Mobile game scene interaction intelligent optimization method based on generative adversarial network

Abstract

The invention discloses a mobile game scene interaction intelligent optimization method based on a generative adversarial network, and aims to improve the game experience of players. Comprising the following steps: S1, collecting behavior data and interaction information of a player in a mobile game, and generating a preliminary behavior feature model; s2, constructing a generative adversarial network model; s3, generating diversified game scene interaction examples according to the trained generator; s4, evaluating and optimizing the generated scene interaction example by using a discriminator; s5, real-time feedback of the player is received, and the behavior feature model of the player is updated; s6, dynamically adjusting a generation strategy based on the updated behavior feature model, and optimizing a scene interaction effect; s7, environment data are integrated, scene interaction is optimized in real time, and the immersion of players is enhanced; and S8, providing user-defined scene interaction options for the players, and allowing the players to adjust according to personal preferences. The interactive experience and game fun of players can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent optimization technology, and in particular to a method for intelligent optimization of mobile phone game scene interactions based on a generative adversarial network. Background Art

[0002] With the rapid development of mobile Internet technology, mobile games have become a widely popular form of entertainment around the world. According to relevant statistics, the size of the global mobile game market has exceeded US$100 billion and is still growing at a rate of more than 10% per year. Game developers are constantly exploring new technologies and methods to enhance players' gaming experience, especially in scene interaction. However, existing mobile games still have some significant defects in the interaction design between players and game scenes, which limits the realization of game immersion and personalized experience.

[0003] At present, most mobile games mainly rely on traditional interactive design methods, which usually use preset rules and static scene elements to achieve the interaction between players and the game environment. Such design concepts can guarantee the basic gameplay and logic of the game to a certain extent, but cannot fully consider the individual differences and preferences of players. Many games only provide fixed scene interaction options, and players' choices are limited by the designer's settings and cannot be flexibly adjusted according to their own game style or emotional state. This limitation makes players often feel boring and monotonous when experiencing the game, and it is difficult to have a deep sense of participation and emotional resonance.

[0004] Traditional interaction design also lacks the ability to analyze and adapt to players' emotions and behaviors in real time. In existing technologies, players' feedback is usually static and lacks dynamic response. Even if some games introduce data analysis mechanisms based on user behavior, it is still difficult to accurately capture and respond to players' emotional changes in a timely manner. Such a design makes it impossible for players to obtain consistent and personalized feedback experience in different game situations, resulting in a decrease in overall user satisfaction. The connection between players' emotional state and game interaction has not been fully understood and applied, resulting in players often feeling passive participation rather than active experience.

[0005] Existing technologies are also insufficient in terms of intelligent optimization of scene interactions. Although some mobile games try to use machine learning and data mining techniques to analyze player behavior, these methods often have difficulty effectively integrating multimodal data and cannot accurately grasp the diverse needs of users. Many games can only rely on players' historical behavior data for recommendations, but fail to deeply analyze and predict players' immediate emotions and behavioral tendencies. Such defects not only affect players' gaming experience, but also limit game developers' breakthroughs in creativity and design. Rule-based recommendation systems often ignore players' emotional responses in specific situations, resulting in insufficient relevance and effectiveness of recommended content.

[0006] Existing game scene interaction designs are often isolated and lack cross-platform synchronization mechanisms. When players play games on different devices or platforms, they often face inconsistent feedback experiences. A game setting adjusted by a player on a mobile phone may not be synchronized on a tablet or computer, resulting in the disruption of the continuity of the experience when the player switches between different devices. This lack of consistent feedback mechanism seriously affects the personalization and continuity of the user experience, restricting the long-term engagement and loyalty of players. As the trend of players using multiple devices becomes more and more obvious, ensuring cross-platform feedback consistency has become an important challenge for game developers.

[0007] Existing technologies are also inadequate in terms of dynamic generation and personalized customization of game scenes. Existing scene generation algorithms often rely on fixed parameter settings and cannot be dynamically adjusted and optimized based on real-time data or user feedback. This makes the changes in game scenes often appear monotonous and mechanical, and cannot stimulate players' desire to explore and participate. Players' exploration and interaction with the environment during the game should be an important part of game design, but the limitations of existing technologies make this part unable to be fully realized.

