Game user behavior prediction system based on deep learning

Abstract

The invention relates to the technical field of user behavior prediction, in particular to a game user behavior prediction system based on deep learning, which comprises a conventional behavior prediction sequencing module, a conventional behavior prediction execution module, an abnormal behavior rapid determination module, a conventional behavior prediction sequencing module and a conventional behavior prediction execution module. The method comprises the following steps: periodically predicting various behaviors of a game user through a long-short-term memory model, ensuring that behaviors with high potential abnormal risks of the game user are preferentially predicted through a behavior prediction sequence, and ensuring orderliness and high efficiency of behavior prediction of the game user; after the abnormal behaviors of the game user are found, the prediction sequence of the behaviors is dynamically adjusted, the abnormal relevance between the abnormal behaviors and other behaviors is deeply analyzed, and it is guaranteed that all the abnormal behaviors of the game user are rapidly locked within a short time.

Description

Technical Field

[0001] The present invention relates to the technical field of user behavior prediction, and more specifically, to a game user behavior prediction system based on deep learning. Background Art

[0002] With the rapid development of Internet technology, online games have become one of the important ways for people to relax and have fun. The number of game users is huge and their behaviors are diverse. How to effectively manage and predict the behaviors of these users is of great significance for improving the gaming experience, ensuring game security, and optimizing operation strategies. Traditional user behavior prediction methods, such as statistical analysis models or simple machine learning algorithms, often have difficulty in capturing the complexity and dynamics of user behavior, especially when dealing with time series data and potential abnormal behaviors.

[0003] In recent years, deep learning technology has shown great potential in the field of game user behavior prediction due to its powerful feature extraction and pattern recognition capabilities. In particular, the long short-term memory (LSTM) model, as a special recurrent neural network (RNN), can effectively handle long-term dependencies in time series data and is suitable for periodic prediction of user behavior. However, existing prediction systems based on deep learning mostly focus on improving the accuracy of predictions, while there is still room for improvement in the orderliness and efficiency of predictions and the rapid identification of abnormal behaviors.

[0004] In actual applications, the behavior of game users often shows certain periodic laws, but there are also potential abnormal behaviors, such as cheating and fraud. These abnormal behaviors not only affect the fairness of the game, but may also damage the reputation and economic interests of the game operator. Therefore, how to achieve efficient prediction and rapid locking of potential abnormal behaviors while ensuring the accuracy of prediction has become a key issue that needs to be solved in the current field of game user behavior prediction.

[0005] In order to solve the above problems, some attempts have been made in the prior art to identify abnormal behaviors by introducing rule engines or anomaly detection algorithms. However, these methods often rely on manually set rules and thresholds and lack adaptability and flexibility. In addition, these methods may suffer from low computational efficiency and high false positive rate when processing large-scale user behavior data.

[0006] Therefore, there is an urgent need for a new prediction system that can comprehensively consider the periodicity of user behavior, potential abnormal risks, and the orderliness and efficiency of prediction. The system should be able to use deep learning technology, especially long-term and short-term memory models, to accurately predict various behaviors of game users, and through reasonable module design and algorithm optimization, to quickly identify and lock abnormal behaviors, thereby providing strong data support and decision-making basis for game operators.

[0007] Based on the above background, the present invention proposes a game user behavior prediction system based on deep learning. Summary of the Invention

[0008] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a game user behavior prediction system based on deep learning.

[0009] To achieve the above purpose, the present invention provides the following technical solutions:

[0010] The game user behavior prediction system based on deep learning includes a regular behavior prediction sequencing module, a regular behavior prediction execution module, and an abnormal behavior quick determination module;

[0011] The regular behavior prediction sequencing module is used to periodically generate a behavior prediction sequence for each game user, and predict each game behavior of the game user in turn based on the behavior prediction sequence;

[0012] After predicting each game behavior of a game user, the regular behavior prediction execution module obtains the behavior prediction value of this game behavior, and determines whether the game behavior is an abnormal game behavior based on the comparison result between the behavior prediction value and the behavior prediction threshold value;

[0013] When an abnormal game behavior occurs, the abnormal behavior quick determination module marks all game behaviors after the abnormal game behavior in the behavior prediction sequence as abnormal behaviors to be solved, obtains the abnormal synchronization value of each abnormal behavior to be solved, sorts all abnormal behaviors to be solved in ascending order according to the value of the abnormal synchronization value, generates an abnormal quick determination sequence for this game user, and predicts the abnormal behaviors to be solved of the game user in turn based on the abnormal quick determination sequence.

