the touch type and touch position in the screen touch information, the constructed touch type sequence and touch position sequence are input into a trained cheat detection model. By performing feature fusion on the constructed touch type sequence and touch position sequence, the cheat suspicion of the touch behavior pattern of the player to be detected during the game operation is directly and accurately outputted, so as to detect whether the player to be detected is a cheating player, thereby helping to improve the accuracy of cheating player detection.

[0100] Screen touch information is a typical spatiotemporal sequence data. Screen touch information may include multiple touch operations. For each touch operation, the information describing the touch operation may be the touch type and touch position. Therefore, a complete screen touch information may include the touch types of multiple touch operations, at least one touch position for each touch operation, and the time when each touch operation is obtained.

[0101] Specifically, the specific type of touch operation and the touch behavior mode under the touch type can be known through the touch type; the touch type includes one or more of a click operation, a long press operation, and a slide operation.

[0102] It is worth noting that when generating screen touch information, in order to characterize the touch type through data, each touch type needs to be converted into a corresponding type identifier.

[0103] For example, "0" can be used to represent a click operation, "1" can be used to represent a long press operation, "2" can be used to represent a sliding operation, and so on.

[0104] At least one touch position can represent at least one position of a touch operation, and the operation trajectory of the touch operation can be obtained through the at least one touch position. In the embodiment of the present application, multiple touch positions in a touch operation can be represented by two-dimensional coordinate data.

[0105] For example, the representation format of the screen touch information is [type identifier, [[X coordinate 1, Y coordinate 1, timestamp 1]…, [X coordinate n, Y coordinate n, timestamp n]]].

[0106] In step S101, screen touch information of the player to be detected during the game operation is obtained.

[0107] It is worth noting that the screen touch information obtained in the embodiment of the present application is the touch information generated by the player by touching the screen during the game operation process. For the player to be detected, the game operation process can be all the screen touch information within a detection cycle, that is, the touch information of the user operating the screen within a period of detection time; it can also be all the screen touch information within the time period of completing a specific game behavior.

[0108] When acquiring screen touch information, for different touch operations, there may be certain deviations in the data presentation and length at both the touch type and touch position levels, and even low-quality data that is meaningless to the detection process may be acquired. Therefore, after acquiring the screen operation information, it is necessary to perform several steps such as information filtering, data normalization, and formatting to pre-process the data so that the data obtained is standardized, thereby better detecting whether the player is a cheating player.

[0109] Since data is easily interfered by external factors during collection, transmission, and storage, the quality of low-quality screen touch information varies, and low-quality screen touch information will introduce unnecessary noise to model training and model inference, so it needs to be filtered in advance.

[0110] Step 1 may be to first filter low-quality screen touch information. After obtaining the screen touch information of the player to be detected, it is necessary to preferentially filter out low-quality screen touch information from the screen touch information.

[0111] In a possible implementation, low-quality screen touch information includes information in which the length of the screen touch information is less than a preset length threshold, and information in which the screen touch information lacks necessary features for representing operation information.

[0112] Specifically, (1) the screen touch information is empty or too short; (2) there are empty values or abnormal values such as "Nan" and "Infinity" in the screen touch information; (3) the structure of the screen touch information is incomplete and necessary fields are missing; (4) the screen touch information is the information in the idle state (that is, it does not move and the coordinates do not change throughout the process).

[0113] The necessary characteristic information in the operation information includes, but is not limited to, touch type, touch position, and time information.

[0114] Because screen touch information is collected from a wide variety of devices and has different settings, resulting in different resolutions, the coordinate values representing the touch location in the screen touch information often have different dimensions, making it difficult to process them uniformly. Therefore, the coordinate values representing the touch location can be normalized.

[0115] Step 2: Normalize the coordinate data representing the touch position.

[0116] Specifically, the coordinate data representing the touch position is two-dimensional coordinate data. For the coordinate data of each touch position, the following processing is performed on the coordinate values on each coordinate axis: (1) the maximum coordinate value and the minimum coordinate value are calculated; (2) the original coordinate value is linearly converted to the range of [0,1].

[0117] For example, taking the X-axis coordinate processing as an example, the coordinate data can be normalized using the following formula:

[0118]

[0119] Among them, X1 is the X-axis coordinate after normalization, X is the X-axis coordinate before normalization, and X max is the maximum coordinate value in the X-axis coordinate, X min The minimum coordinate value in the X-axis coordinate.

[0120] In step S102, the touch type in each touch operation is extracted, and a touch type sequence is constructed according to the order in which each touch operation occurs. At least one touch position in each touch operation is extracted, and a touch position sequence is constructed according to the order in which each touch operation occurs.

[0121] In a preferred embodiment, in order to characterize the touch type and at least one touch position as timing information that changes over time, the touch type can be extracted from the screen touch information to form a touch type sequence, and at least one touch position can be extracted to form a touch position sequence. The touch type sequence and touch position sequence are input into the plug-in detection model.

[0122] Specifically, the touch type is extracted from the original screen touch information as an independent sequence, which is recorded as the touch type sequence A = {a1, a2, ..., a |A|}, where a∈A represents the touch type identifier (ID); for the touch position (represented by two-dimensional coordinates), its two-layer hierarchical sequence structure is retained and the timestamp field is removed, that is, only the normalized X coordinate value and Y coordinate value are used, which is recorded as the coordinate sequence {B (1) , B (2) ,...,B (|B|)}, where B (i) ∈B represents the coordinate sequence within a touch operation, recorded as where b (i) ∈B (i) Indicates the coordinates of touch operation i, that is, b (i) =(x (i) ,y (i) ).