[0008] In summary, the existing mobile game scene interaction technology still has many shortcomings in terms of personalization, intelligence and feedback consistency. In order to improve the game experience of players, there is an urgent need for a new method that can achieve dynamic scene interaction optimization through intelligent technology to meet the personalized needs and emotional feedback of players in different situations. The mobile game scene interaction intelligent optimization method based on generative adversarial network came into being, aiming to overcome the shortcomings of existing technologies through innovative algorithms and models, and provide a richer and more personalized game interaction experience.

[0009] Therefore, how to provide an intelligent optimization method for mobile game scene interaction based on generative adversarial networks is an urgent problem that technicians in this field need to solve. Summary of the invention

[0010] One object of the present invention is to propose a method for intelligent optimization of mobile game scene interaction based on generative adversarial networks. The present invention makes full use of generative adversarial networks, behavioral data analysis and multimodal feedback mechanisms, and describes in detail the implementation process of optimizing the interactive experience of game scenes through intelligent algorithms. The method has the advantages of high personalization, realistic interactive effects and strong player immersion.

[0011] According to an embodiment of the present invention, a method for intelligent optimization of mobile game scene interaction based on a generative adversarial network comprises the following steps:

[0012] S1. Collect players’ behavior data and interaction information in mobile games and generate a preliminary behavior feature model;

[0013] S2. Construct a generative adversarial network model, including a generator and a discriminator, to train the preliminary behavior feature model. The generator adopts a nonlinear projection method based on hypersphere mapping, and the discriminator adopts a dynamic network structure based on graph theory.

[0014] S3, generate diverse game scene interaction examples based on the trained generator;

[0015] S4. Use the discriminator to evaluate the generated scene interaction examples and optimize the quality to meet the player experience requirements;

[0016] S5, receiving real-time feedback from players, updating the player's behavior feature model, and constructing a behavior feature matrix to describe the multi-dimensional features of user behavior;

[0017] S6. Based on the updated behavior feature model, dynamically adjust the generation strategy to optimize the scene interaction effect;

[0018] S7. Integrate environmental data, optimize scene interaction in real time, and optimize player immersion;

[0019] S8. Provide players with customized scene interaction options, allowing players to adjust according to personal preferences. The customized scene interaction configuration list contains the scene elements and adjustment parameters preferred by players.

[0020] Optionally, the S1 specifically includes:

[0021] S11, the behavior data set is D = {d1, d2, ..., d n}, a behavior data processing method QR(D) based on quantum state reconstruction is introduced to process the behavior data by constructing the coherent superposition form of the quantum state ψ;

[0022] S12, using a method based on topological data analysis to build a preliminary behavioral feature model, and analyzing topological features by calculating the persistent homology of the data;

[0023] S13, the feature set is F = {f1, f2, ..., f m}, and adopt a feature optimization method based on spectral graph theory to evaluate the relationship between features and improve them through the Laplacian operator of the features;

[0024] S14. Apply a behavior pattern recognition algorithm based on chaotic system to identify the player’s behavior pattern P = {p1, p2, …, p k}:

[0025]

[0026] Among them, P represents the set of identified player behavior patterns. By analyzing and clustering player behaviors, the patterns obtained can help optimize the gaming experience. C represents the cluster set. The algorithm divides the behavior data into multiple clusters. C is the set of all possible clusters. k is the number of clusters, and L is the number of possible clusters. Lyapunov (C j ) represents cluster C j The Lyapunov index is used to measure the stability and chaos of a dynamic system. The larger the value, the more unstable the system is and the faster the behavior pattern changes. Lyapunov (C j ) is used to evaluate the dynamic characteristics and diversity of clustering, Entropy (C j ) represents cluster C j The entropy is an indicator of information uncertainty or complexity. The higher the entropy, the more complex and diverse the behavior patterns in the clustering. The δ adjustment parameter is used to balance the weight between the Lyapunov exponent and the entropy.