[0014] Furthermore, the process of generating the behavior prediction sequence of the game user is as follows: Obtain the potential abnormal measurement value of a game user facing various game behaviors, sort various game behaviors in descending order according to the value of the potential abnormal measurement value, and then generate the behavior prediction sequence of this game user.

[0015] Further, the process of obtaining the potential deviation value of game behavior is as follows: Obtain the behavior prediction values of a game behavior in the previous j cycles, sort all the behavior prediction values in the order of the cycles, compare the subsequent behavior prediction value with the previous one after sorting. When the subsequent behavior prediction value is greater than the previous one, increase the abnormal exacerbation count by one. When the subsequent behavior prediction value is less than or equal to the previous one, do nothing and mark the abnormal exacerbation count as Lvm. Calculate the difference between two adjacent behavior prediction values after sorting and take the absolute value to obtain the behavior swing value. Sum up all the behavior swing values and take the average to obtain the behavior prediction swing value Sew. Sum up all the behavior prediction values and take the average to obtain the behavior prediction mean Mds. Use the formula Bzx = Lvm * (Sew * y1 + Mds * y2) to obtain the potential deviation value Bzx of this game behavior, where y1 is the prediction swing coefficient and y2 is the behavior prediction coefficient.

[0016] Further, after predicting a game behavior of each game user, obtain the behavior prediction value of this game behavior. Specifically: Obtain the game behavior feature set of the game user for a game behavior, obtain the game behavior prediction model of the game user for this game behavior, use the game behavior feature set as the input data of the game behavior prediction model, and output the behavior prediction value of this game behavior.

[0017] Further, the process of obtaining the game behavior feature set of a game user for a game behavior is as follows: Obtain the game behavior features of the game user for the same game behavior in the previous x cycles, and form the game behavior feature set with all the game behavior features in the form of a time series data set.

[0018] Further, the process of obtaining the game behavior features of a game behavior in one cycle is as follows: Collect all the game behavior data of the game user for a game behavior in one cycle, and perform feature extraction on all the game behavior data to obtain the game behavior features.

[0019] Further, the method for obtaining the abnormal synchronization value of an abnormal behavior to be solved is as follows: Determine an abnormal behavior to be solved, obtain the historical behavior analysis value of the abnormal behavior to be solved, synchronously obtain the historical behavior analysis value of the abnormal game behavior, calculate the difference between the historical behavior analysis value of the abnormal behavior to be solved and the historical behavior analysis value of the abnormal game behavior and take the absolute value to obtain the abnormal synchronization value of the abnormal behavior to be solved.

[0020] Further, the process of obtaining the historical behavior analysis value is as follows: Obtain the behavior prediction values of a certain game behavior in the previous d cycles, set the high value and low value of behavior prediction. When the behavior prediction value is greater than or equal to the high value of behavior prediction, increase the prediction deviation count by one, and mark the prediction deviation count as Phm. When the behavior prediction value is less than or equal to the low value of behavior prediction, increase the prediction ideal count by one, and mark the prediction ideal count as Xus. When the behavior prediction value is between the high value and low value of behavior prediction, mark this behavior prediction value as the normal prediction value, sum up all the normal prediction values and take the average to obtain the average normal prediction value Fzw. Use the formula to obtain the historical behavior analysis value Ldt of this game behavior, where a1 is the prediction deviation coefficient, a2 is the prediction ideal coefficient, and a3 is the normal prediction coefficient.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. Set up the normal behavior prediction sequencing module and the normal behavior prediction execution module, and perform periodic prediction on various behaviors of game users through the long short-term memory model. Ensure the prediction of behaviors with high potential abnormal risks of game users first through the behavior prediction sequence, and ensure the orderliness and efficiency of game user behavior prediction;

[0023] 2. Set up the abnormal behavior quick determination module. After discovering the abnormal behavior of game users, dynamically adjust the prediction order of behaviors, and deeply analyze the abnormal correlation between abnormal behaviors and other behaviors to ensure quickly locking all abnormal behaviors of game users in a short time. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is the system operation flowchart of the game user behavior prediction system based on deep learning;

[0025] Figure 2 is the system module diagram of the game user behavior prediction system based on deep learning;

[0026] Figure 3 is the generation flowchart of the abnormal quick determination sequence. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] Refer to Figures 1 - 3 , the game user behavior prediction system based on deep learning includes a normal behavior prediction sequencing module, a normal behavior prediction execution module, and an abnormal behavior quick determination module.