[0123] In step S103, the touch type sequence and the touch position sequence are input into a pre-trained cheat detection model, so that the cheat detection model performs feature extraction and feature fusion according to the touch type sequence and the touch position sequence respectively, and then outputs the cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation.

[0124] After obtaining the screen touch information and constructing the touch type sequence and touch position sequence, the touch type sequence and touch position sequence are input into a pre-trained cheat detection model, and the cheat detection model outputs the cheat suspicion of the touch behavior pattern during the game operation.

[0125] In a possible implementation, it is also possible not to construct a touch type sequence and a touch position sequence, and directly obtain the touch type and at least one touch position for completing the touch operation from the screen touch information, and input the touch type and at least one touch position in the screen touch information into the plug-in detection model in the order in which each touch operation occurs during the game operation to perform the process of plug-in suspicion detection.

[0126] Analysis shows significant differences in screen touch information between normal players and cheaters during gameplay. While normal players' touch information is typically random and lacks a fixed pattern, cheaters' touch information often exhibits distinct patterns, such as repeated points, repeated lines, parallel lines, clusters, and regular polygons. Therefore, a trained cheat detection model is required. A trained cheat detection model can learn the behavioral differences between normal and cheaters, and then characterize whether a player is cheating by determining the suspicion of cheating in the touch behavior patterns during gameplay.

[0127] Specifically, the cheat suspicion degree of the touch behavior pattern represents whether the changes in the touch position and / or touch type of the player to be detected conform to a regular change pattern during the process of completing the game operation; the more the touch position conforms to the regular change pattern, the higher the corresponding cheat suspicion degree.

[0128] The change in the touch position of the player to be detected during the game operation process conforms to a regular change pattern if the change in the touch position of the player to be detected during the game operation process satisfies at least one of the following conditions.

[0129] The distance between different touch positions of the player to be detected during the game operation is less than a preset distance threshold (repeated points); among the multiple tracks formed by the touch position changes, the track direction and / or track length of each track are consistent with those of the other tracks (repeated lines, parallel lines); the tracks formed by the touch position changes form a regular polygon (the track is a regular polygon).

[0130] The change in the touch type of the to-be-detected player during the completion of the game operation conforms to a regular change pattern if the change in the touch type of the to-be-detected player during the completion of the game operation satisfies at least one of the following conditions:

[0131] During the game operation, the number of times the same touch type appears is greater than a preset threshold; the time interval between the appearances of the same touch type is less than a preset time threshold.

[0132] In one embodiment, the pre-trained cheat detection model can be divided into three parts, namely, a touch type processing sub-model, a touch position processing sub-model and a cheat prediction sub-model. The touch type processing sub-model is used to process the touch type input into the model to obtain type change information; the touch position processing sub-model is used to process at least one touch position input into the model to obtain position change information; the cheat prediction sub-model is used to combine the type change information and the position change information to determine the cheat suspicion of the touch behavior pattern of the player to be detected during the game operation.

[0133] Specifically, see Figure 2 , Figure 2 This is a flow chart of another method for detecting cheats in games provided by an embodiment of the present application. Figure 2 As shown in FIG, step S103 includes:

[0134] S1031: Input the touch type sequence into the touch type processing sub-model, so that the touch type processing sub-model outputs type change information representing the touch type change.

[0135] S1032: Input the touch position sequence into the touch position processing sub-model, so that the touch position processing sub-model outputs position change information representing touch position changes.

[0136] S1033. Input the type change information and the position change information into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the cheat suspicion of the touch behavior pattern of the player to be detected during the game operation.

[0137] In step S1031, the touch type processing sub-model includes an embedded feature representation layer, a temporal feature extraction layer, and a first feature fusion layer; specifically, step S1031 includes:

[0138] The touch type sequence is input into the embedded feature representation layer, so that the embedded feature representation layer converts the touch type of each touch operation into a corresponding first type feature vector.

[0139] In a possible implementation, the embedded feature representation layer may be an Embedding layer, in which each touch type (the input model is a touch type identification ID) is converted into a corresponding embedding vector (a first type feature vector).

[0140] A specific implementation method may be: taking the touch type sequence as an input to the cheat detection model as an example, the first step of using the embedding layer is to encode each touch type in the touch type sequence through the index, and then create an embedding matrix. It is necessary to decide how many "latent factors" need to be assigned to each index according to the needs, which generally means the length of the first type of feature vector that needs to be output (which can be set according to the model output requirements during the model training process), and then embed the encoded touch type code into the created embedding matrix to obtain the first type of feature vector.

[0141] The obtained at least one first type feature vector is input into the temporal feature extraction layer, so that the temporal feature extraction layer extracts features from the at least one first type feature vector in sequence and outputs at least one second type feature vector.

[0142] In one possible implementation, the time series feature extraction layer uses a long short-term memory network layer (Long Short-Term Memory, LSTM), which is essentially a time recurrent neural network (Recurrent Neural Network, RNN) and is also one of the basic components (network layers) in the field of deep learning. The use of LSTM can effectively solve the problems of gradient disappearance and gradient explosion in the long sequence training process.

[0143] In the timing feature extraction layer, feature extraction is performed in sequence through the timing feature extraction layer (including multiple timing feature extraction units, and the number of timing feature extraction units is consistent with the number of touch types) according to the order in which each touch operation occurs. It is worth noting that during the processing, the output of the previous timing feature extraction unit is the output of the next timing feature extraction unit.