[0027] Optionally, the S2 specifically includes:

[0028] S21. Construct a generative adversarial network model, including a generator G and a discriminator D. The generator adopts a nonlinear projection method based on hypersphere mapping:

[0029]

[0030] Among them, z is random noise, c i is the weight coefficient associated with the sample, |d i > is a high-dimensional feature vector, φ i The output of the G(z) generator is the corresponding angle parameter, which represents the generated game scene interaction sample. represents a nonlinear mapping or transformation function, which maps the internal representation of the generator to the generated sample space. n represents the dimension or number of feature vectors, that is, the number of basic features used for generation. c i With each sample feature |d i >The relevant weight coefficient controls the influence of the feature on the generated sample, sin(φ i ) is an angle parameter related to the feature, which indicates the role of the feature in the generation process. The sine function of the angle is used to introduce periodic changes and enhance the diversity of the generated samples. i >High-dimensional feature vector, representing each feature in the preliminary behavioral feature model;

[0031] S22, the discriminator adopts a dynamic network structure based on graph theory and outputs the credibility p:

[0032]

[0033] in, is the topological feature map of the sample, n j is a node in the network, W j is the weight matrix, is the similarity kernel function, β is the dynamic adjustment factor, and D(y) is the output of the discriminator, which indicates the credibility of the input sample y. represents a normalization or standardization function used to map features to a standard range. The topological feature map of sample y represents the topological structure or position of the sample in the feature space. The dynamic adjustment factor β is used to adjust the weight of similarity calculation. The number of nodes in the m network represents the complexity of the discriminator network structure. Similarity kernel function, used to calculate the similarity between the input sample y and the discriminator network node n j The similarity between them reflects the proximity of samples in the feature space, W j Weight matrix, sigmoid(W j ) function is used to map the weight value to the interval [0,1], generate probability values, and optimize the nonlinear characteristics of the discriminator output;

[0034] S23. The loss function of the generator is constructed by introducing a coherence measure based on information theory:

[0035]

[0036] in, The overall loss function represents the joint loss of the generator G and the discriminator D, guiding the training of the generator. Expected symbol, indicating the calculation of the average value under a certain distribution, y~p data (y) represents the distribution of real data p data The real sample y sampled in the logD(y) discriminator is the logarithm of the predicted probability of the real sample y, which reflects the discriminator's ability to judge the real sample. The higher the value, the more confident the discriminator is that y is a real sample. z (z) represents the random noise distribution p z The noise vector z sampled in, G(z) is the generated output of the generator G for the noise z, which represents the generated fake sample, and log(1-D(G(z))) is the logarithm of the predicted probability of the discriminator for the generated sample G(z), which reflects the discriminator's ability to judge the generated samples;

[0037] S24, the loss function of the discriminator is improved by introducing an adaptive adjustment mechanism based on fuzzy logic;

[0038] S25. Introduce feedback adjustment mechanism based on chaos theory during training;

[0039] S26. Through models and mechanisms, efficient training of the generator and discriminator is achieved, highly diverse and realistic game scene interaction samples are generated, and the player's immersion and interactive experience are optimized.

[0040] Optionally, the S5 specifically includes:

[0041] S51, generate a multidimensional matrix of behavioral characteristics, and construct a behavioral characteristic matrix X t , used to describe the multidimensional characteristics of user behavior, let t be the time step:

[0042]

[0043] in, represents the i-th behavior feature, N is the total number of features, and the matrix is ​​updated with the time step t;

[0044] S52, Generation of feedback topology model and feedback flow mapping, using feedback topology generation model The model generates the topological structure of the feedback path according to the dynamic changes of user behavior characteristics. The momentum evolution operator included in the model It is used to generate dynamic feedback at different feedback time scales τ and optimize each feedback flow J τ With the feature matrix X t Corresponding:

[0045]