[0028] Regular Behavior Prediction Sequential Generation Module: Periodically generate a behavior prediction sequence for each game user (each game user corresponds to a behavior prediction sequence), and sequentially predict each game behavior of the game user based on the behavior prediction sequence (the types of game behaviors include interaction behaviors, consumption behaviors, collection behaviors, leisure behaviors, etc.).

[0029] The process of generating the behavior prediction sequence of a game user is as follows: Obtain the potential deviation values of a game user for various game behaviors, sort various game behaviors in descending order according to the values of the potential deviation values, and then generate the behavior prediction sequence of the game user.

[0030] The process of obtaining the potential deviation value of a game behavior is as follows: Obtain the behavior prediction values of a game behavior in the previous j cycles, sort all the behavior prediction values in the order of the cycles, compare the latter behavior prediction value with the former behavior prediction value after sorting. When the latter behavior prediction value is greater than the former behavior prediction value, increase the abnormal aggravation count by one. When the latter behavior prediction value is less than or equal to the former behavior prediction value, do not process it, mark the abnormal aggravation count as Lvm, calculate the difference between two adjacent behavior prediction values after sorting and take the absolute value to obtain the behavior swing value, sum up all the behavior swing values and take the average to obtain the behavior prediction swing value Sew, sum up all the behavior prediction values and take the average to obtain the behavior prediction mean Mds, and use the formula Bzx = Lvm * (Sew * y1 + Mds * y2) to obtain the potential deviation value Bzx of this game behavior, where y1 is the prediction swing coefficient, y2 is the behavior prediction coefficient, the value of y1 is 0.84, and the value of y2 is 0.43.

[0031] Regular Behavior Prediction Execution Module: After predicting each game behavior of a game user, obtain the behavior prediction value of this game behavior, set the behavior prediction boundary value (the behavior prediction boundary value is a threshold). When the behavior prediction value of the game behavior is greater than the behavior prediction boundary value, mark this game behavior as an abnormal game behavior. When the behavior prediction value of the game behavior is less than or equal to the behavior prediction boundary value, continue to predict the next game behavior based on the behavior prediction sequence.

[0032] After predicting each game behavior of a game user, obtain the behavior prediction value of this game behavior. Specifically: Obtain the game behavior feature set of the game user for a game behavior, obtain the game behavior prediction model of the game user for this game behavior, use the game behavior feature set as the input data of the game behavior prediction model, and output the behavior prediction value of this game behavior.

[0033] The process of obtaining the game behavior feature set for a game behavior by a game user is as follows: Obtain the game behavior features of the game user for the same game behavior in the previous x cycles (each cycle corresponds to a game behavior feature), and form the game behavior feature set by taking all the game behavior features in the form of a time series data set.

[0034] The process of obtaining the game behavior feature of a game behavior in one cycle is as follows: Collect all the game behavior data of the game user for a game behavior in one cycle, and perform feature extraction on all the game behavior data to obtain the game behavior feature.

[0035] Different game behaviors of the same game user respectively correspond to different game behavior prediction models. For example, the interaction behavior and consumption behavior of game user A respectively correspond to two game behavior prediction models. The same game behavior of different game users also corresponds to different game behavior prediction models. For example, the interaction behavior of game user A and the interaction behavior of game user B respectively correspond to two game behavior prediction models. In this embodiment, taking the interaction behavior of game user A as an example, the construction process of the game behavior prediction model for the interaction behavior of game user A is disclosed: Collect multiple game behavior feature sets of game user A for the interaction behavior, construct an LSTM model, use the game behavior feature set as the training data of the LSTM model, assign a behavior prediction value to each training data, and the value range of the behavior prediction value is (1.1 - 3.9). The closer the value of the behavior prediction value is to 3.9, the more abnormal the interaction behavior of game user A is. The closer the value of the behavior prediction value is to 1.1, the more normal the interaction behavior of game user A is. Divide the training data into a training set, a validation set, and a test set according to the set ratio of 5:1:1, and train the training set, the validation set, and the test set. After the training is completed, the game behavior prediction model for the interaction behavior of game user A is constructed.