[0144] In a possible implementation, the hidden state H of each touch type can be extracted through the temporal feature extraction layer. A , (second type eigenvector).

[0145] The at least one second-type feature vector is input into the first feature fusion layer, so that the first feature fusion layer outputs the type change information according to the at least one second-type feature vector.

[0146] In one possible implementation, the first feature fusion layer is an Attention mechanism layer, and most preferably a temporal attention mechanism. The principle of Attention is to calculate the degree of matching between the current input sequence and the output vector. A higher matching degree, that is, a higher relative score of the attention focus point, is obtained.

[0147] At least one second type feature vector is input into the first feature fusion layer, so that the first feature fusion layer outputs the type change information according to the at least one second type feature vector. In the embodiment of the present application, the type change information is a vector V A , V A The expression is as follows:

[0148] H A =LSTM(Embedding(A));

[0149] V A =Attention(H A );

[0150] The above expression expresses that the hidden state H A The input touch type sequence is first processed by the Embedding layer and then by the LSTM layer; the type change information (V A ) is composed of the hidden state H A Input to the Attention layer for feature processing.

[0151] In step S1032, in this embodiment of the present application, the number of first position feature extraction layers and the number of second position feature extraction layers included in the touch position processing submodel are consistent with the number of position coordinates included in each touch position. Since the touch position of each touch type is represented by a two-dimensional coordinate vector, the touch position processing submodel includes one first position feature extraction layer and one second position feature extraction layer. Position fusion vectors are extracted through the two position feature extraction layers, and then position change information is obtained. Specifically, step S1032 includes:

[0152] Step S10321: input the touch position sequence into the first position feature extraction layer, so that the first position feature extraction layer converts at least one touch position into a feature vector, performs weighted fusion, and outputs at least one first position representation vector.

[0153] In a possible implementation, the first position feature extraction layer includes a first temporal sub-extraction layer and a first feature sub-fusion layer, and step S10321 includes:

[0154] The touch position sequence is input into the first sequential sub-extraction layer, so that the first sequential sub-extraction layer classifies at least one touch position according to different touch operations, performs feature extraction, and outputs at least one position feature vector.

[0155] In a possible implementation, the first temporal sub-extraction layer is also an LSTM layer. In the first temporal sub-extraction layer, at least one touch position is classified according to the touch operation and then feature extraction is performed to output at least one position feature vector.

[0156] According to the order in which each touch operation occurs during the game operation, at least one position feature vector is input into the first feature sub-fusion layer, so that the first feature sub-fusion layer performs weighted fusion on the at least one position feature vector according to the weight coefficients of different dimensions in the preset vector, and outputs at least one first position representation vector.

[0157] In one possible implementation, the first feature sub-fusion layer is also an Attention layer. During the model training process, it is necessary to determine the weight coefficients corresponding to different dimensions in the position feature vector by learning the behavior patterns represented in different samples, and then perform weighted processing within the Attention layer to obtain the first position representation vector.

[0158] Specifically, taking the touch position sequence input into the cheat detection model as an example, traverse B (i) ∈B, the touch position sequence B within the same touch operation (i) Input to the first LSTM layer to get the hidden state H of each time step B (i) , and obtain the intermediate feature vector V through the Attention layer B (i) ; Finally, we get the representation vector sequence {V B (1) , V B (2) ,…V B (i) ,…V B (│B│)}(the first position represents a vector).

[0159] The Attention layer used adopts a self-attention mechanism (Self-Attention). For the hidden state of each time step, the attention calculation is as follows:

[0160] u t =Tanh(W u H t +b u );

[0161]

[0162]

[0163] Among them, μt is the high-dimensional mapping of the hidden state at time step t; H t is hidden state; W μ is used to calculate the parameter μ t The learnable weight parameters of b μ is used to calculate the parameter μ t The learnable bias parameter W α is used to calculate the parameter α t The learnable weight parameter α t is the attention weight at time step t;

[0164] Step S10322: input the at least one first position representation vector into the second position feature extraction layer, so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient, and outputs the position change information.

[0165] In one possible implementation, the first position feature extraction layer includes a second temporal sub-extraction layer and a second feature sub-fusion layer; step S10322 includes:

[0166] According to the order in which each touch operation occurs during the game operation, at least one first position representation vector is input into the second timing sub-extraction layer, so that the second timing sub-extraction layer classifies the at least one first position representation vector according to different touch operations, performs feature extraction, and outputs at least one touch type feature vector.

[0167] In a possible implementation, the second temporal sub-extraction layer is also an LSTM layer. In the second temporal sub-extraction layer, after classifying at least one first position representation vector according to different touch operations, feature extraction is performed to output at least one touch type feature vector.

[0168] In the second timing sub-extraction layer, feature extraction is performed in sequence through the second timing sub-extraction layer (including multiple timing feature extraction units, and the number of timing feature extraction units is consistent with the number of touch types) according to the order of occurrence of each touch operation. It is worth noting that during the processing, the output of the previous timing feature extraction unit is the output of the next timing feature extraction unit.

[0169] According to the order in which each touch operation occurs during the game operation, at least one touch type feature vector is input into the second feature sub-fusion layer, so that the second feature sub-fusion layer performs weighted fusion on the weight coefficients of different dimensions in at least one touch type feature vector and outputs at least one position change information.

[0170] In one possible implementation, the second feature sub-fusion layer is also an Attention layer. During the model training process, it is necessary to determine the weight coefficients corresponding to different dimensions in the touch type feature vector by learning the behavior patterns represented in different samples, and then perform weighted processing within the Attention layer to obtain position change information.