[0046] in, is a hypercomplex mapping of the behavioral feature matrix, X t Convert to a higher-dimensional feature expression space to enhance the accuracy and personalization of feedback. represents the feedback path generating function, is the momentum evolution operator, which represents the dynamic response of feedback on different time scales, The behavior feature matrix X t Mapping to hypercomplex space } operators, making the feedback mechanism multi-dimensional;

[0047] S53, Generation and adjustment of behavior feedback coupling chain. In the feedback generation process, the coupling relationship between behavior characteristics and feedback chain is constructed. The coupling chain can be operated by tensor product T ij accomplish:

[0048]

[0049] in, Represents the coupling relationship matrix, which is used to represent the relationship between different feedback paths. ij(Δt) represents the feedback coupling strength between the i-th path and the j-th path at the time difference Δt, T ij is the intermediate feedback relationship calculated in the local coupling chain by the tensor product of the behavior characteristic matrix and the feedback matrix, is a local feedback matrix used to generate feedback for the behavior feature matrix X at each stage t Make updates;

[0050] S54, Optimization selection of quaternary hypercomplex feedback path, optimizing the feedback path to the quaternary hypercomplex space The optimal path set in Path Set Satisfies the minimum norm criterion:

[0051]

[0052] Here, ∥·∥ represents the norm of the feedback path, and the optimization is achieved through the complex path combination of the quaternary hypercomplex space. The quaternary hypercomplex space representing the feedback path enhances the feedback mechanism by increasing the complex dimension of the path. It is the optimal choice of feedback path, adopting the minimum norm criterion to optimize the efficiency of behavioral feedback;

[0053] S55, non-commutative feedback objective function construction, non-commutative feedback objective function L nc , the dynamic combination of multiple feedback paths solves the optimal feedback solution:

[0054]

[0055] in, represents the hypercomplex product of the feedback path, represents non-commutative superposition, L nc As the optimal objective function of the feedback mechanism, the personalized performance of feedback is improved through the non-commutative operation of each path. The optimal path set, is the momentum evolution operator.

[0056] Optionally, the S8 specifically includes:

[0057] S81. Collect player scene interaction dataset D scene , contains the player's interaction records with scenes in the game, including multimodal data features of time, interaction type, interaction frequency and emotional state;

[0058] S82: the scene interaction dataset D scene Perform preprocessing to extract the player's scene preference feature set P and arrange them in chronological order to form a time-series data stream;

[0059] S83. Clustering model using multimodal features Perform cluster analysis on the scene preference feature set P in the time series data stream to obtain the main scene interaction category set in different time periods

[0060] S84. Clustering results Adopting adaptive adjustment algorithm Analyze the players’ preference strength in different interaction scenarios and generate a custom scenario interaction configuration list C config ,The custom scene interaction configuration list contains the player’s preferred scene elements and adjustment parameters;

[0061] S85. Provide players with customized scene interaction options Allow players to dynamically adjust scene elements according to personal preferences and update the custom scene interaction configuration list C config , optimized to match the players’ latest preferences;

[0062] S86. Configure list C according to custom scene interaction config , dynamically adjust the interaction logic of the game scene Optimize players' customized preferences to be applied to game interactions;

[0063] S87. When players interact with the scene, monitor the interaction effect and the player's emotional feedback in real time to generate a feedback data set D feedback , and optimize custom configurations based on players' emotional changes and interactive effects;

[0064] S88. Through the cross-platform data synchronization method, the player's custom scene interaction configuration list C config With the feedback dataset D feedback Synchronize to different devices or platforms of players to optimize the scene interaction effects that players can maintain customization on different devices.

[0065] Beneficial Effects

[0066] 1. Improve the player experience. This invention collects and analyzes the player's behavior data and interaction information to generate a preliminary behavior feature model, thereby deeply understanding the player's preferences and needs. Based on this model, a generative adversarial network is used to generate a variety of game scene interaction examples to ensure that players can experience personalized and rich interactions in the game. This player-centric design concept makes the game more attractive and interesting, greatly improving the player's sense of immersion.