[0036] Set up a regular behavior prediction sequencing module and a regular behavior prediction execution module, and perform periodic prediction on various behaviors of the game user through a long short-term memory model. Ensure that the behaviors with high potential abnormal risks of the game user are preferentially predicted through the behavior prediction sequence, and ensure the orderliness and efficiency of the game user behavior prediction.

[0037] Abnormal behavior quick determination module: When an abnormal game behavior occurs, mark all the game behaviors after the abnormal game behavior in the behavior prediction sequence as abnormal behaviors to be solved, obtain the abnormal synchronization values of each abnormal behavior to be solved, sort all the abnormal behaviors to be solved in ascending order according to the numerical values of the abnormal synchronization values, generate the abnormal quick determination sequence of this game user, and predict the abnormal behaviors to be solved of the game user in turn based on the abnormal quick determination sequence.

[0038] The method for obtaining the abnormal synchronization value of an abnormal behavior to be resolved is as follows: Determine an abnormal behavior to be resolved, obtain the historical behavior analysis value of the abnormal behavior to be resolved, synchronously obtain the historical behavior analysis value of the abnormal game behavior, calculate the difference between the historical behavior analysis value of the abnormal behavior to be resolved and the historical behavior analysis value of the abnormal game behavior and take the absolute value to obtain the abnormal synchronization value of the abnormal behavior to be resolved.

[0039] The process of obtaining the historical behavior analysis value is as follows: Obtain the behavior prediction value of a game behavior in the previous d cycles, set the high behavior prediction value and the low behavior prediction value (the high behavior prediction value is greater than the low behavior prediction value, and both the high behavior prediction value and the low behavior prediction value are system-preset exponents). When the behavior prediction value is greater than or equal to the high behavior prediction value, increase the prediction overabnormal times by one, and mark the prediction overabnormal times as Phm. When the behavior prediction value is less than or equal to the low behavior prediction value, increase the prediction ideal times by one, and mark the prediction ideal times as Xus. When the behavior prediction value is between the high behavior prediction value and the low behavior prediction value, mark this behavior prediction value as the normal prediction value, sum up all the normal prediction values and take the average to obtain the average normal prediction value Fzw. Use the formula to obtain the historical behavior analysis value Ldt of this game behavior, where a1 is the prediction overabnormal coefficient, a2 is the prediction ideal coefficient, a3 is the normal prediction coefficient, the value of a1 is 1.03, the value of a2 is 1.01, and the value of a3 is 0.81.

[0040] Set up an abnormal behavior quick determination module. After detecting the abnormal behavior of the game user, dynamically adjust the prediction order of the behavior, and deeply analyze the abnormal correlation between the abnormal behavior and the other behaviors to ensure that all the abnormal behaviors of the game user can be quickly locked in a short time.

[0041] The above formulas are all dimensionless and take their numerical calculations. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0042] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that contains a collection of one or more available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0043] It should be understood that in various embodiments of the present application, the order numbers of the above processes do not indicate the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0044] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0045] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0046] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0047] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing 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 methods described in the various embodiments of the present application. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs, and other various media that can store program codes.

[0048] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present application can easily think of changes or substitutions, which should all be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A game user behavior prediction system based on deep learning, characterized in that, It includes a regular behavior prediction sequencing module, a regular behavior prediction execution module, and an abnormal behavior quick determination module; The regular behavior prediction sequencing module is used to periodically generate a behavior prediction sequence for each game user, and sequentially predict each game behavior of the game user based on the behavior prediction sequence; After predicting a game behavior of a game user each time, the regular behavior prediction execution module obtains the behavior prediction value of this game behavior, and determines whether the game behavior is an abnormal game behavior based on the comparison result between the behavior prediction value and the behavior prediction threshold value; When an abnormal game behavior occurs, the abnormal behavior quick determination module marks all game behaviors after the abnormal game behavior in the behavior prediction sequence as abnormal behaviors to be solved, obtains the abnormal synchronization values of each abnormal behavior to be solved, sorts all abnormal behaviors to be solved in ascending order according to the values of the abnormal synchronization values, generates an abnormal quick determination sequence for this game user, and sequentially predicts the abnormal behaviors to be solved of the game user based on the abnormal quick determination sequence.