[0171] Output the position change information. In the embodiment of the present application, the position change information is a vector V B , V B The expression is as follows:

[0172]

[0173]

[0174]

[0175] V B =Attention 2nd (H B );

[0176] The above expression expresses that the hidden state H B (i) The intermediate feature vector V is obtained by processing the input touch position sequence through the first temporal sub-extraction layer (LSTM); B (i) Is H B (i) Input to the first feature sub-fusion layer (Attention) for feature processing; H B is represented by the first position vector ({V B (1) , V B (2) ,…V B (i) ,…V B (│B│)}) Input the second temporal sub-extraction layer (LSTM) to obtain the position change information (V B ) is H B Input to the second feature sub-fusion layer (Attention) for feature processing.

[0177] In step S1033, the cheat prediction sub-model includes multiple fully connected layers, and the type change information and the position change information are input into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the cheat suspicion of the touch behavior pattern of the player to be detected in the process of completing the game operation.

[0178] In a possible implementation manner, the type change information and the position change information are both vectors. After the type change information and the position change information are concatenated, a vector V is obtained. The calculation formula of V is: V=Concat(V A , V B ); After the vector V is calculated through the classification function, the regularity degree Y is obtained. The calculation formula of Y is: Y = Sigmoid(FCs(V)). The regularity degree is used to represent the probability that the player to be detected is a cheating player.

[0179] Among them, Sigmoid is one of the basic components (activation functions) in the field of deep learning, used to map the input to a real number between [0, 1], representing the probability of meeting the target condition; FC is the abbreviation of "Fully Connected Layer", that is, the fully connected layer is one of the basic components (network layers) in the field of deep learning, used to map the feature vector obtained from the feature extraction layer before this layer to the next layer, or to the final classifier; Concat is the splicing function.

[0180] In one possible implementation, when training the cheat detection model, a supervised training method is adopted, and the collected raw screen touch data is appropriately preprocessed (consistent with the prediction process, the preprocessing includes low-quality sample filtering, normalization and data formatting) to obtain a standardized data set. On this basis, based on expert experience and prior knowledge, the samples in the data set are labeled for multiple rounds, and finally binary classification labeled data with high confidence is obtained, in which the positive samples are cheat players and the negative samples are normal players. At this point, the labeled data set is divided into training set, validation set and test set, and the cross entropy loss is used to guide model learning through the general training paradigm of binary supervised learning. Finally, after multiple rounds of parameter adjustment and training, a cheat detection model that meets expectations can be obtained.

[0181] During the training of the cheat model, the structure of the cheat detection model has been determined. Figure 3 , Figure 3 This is a schematic diagram of the structure of the cheat detection model provided in the embodiment of the present application, as shown in FIG. Figure 3 As shown in FIG, the cheat detection model includes a touch type processing sub-model 310, a touch position processing sub-model 320 and a cheat prediction sub-model 330. The input to the touch type processing sub-model is the acquired multiple touch types (such as Figure 3As shown in , type 1, type 2 and type 3), the touch type processing sub-model 310 includes an embedded feature representation layer 311 (including multiple embedded feature representation units 3111), a temporal feature extraction layer 312 (including multiple temporal feature extraction units 3121) and a first feature fusion layer 313; the touch position processing sub-model 320 includes a first position feature extraction layer 321 and a second position feature extraction layer 322 (the input to the first position feature extraction layer 321 is the acquired multiple coordinates representing the touch position (such as Figure 3 As shown in , coordinates 1, 2, and 3); the input to the second position feature extraction layer 322 is a plurality of events output by the first position feature extraction layer 321 (such as Figure 3 As shown in , event 1, event 2 and event 3); the first position feature extraction layer 321 includes a first timing sub-extraction layer 3211 (including multiple first timing sub-extraction units 32111) and a first feature sub-fusion layer 3212; the second position feature extraction layer 322 includes a second timing sub-extraction layer 3221 (including multiple second timing sub-extraction units 32211) and a second feature sub-fusion layer 3222. The screen touch information is processed by each layer in each sub-model to obtain the degree of regularity that characterizes whether the player to be detected is a plug-in player.

[0182] In step S104, if the cheating suspicion degree is greater than a preset threshold, it is determined that the player to be detected is a cheating player. In the embodiment of the present application, the cheating suspicion degree of the touch behavior pattern is used to characterize the probability that the player to be detected is a cheating player. The higher the cheating suspicion degree of the touch behavior pattern, the higher the probability that the player to be detected is a cheating player. Therefore, in a possible implementation manner, the cheating suspicion degree of the touch behavior pattern can be a probability value.

[0183] Preferably, in an embodiment of the present application, the preset threshold can be determined based on historical operation data combined with expert prior knowledge. The preset thresholds corresponding to different games or different operation actions in the same game can be different. That is, when selecting the preset threshold, it can be determined in combination with the game operation process of obtaining screen touch information.

[0184] For example, in the two operations of skydiving and shooting, the screen touch information generated by the player will be different (the touch type and touch position changes will also be different). Therefore, different preset thresholds need to be set for different game operation processes to prevent the situation where the screen touch information of normal players in certain specific game actions is very regular, but is mistakenly judged as a cheating player because of its excessive regularity, thereby further improving the accuracy of detecting cheating players.