[0067] 2. Dynamic adaptability. The present invention has a real-time feedback mechanism, which can update the behavior feature model according to the real-time feedback of the players and dynamically adjust the generation strategy. This feature ensures that the game content can quickly respond to the changes of the players, adapt to the gaming habits and preferences of different players, and further enhance the interactivity and fun of the game. Each operation and feedback of the players can directly affect the progress of the game, forming a highly interactive gaming environment.

[0068] 3. Efficient scene optimization: By introducing a discriminator to evaluate the generated scene interaction examples, the interaction quality can be effectively optimized to ensure that the generated game scene meets the player's experience needs. The discriminator's evaluation not only improves the authenticity and fun of the scene, but also improves the player's satisfaction in the game. In addition, by integrating environmental data in real time and optimizing scene interactions, the game's expressiveness can be significantly improved, greatly improving the game's playability and durability.

[0069] 4. Personalized interaction options provide players with customized scene interaction options, allowing players to adjust according to personal preferences. This personalized setting not only enhances the player's sense of participation, but also enables each player to enjoy a unique gaming experience. Players can adjust the elements in the scene according to their preferences and needs, ensuring the uniqueness and personalization of the gaming experience, and enhancing the emotional connection between players and the game.

[0070] 5. Multimodal data fusion, by introducing a behavior data processing method based on quantum state reconstruction and topological data analysis, optimizes the generation and update of player behavior characteristics. This technology improves the recognition accuracy of player behavior patterns, allowing the game to more accurately reflect the needs of players and provide more targeted feedback. The system can analyze the player's behavior in different scenarios and identify the player's potential points of interest, thereby optimizing the game content and improving the interactive quality of the game.

[0071] 6. Cross-device consistency. The present invention uses a cross-platform data synchronization method to ensure that players can maintain consistent customized scene interaction effects on different devices or platforms. This feature improves the continuity and convenience of the user experience, allowing players to seamlessly switch between multiple devices while maintaining a personalized gaming experience. Whether players are on mobile phones, tablets or other devices, they can enjoy a smooth gaming experience, which undoubtedly improves the user stickiness of the game.

[0072] 7. Security and privacy protection. In the process of collecting and processing player data, the present invention pays special attention to the security and privacy protection of players. By adopting advanced data encryption technology and anonymization processing, it ensures that the player's data will not be abused, and enhances the player's trust in the game.

[0073] 8. Promote game development and innovation. The optimization method proposed in this invention provides game developers with new ideas and tools, enabling them to apply intelligent algorithms more flexibly in game design. This not only promotes the continuous innovation of games, but also provides developers with a data-driven decision-making basis, which helps to improve the market competitiveness of games. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0075] Figure 1 A summary diagram of a method for intelligent optimization of mobile game scene interaction based on a generative adversarial network proposed by the present invention;

[0076] Figure 2 The flowchart of claim 3 of the method for intelligent optimization of mobile game scene interaction based on generative adversarial network proposed by the present invention;

[0077] Figure 3 This is the flow chart of claim 4 of the method for intelligent optimization of mobile game scene interaction based on generative adversarial networks proposed by the present invention. DETAILED DESCRIPTION

[0078] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0079] refer to Figure 1-3 , a method for intelligent optimization of mobile game scene interaction based on generative adversarial network, comprising the following steps:

[0080] S1. Collect players’ behavior data and interaction information in mobile games and generate a preliminary behavior feature model;

[0081] S2. Construct a generative adversarial network model, including a generator and a discriminator, to train the preliminary behavior feature model. The generator adopts a nonlinear projection method based on hypersphere mapping, and the discriminator adopts a dynamic network structure based on graph theory.

[0082] S3, generate diverse game scene interaction examples based on the trained generator;

[0083] S4. Use the discriminator to evaluate the generated scene interaction examples and optimize the quality to meet the player experience requirements;

[0084] S5, receiving real-time feedback from players, updating the player's behavior feature model, and constructing a behavior feature matrix to describe the multi-dimensional features of user behavior;

[0085] S6. Based on the updated behavior feature model, dynamically adjust the generation strategy to optimize the scene interaction effect;

[0086] S7. Integrate environmental data, optimize scene interaction in real time, and optimize player immersion;

[0087] S8. Provide players with customized scene interaction options, allowing players to adjust according to personal preferences. The customized scene interaction configuration list contains the scene elements and adjustment parameters preferred by players.