2. The game user behavior prediction system based on deep learning according to claim 1, wherein The process of generating the behavior prediction sequence of a game user is as follows: Obtain the potential anomaly detection values of a game user for various game behaviors, sort various game behaviors in descending order according to the values of the potential anomaly detection values, and then generate the behavior prediction sequence of this game user.

3. The game user behavior prediction system based on deep learning according to claim 2, wherein, The process of obtaining the potential anomaly detection value of a game behavior is as follows: Obtain the behavior prediction values of a game behavior in the previous j cycles, sort all behavior prediction values in the order of the cycles, compare the latter behavior prediction value with the previous one after sorting. When the latter behavior prediction value is greater than the previous one, increase the anomaly aggravation count by one. When the latter behavior prediction value is less than or equal to the previous one, do nothing and mark the anomaly aggravation count as Lvm. Calculate the difference between two adjacent behavior prediction values after sorting and take the absolute value to obtain the behavior swing value. Sum up all behavior swing values and take the average to obtain the behavior prediction swing value Sew. Sum up all behavior prediction values and take the average to obtain the behavior prediction mean Mds. Use the formula Bzx = Lvm * (Sew * y1 + Mds * y2) to obtain the potential anomaly detection value Bzx of this game behavior, where y1 is the prediction swing coefficient and y2 is the behavior prediction coefficient.

4. The game user behavior prediction system based on deep learning according to claim 1, characterized in that, After predicting a game behavior of a game user each time, the behavior prediction value of this game behavior is obtained specifically as follows: Obtain the game behavior feature set of the game user for a game behavior, obtain the game behavior prediction model of the game user for this game behavior, use the game behavior feature set as the input data of the game behavior prediction model, and output the behavior prediction value of this game behavior.

5. The game user behavior prediction system based on deep learning according to claim 4, characterized in that The process of obtaining the game behavior feature set of a game user for a game behavior is as follows: Obtain the game behavior features of the game user for the same game behavior in the previous x cycles, and form the game behavior feature set with all game behavior features in the form of a time series data set.

6. The game user behavior prediction system based on deep learning according to claim 5, wherein The process of obtaining the game behavior characteristics in one cycle for a game behavior is as follows: Collect all the game behavior data of a game user for a game behavior in one cycle, and perform feature extraction on all the game behavior data to obtain the game behavior characteristics.

7. The game user behavior prediction system based on deep learning according to claim 1, characterized in that, The method for obtaining the anomaly synchronization value of an anomaly to-be-solved behavior is as follows: Determine an anomaly to-be-solved behavior, obtain the historical behavior analysis value of the anomaly to-be-solved behavior, synchronously obtain the historical behavior analysis value of the abnormal game behavior, calculate the difference between the historical behavior analysis value of the anomaly to-be-solved behavior and the historical behavior analysis value of the abnormal game behavior and take the absolute value to obtain the anomaly synchronization value of the anomaly to-be-solved behavior.

8. The game user behavior prediction system based on deep learning according to claim 7, wherein The process of obtaining the historical behavior analysis value is as follows: Obtain the behavior prediction value of a certain game behavior in the previous d cycles, set the high value and low value of behavior prediction. When the behavior prediction value is greater than or equal to the high value of behavior prediction, increase the prediction abnormal times by one, and mark the prediction abnormal times as Phm. When the behavior prediction value is less than or equal to the low value of behavior prediction, increase the prediction ideal times by one, and mark the prediction ideal times as Xus. When the behavior prediction value is between the high value and low value of behavior prediction, mark this behavior prediction value as the normal prediction value, sum up all the normal prediction values and take the average to obtain the average normal prediction value Fzw. Use the formula to obtain the historical behavior analysis value Ldt of this game behavior, where a1 is the prediction abnormal coefficient, a2 is the prediction ideal coefficient, and a3 is the normal prediction coefficient.

Images