[0185] In one possible implementation, see Figure 4 , Figure 4 This is a flow chart of another method for detecting cheats in games provided by an embodiment of the present application. Figure 4 As shown in , the method for detecting cheats in games provided by the embodiments of the present application includes:

[0186] S401. Acquire screen touch information of a player to be detected during a game operation; the screen touch information includes a touch type of each touch operation completed during the game operation and at least one touch position for completing the touch operation.

[0187] S402: Extracting the touch type in each touch operation, constructing a touch type sequence according to the order in which the touch operations occur, and extracting at least one touch position in each touch operation, constructing a touch position sequence according to the order in which the touch operations occur;

[0188] S403. Input the touch type sequence and the touch position sequence into a pre-trained cheat detection model, so that the cheat detection model performs feature extraction and feature fusion according to the touch type sequence and the touch position sequence respectively, and outputs the cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation.

[0189] S404: If the cheating suspicion level is greater than a preset threshold, determine that the player to be detected is a cheating player.

[0190] Among them, the description of S401 to S404 can refer to the description of S101 to S104, and can achieve the same technical effect, so it will not be repeated here.

[0191] S405. Within a preset time period, if the number of times the player to be detected is determined to be a cheating player is greater than a preset threshold, a warning message for the player to be detected is sent to other players to remind other users that the player to be detected is in an abnormal state.

[0192] In one possible implementation, the preset time period can be a detection cycle, or a natural cycle (one day, one week), etc. The preset time period can be set according to the plug-in detection requirements and is not specifically limited in the embodiments of the present application.

[0193] In an embodiment of the present application, when the player to be detected is determined to be an abnormal player within a preset time period, in addition to generating a warning message and sending it to other players to remind them of the abnormal status of the player to be detected, the player to be detected will be directly banned, and the cheating player will be prohibited from operating again, which will affect the fairness of the game.

[0194] See also Figure 5 , Figure 5A flow chart of data processing and cheat player prediction provided in the embodiment of the present application is shown as follows: Figure 5 As shown, the technical solutions provided in the embodiments of the present application include:

[0195] Step 1: Data collection, collecting screen operation information during game operation;

[0196] Step 2: Data preprocessing: Data preprocessing is performed on the collected data. Data preprocessing includes filtering low-quality data, normalizing data, and formatting data. After step 2, if the model training process is being executed, jump to step 3; if the inference process is performed using the trained model, jump to step 5.

[0197] Step 3: Model construction, setting up sub-model structures for processing touch type and touch position in the model respectively, so that the trained model can separately process the touch type and touch position in the screen touch information.

[0198] Step 4: Model training. After preprocessing the obtained sample information, multiple rounds of labeling are performed to obtain positive samples (cheating players) and negative samples (normal players); the labeled data set is divided into training set, validation set and test set, and the cross-entropy loss is used to guide the model training process through the general training paradigm of binary supervised learning.

[0199] Step 5: Model inference. The trained cheat detection model outputs the cheat suspicion of the player's touch behavior pattern during the game operation. The detection output is greater than the threshold θ. If it is greater than the threshold θ, the player is determined to be a cheating player. If it is not greater than the threshold θ, the player is determined to be a normal player.

[0200] An embodiment of the present application provides a method for detecting cheats in a game, which obtains screen touch information of a player to be detected during a game operation; the screen touch information includes the touch type of each touch operation that completes the game operation process and at least one touch position for completing the touch operation; a touch type sequence and a touch position sequence are constructed according to the order in which each touch operation occurs during the game operation process; the touch type sequence and the touch position sequence are input into a pre-trained cheat detection model, so that the cheat detection model performs feature extraction and feature fusion according to the touch type sequence and the touch position sequence, respectively, and outputs a cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation process; if the cheat suspicion degree is greater than a preset threshold, the player to be detected is determined to be a cheating player. In this way, the touch type and touch position in the acquired screen touch information are used to input the constructed touch type sequence and touch position sequence into the trained plug-in detection model. By performing feature fusion on the constructed touch type sequence and touch position sequence, the plug-in suspicion of the touch behavior pattern of the player to be detected during the game operation is directly and accurately output to detect whether the player to be detected is a plug-in player, which helps to improve the accuracy of plug-in player detection.

[0201] See also Figure 6 、 Figure 7 , Figure 6 This is one of the structural diagrams of a cheat detection device for a game provided by an embodiment of the present application. Figure 7 This is a second structural diagram of a cheat detection device in a game provided by an embodiment of the present application. Figure 6 As shown in , the cheat detection device 600 includes:

[0202] The information acquisition module 610 is used to obtain screen touch information of the player to be detected during the game operation process; the screen touch information includes the touch type of each touch operation completed during the game operation process and at least one touch position where the touch operation is completed;

[0203] A sequence construction module 620 is configured to extract the touch type in each touch operation, construct a touch type sequence according to the order in which the touch operations occur, and extract at least one touch position in each touch operation, and construct a touch position sequence according to the order in which the touch operations occur;

[0204] The cheating suspicion output module 630 is configured to input the touch type sequence and the touch position sequence into a pre-trained cheating detection model, so that the cheating detection model extracts and fuses features based on the touch type sequence and the touch position sequence, and outputs a cheating suspicion level of the touch behavior pattern of the player to be detected during the game operation.

[0205] The cheat detection module 640 is used to determine that the player to be detected is a cheating player if the cheat suspicion degree is greater than a preset threshold.

[0206] Further, such as Figure 7 As shown, the cheat detection device 600 further includes an early warning module 650, which is used to:

[0207] Within a preset time period, if the number of times the player to be detected is determined to be a cheating player is greater than a preset threshold, a warning message for the player to be detected is sent to other players to remind other users that the player to be detected is in an abnormal state.