[0088] In this implementation, S1 specifically includes:

[0089] S11, the behavior data set is D = {d1, d2, ..., d n}, a behavior data processing method QR(D) based on quantum state reconstruction is introduced to process the behavior data by constructing the coherent superposition form of the quantum state ψ;

[0090] S12, using a method based on topological data analysis to build a preliminary behavioral feature model, and analyzing topological features by calculating the persistent homology of the data;

[0091] S13, the feature set is F = {f1, f2, ..., f m}, and adopt a feature optimization method based on spectral graph theory to evaluate the relationship between features and improve them through the Laplacian operator of the features;

[0092] S14. Apply a behavior pattern recognition algorithm based on chaotic system to identify the player’s behavior pattern P = {p1, p2, …, p k}:

[0093]

[0094] Among them, P represents the set of identified player behavior patterns. By analyzing and clustering player behaviors, the patterns obtained can help optimize the gaming experience. C represents the cluster set. The algorithm divides the behavior data into multiple clusters. C is the set of all possible clusters. k is the number of clusters, and L is the number of possible clusters. Lyapunov (C j ) represents cluster C j The Lyapunov index is used to measure the stability and chaos of a dynamic system. The larger the value, the more unstable the system is and the faster the behavior pattern changes. Lyapunov (C j ) is used to evaluate the dynamic characteristics and diversity of clustering, Entropy (C j ) represents cluster C jThe entropy is an indicator of information uncertainty or complexity. The higher the entropy, the more complex and diverse the behavior patterns in the clustering. The δ adjustment parameter is used to balance the weight between the Lyapunov exponent and the entropy.

[0095] In this implementation, S2 specifically includes:

[0096] S21. Construct a generative adversarial network model, including a generator G and a discriminator D. The generator adopts a nonlinear projection method based on hypersphere mapping:

[0097]

[0098] Among them, z is random noise, c i is the weight coefficient associated with the sample, |d i > is a high-dimensional feature vector, φ i The output of the G(z) generator is the corresponding angle parameter, which represents the generated game scene interaction sample. represents a nonlinear mapping or transformation function, which maps the internal representation of the generator to the generated sample space. n represents the dimension or number of feature vectors, that is, the number of basic features used for generation. c i With each sample feature |d i >The relevant weight coefficient controls the influence of the feature on the generated sample, sin(φ i ) is an angle parameter related to the feature, which indicates the role of the feature in the generation process. The sine function of the angle is used to introduce periodic changes and enhance the diversity of the generated samples. i >High-dimensional feature vector, representing each feature in the preliminary behavioral feature model;

[0099] S22, the discriminator adopts a dynamic network structure based on graph theory and outputs the credibility p:

[0100]

[0101] in, is the topological feature map of the sample, n j is a node in the network, W j is the weight matrix, is the similarity kernel function, β is the dynamic adjustment factor, and D(y) is the output of the discriminator, which indicates the credibility of the input sample y. represents a normalization or standardization function used to map features to a standard range. The topological feature map of sample y represents the topological structure or position of the sample in the feature space. The dynamic adjustment factor β is used to adjust the weight of similarity calculation. The number of nodes in the m network represents the complexity of the discriminator network structure. Similarity kernel function, used to calculate the similarity between the input sample y and the discriminator network node n j The similarity between them reflects the proximity of samples in the feature space, W j Weight matrix, sigmoid(W j ) function is used to map the weight value to the interval [0,1], generate probability values, and optimize the nonlinear characteristics of the discriminator output;

[0102] S23. The loss function of the generator is constructed by introducing a coherence measure based on information theory:

[0103]