[0208] Furthermore, the cheat suspicion degree of the touch behavior pattern indicates whether the changes in the touch position and / or touch type of the player to be detected during the completion of the game operation conform to a regular change pattern;

[0209] The change in the touch position of the to-be-detected player during the game operation process conforms to a regular change pattern if the change in the touch position of the to-be-detected player during the game operation process satisfies at least one of the following conditions:

[0210] The distance between different touch positions of the player to be detected during the game operation is less than a preset distance threshold; among the multiple tracks formed by the touch position changes, the track direction and / or track length of each track are consistent with the other tracks; the tracks formed by the touch position changes form a regular polygon;

[0211] The change in the touch type of the to-be-detected player during the completion of the game operation conforms to a regular change pattern if the change in the touch type of the to-be-detected player during the completion of the game operation satisfies at least one of the following conditions:

[0212] During the game operation, the number of times the same touch type appears is greater than a preset threshold; the time interval between the appearances of the same touch type is less than a preset time threshold.

[0213] Furthermore, the cheat detection model includes a touch type processing sub-model, a touch position processing sub-model and a cheat prediction sub-model;

[0214] The cheating suspicion output module 630 is configured to input the touch type sequence and the touch position sequence into a pre-trained cheating detection model, so that the cheating detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, and outputs the cheating suspicion of the touch behavior pattern of the player to be detected during the game operation. The cheating suspicion output module 630 is configured to:

[0215] Inputting the touch type sequence into the touch type processing sub-model, so that the touch type processing sub-model outputs type change information representing the touch type change;

[0216] Inputting the touch position sequence into the touch position processing sub-model, so that the touch position processing sub-model outputs position change information representing touch position changes;

[0217] The type change information and the position change information are input into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the cheat suspicion of the touch behavior pattern of the player to be detected during the process of completing the game operation.

[0218] Furthermore, the touch type processing sub-model includes an embedded feature representation layer, a temporal feature extraction layer, and a first feature fusion layer;

[0219] When the cheating suspicion output module 630 is used to input the touch type sequence into the touch type processing sub-model so that the touch type processing sub-model outputs type change information representing the touch type change, the cheating suspicion output module 630 is used to:

[0220] Inputting the touch type sequence into the embedded feature representation layer, so that the embedded feature representation layer converts the touch type of each touch operation into a corresponding first type feature vector;

[0221] Inputting the obtained at least one first-type feature vector into the temporal feature extraction layer, so that the temporal feature extraction layer sequentially extracts features from the at least one first-type feature vector and outputs at least one second-type feature vector;

[0222] The at least one second-type feature vector is input into the first feature fusion layer, so that the first feature fusion layer outputs the type change information according to the at least one second-type feature vector.

[0223] Furthermore, the touch position processing sub-model includes a first position feature extraction layer and a second position feature extraction layer;

[0224] When the cheating suspicion output module 630 is used to input the touch position sequence into the touch position processing sub-model so that the touch position processing sub-model outputs position change information representing the touch position change, the cheating suspicion output module 630 is used to:

[0225] Inputting the touch position sequence into the first position feature extraction layer, so that the first position feature extraction layer converts at least one touch position into a feature vector, performs weighted fusion, and outputs at least one first position representation vector;

[0226] The at least one first position representation vector is input into the second position feature extraction layer, so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient, and outputs the position change information.

[0227] Furthermore, the first position feature extraction layer includes a first temporal sub-extraction layer and a first feature sub-fusion layer;

[0228] When the cheating suspicion output module 630 is configured to input the touch position sequence into the first position feature extraction layer, so that the first position feature extraction layer converts the touch position sequence into a feature vector according to at least one touch position, performs weighted fusion, and outputs at least one first position representation vector, the cheating suspicion output module 630 is configured to:

[0229] Inputting the touch position sequence into the first sequential sub-extraction layer, so that the first sequential sub-extraction layer classifies at least one touch position according to different touch operations, performs feature extraction, and outputs at least one position feature vector;

[0230] According to the order in which each touch operation occurs during the game operation, at least one position feature vector is input into the first feature sub-fusion layer, so that the first feature sub-fusion layer performs weighted fusion on the at least one position feature vector according to the weight coefficients of different dimensions in the preset vector, and outputs at least one first position representation vector.

[0231] Furthermore, the second position feature extraction layer includes a second temporal sub-extraction layer and a second feature sub-fusion layer;

[0232] When the cheating suspicion output module 630 is used to input the at least one first position representation vector into the second position feature extraction layer so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient and outputs the position change information, the cheating suspicion output module 630 is used to:

[0233] Inputting at least one first position representation vector into the second sequential sub-extraction layer according to the order in which each touch operation occurs during the game operation, so that the second sequential sub-extraction layer classifies the at least one first position representation vector according to different touch operations, performs feature extraction, and outputs at least one touch type feature vector;

[0234] According to the order in which each touch operation occurs during the game operation, at least one touch type feature vector is input into the second feature sub-fusion layer, so that the second feature sub-fusion layer performs weighted fusion on the weight coefficients of different dimensions in at least one touch type feature vector and outputs at least one position change information.

[0235] Furthermore, the number of the first position feature extraction layers and the number of the second position feature extraction layers are respectively consistent with the number of position coordinates included in each touch position.

[0236] Furthermore, when the cheating suspicion output module 630 is used to input the type change information and the position change information into the cheating prediction sub-model, so that the cheating prediction sub-model combines the type change information and the position change information to output the degree of regularity of the touch position changes of the to-be-detected player during the completion of the game operation, the cheating suspicion output module 630 is used to:

[0237] The type change information and the position change information are input into the plug-in prediction sub-model, so that the plug-in prediction sub-model performs feature extraction after splicing the type change information and the position change information, determines the touch type change law and the touch position change law of the touch operation, and outputs the plug-in suspicion degree of the touch behavior pattern according to the type change law and the touch position change law of the touch operation.

[0238] The embodiment of the present application provides a cheat detection device for a game, which obtains screen touch information of a player to be detected during a game operation; the screen touch information includes the touch type of each touch operation completed during the game operation process and at least one touch position for completing the touch operation; the touch type and at least one touch position are input into a pre-trained cheat detection model according to the order in which each touch operation occurs during the game operation process, so that the cheat detection model outputs a cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation process based on the touch type and at least one touch position; if the cheat suspicion degree is greater than a preset threshold, the player to be detected is determined to be a cheating player. In this way, by inputting a constructed touch type sequence and a touch position sequence from the acquired screen touch information into the trained cheat detection model, and by performing feature fusion on the constructed touch type sequence and touch position sequence, the cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation process is directly and accurately outputted, so as to detect whether the player to be detected is a cheating player, thereby helping to improve the accuracy of cheating player detection.

[0239] See also Figure 8 , Figure 8This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 8 As shown in FIG, the electronic device 800 includes a processor 810, a memory 820 and a bus 830.

[0240] The memory 820 stores machine-readable instructions executable by the processor 810. When the electronic device 800 is running, the processor 810 communicates with the memory 820 via the bus 830. When the machine-readable instructions are executed by the processor 810, the above-mentioned Figure 1 、 Figure 2 as well as Figure 4 The steps of the method for detecting cheats in games in the method embodiment shown are described in detail in the method embodiment, and will not be described in detail here.

[0241] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 、 Figure 2 as well as Figure 4 The steps of the method for detecting cheats in games in the method embodiment shown are described in detail in the method embodiment, and will not be described in detail here.

[0242] Those skilled in the art will 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.

[0243] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0244] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0245] 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.

[0246] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0247] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-mentioned embodiments within the technical scope disclosed in the present application, or perform equivalent replacements for some of the technical features thereof. These modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for detecting cheats in games, characterized in that: The cheat detection method includes: Obtaining screen touch information of the player to be detected during the game operation process; the screen touch information includes the touch type of each touch operation completed during the game operation process and at least one touch position where the touch operation is completed; wherein the touch type includes one or more of a click operation, a long press operation, and a slide operation; Extracting the touch type in each touch operation, and constructing a touch type sequence according to the order in which the touch operations occur; and extracting at least one touch position in each touch operation, and constructing a touch position sequence according to the order in which the touch operations occur; Inputting the touch type sequence and the touch position sequence into a pre-trained cheat detection model, so that the cheat detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, respectively, and outputs a cheat suspicion level of the touch behavior pattern of the player to be detected during the game operation; If the cheat suspicion level is greater than a preset threshold, the player to be detected is determined to be a cheating player; The cheat suspicion degree of the touch behavior pattern indicates whether the changes in the touch position and / or touch type of the player to be detected during the completion of the game operation conform to a regular change pattern; The change in the touch position of the to-be-detected player during the game operation process conforms to a regular change pattern if the change in the touch position of the to-be-detected player during the game operation process satisfies at least one of the following conditions: The distance between different touch positions of the player to be detected during the game operation is less than a preset distance threshold; among the multiple tracks formed by the touch position changes, the track direction and / or track length of each track are consistent with the other tracks; the tracks formed by the touch position changes form a regular polygon; The change in the touch type of the to-be-detected player during the completion of the game operation conforms to a regular change pattern if the change in the touch type of the to-be-detected player during the completion of the game operation satisfies at least one of the following conditions: During the game operation, the number of times the same touch type appears is greater than a preset threshold; the time interval between the appearances of the same touch type is less than a preset time threshold.

2. The cheat detection method according to claim 1, characterized in that: The cheat detection model includes a touch type processing sub-model, a touch position processing sub-model and a cheat prediction sub-model; Inputting the touch type sequence and the touch position sequence into a pre-trained cheat detection model, so that the cheat detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, and outputs a cheat suspicion degree of the touch behavior pattern of the player to be detected during the game operation, including: Inputting the touch type sequence into the touch type processing sub-model, so that the touch type processing sub-model outputs type change information representing the touch type change; Inputting the touch position sequence into the touch position processing sub-model, so that the touch position processing sub-model outputs position change information representing touch position changes; The type change information and the position change information are input into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the cheat suspicion of the touch behavior pattern of the player to be detected during the process of completing the game operation.

3. The cheat detection method according to claim 2, characterized in that: The touch type processing sub-model includes an embedded feature representation layer, a temporal feature extraction layer, and a first feature fusion layer; Inputting the touch type sequence into the touch type processing sub-model so that the touch type processing sub-model outputs type change information representing the touch type change includes: Inputting the touch type sequence into the embedded feature representation layer, so that the embedded feature representation layer converts the touch type of each touch operation into a corresponding first type feature vector; Inputting the obtained at least one first-type feature vector into the temporal feature extraction layer, so that the temporal feature extraction layer sequentially extracts features from the at least one first-type feature vector and outputs at least one second-type feature vector; The at least one second-type feature vector is input into the first feature fusion layer, so that the first feature fusion layer outputs the type change information according to the at least one second-type feature vector.

4. The cheat detection method according to claim 2, characterized in that: The touch position processing sub-model includes a first position feature extraction layer and a second position feature extraction layer; Inputting the touch position sequence into the touch position processing sub-model so that the touch position processing sub-model outputs position change information representing touch position changes includes: Inputting the touch position sequence into the first position feature extraction layer, so that the first position feature extraction layer converts at least one touch position into a feature vector, performs weighted fusion, and outputs at least one first position representation vector; The at least one first position representation vector is input into the second position feature extraction layer, so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient, and outputs the position change information.

5. The cheat detection method according to claim 4, characterized in that: The first position feature extraction layer includes a first temporal sub-extraction layer and a first feature sub-fusion layer; The step of inputting the touch position sequence into the first position feature extraction layer so that the first position feature extraction layer converts the touch position sequence into a feature vector according to at least one touch position, performs weighted fusion, and outputs at least one first position representation vector, including: Inputting the touch position sequence into the first sequential sub-extraction layer, so that the first sequential sub-extraction layer classifies at least one touch position according to different touch operations, performs feature extraction, and outputs at least one position feature vector; According to the order in which each touch operation occurs during the game operation, at least one position feature vector is input into the first feature sub-fusion layer, so that the first feature sub-fusion layer performs weighted fusion on the at least one position feature vector according to the weight coefficients of different dimensions in the preset vector, and outputs at least one first position representation vector.

6. The cheat detection method according to claim 4, characterized in that: The second position feature extraction layer includes a second temporal sub-extraction layer and a second feature sub-fusion layer; Inputting the at least one first position representation vector into the second position feature extraction layer so that the second position feature extraction layer weights and fuses each first position representation vector according to a preset weight coefficient and outputs the position change information, includes: Inputting at least one first position representation vector into the second sequential sub-extraction layer according to the order in which each touch operation occurs during the game operation, so that the second sequential sub-extraction layer classifies the at least one first position representation vector according to different touch operations, performs feature extraction, and outputs at least one touch type feature vector; According to the order in which each touch operation occurs during the game operation, at least one touch type feature vector is input into the second feature sub-fusion layer, so that the second feature sub-fusion layer performs weighted fusion on the weight coefficients of different dimensions in at least one touch type feature vector and outputs at least one position change information.

7. The cheat detection method according to claim 4, characterized in that: The number of the first position feature extraction layers and the number of the second position feature extraction layers are respectively consistent with the number of position coordinates included in each touch position.

8. The cheat detection method according to claim 2, characterized in that: Inputting the type change information and the position change information into the cheat prediction sub-model, so that the cheat prediction sub-model combines the type change information and the position change information to output the cheat suspicion level of the touch behavior pattern of the to-be-detected player during the completion of the game operation, includes: The type change information and the position change information are input into the plug-in prediction sub-model, so that the plug-in prediction sub-model will perform feature extraction after splicing the type change information and the position change information, determine the touch type change law and the touch position change law of the touch operation, and output the plug-in suspicion degree of the touch behavior pattern according to the type change law and the touch position change law of the touch operation.

9. The cheat detection method according to claim 1, characterized in that: The detection method further comprises: Within a preset time period, if the number of times the player to be detected is determined to be a cheating player is greater than a preset threshold, a warning message for the player to be detected is sent to other players to remind other users that the player to be detected is in an abnormal state.

10. A device for detecting cheats in games, characterized in that: The external plug-in detection device includes: An information acquisition module is configured to acquire screen touch information of a player to be detected during a game operation; the screen touch information includes the touch type of each touch operation completed during the game operation and at least one touch position at which the touch operation is completed; wherein the touch type includes one or more of a click operation, a long press operation, and a slide operation; A sequence construction module is used to extract the touch type in each touch operation, construct a touch type sequence according to the order in which the touch operations occur, and extract at least one touch position in each touch operation, and construct a touch position sequence according to the order in which the touch operations occur; a cheating suspicion output module, configured to input the touch type sequence and the touch position sequence into a pre-trained cheating detection model, so that the cheating detection model performs feature extraction and feature fusion based on the touch type sequence and the touch position sequence, and outputs a cheating suspicion level of the touch behavior pattern of the player to be detected during the game operation; A cheat detection module, configured to determine that the player to be detected is a cheating player if the cheat suspicion level is greater than a preset threshold; The cheat suspicion degree of the touch behavior pattern indicates whether the changes in the touch position and / or touch type of the player to be detected during the completion of the game operation conform to a regular change pattern; The change in the touch position of the to-be-detected player during the game operation process conforms to a regular change pattern if the change in the touch position of the to-be-detected player during the game operation process satisfies at least one of the following conditions: The distance between different touch positions of the player to be detected during the game operation is less than a preset distance threshold; among the multiple tracks formed by the touch position changes, the track direction and / or track length of each track are consistent with the other tracks; the tracks formed by the touch position changes form a regular polygon; The change in the touch type of the to-be-detected player during the completion of the game operation conforms to a regular change pattern if the change in the touch type of the to-be-detected player during the completion of the game operation satisfies at least one of the following conditions: During the game operation, the number of times the same touch type appears is greater than a preset threshold; the time interval between the appearances of the same touch type is less than a preset time threshold.

11. An electronic device, characterized in that: include: A processor, a storage medium and a bus, wherein the storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the storage medium communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of the cheat detection method in the game as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the method for detecting cheats in a game as described in any one of claims 1 to 9.

Images