[0104] in, The overall loss function represents the joint loss of the generator G and the discriminator D, guiding the training of the generator. Expected symbol, indicating the calculation of the average value under a certain distribution, y~p data (y) represents the distribution of real data p data The real sample y sampled in the logD(y) discriminator is the logarithm of the predicted probability of the real sample y, which reflects the discriminator's ability to judge the real sample. The higher the value, the more confident the discriminator is that y is a real sample. z (z) represents the random noise distribution p z The noise vector z sampled in, G(z) is the generated output of the generator G for the noise z, which represents the generated fake sample, and log(1-D(G(z))) is the logarithm of the predicted probability of the discriminator for the generated sample G(z), which reflects the discriminator's ability to judge the generated samples;

[0105] S24, the loss function of the discriminator is improved by introducing an adaptive adjustment mechanism based on fuzzy logic;

[0106] S25. Introduce feedback adjustment mechanism based on chaos theory during training;

[0107] S26. Through models and mechanisms, efficient training of the generator and discriminator is achieved, highly diverse and realistic game scene interaction samples are generated, and the player's immersion and interactive experience are optimized.

[0108] In this implementation, the S5 specifically includes:

[0109] S51, generate a multidimensional matrix of behavioral characteristics, and construct a behavioral characteristic matrix X t , used to describe the multidimensional characteristics of user behavior, let t be the time step:

[0110]

[0111] in, represents the i-th behavior feature, N is the total number of features, and the matrix is ​​updated with the time step t;

[0112] S52, Generation of feedback topology model and feedback flow mapping, using feedback topology generation model The model generates the topological structure of the feedback path according to the dynamic changes of user behavior characteristics. The momentum evolution operator included in the model It is used to generate dynamic feedback at different feedback time scales τ and optimize each feedback flow J τ With the feature matrix X t Corresponding:

[0113]

[0114] in, is a hypercomplex mapping of the behavioral feature matrix, X t Convert to a higher-dimensional feature expression space to enhance the accuracy and personalization of feedback. represents the feedback path generating function, is the momentum evolution operator, which represents the dynamic response of feedback on different time scales, The behavior feature matrix X t Mapping to hypercomplex space } operators, making the feedback mechanism multi-dimensional;

[0115] S53, Generation and adjustment of behavior feedback coupling chain. In the feedback generation process, the coupling relationship between behavior characteristics and feedback chain is constructed. The coupling chain can be operated by tensor product T ij accomplish:

[0116]

[0117] in, Represents the coupling relationship matrix, which is used to represent the relationship between different feedback paths. ij (Δt) represents the feedback coupling strength between the i-th path and the j-th path at the time difference Δt, T ij is the intermediate feedback relationship calculated in the local coupling chain by the tensor product of the behavior characteristic matrix and the feedback matrix, is a local feedback matrix used to generate feedback for the behavior feature matrix X at each stage t Make updates;

[0118] S54, Optimization selection of quaternary hypercomplex feedback path, optimizing the feedback path to the quaternary hypercomplex space The optimal path set in Path Set Satisfies the minimum norm criterion:

[0119]

[0120] Here, ∥·∥ represents the norm of the feedback path, and the optimization is achieved through the complex path combination of the quaternary hypercomplex space. The quaternary hypercomplex space representing the feedback path enhances the feedback mechanism by increasing the complex dimension of the path. It is the optimal choice of feedback path, adopting the minimum norm criterion to optimize the efficiency of behavioral feedback;

[0121] S55, non-commutative feedback objective function construction, non-commutative feedback objective function L nc , the dynamic combination of multiple feedback paths solves the optimal feedback solution:

[0122]

[0123] in, represents the hypercomplex product of the feedback path, represents non-commutative superposition, L nc As the optimal objective function of the feedback mechanism, the personalized performance of feedback is improved through the non-commutative operation of each path. The optimal path set, is the momentum evolution operator.

[0124] In this implementation manner, S8 specifically includes:

[0125] S81. Collect player scene interaction dataset D scene , contains the player's interaction records with scenes in the game, including multimodal data features of time, interaction type, interaction frequency and emotional state;

[0126] S82: the scene interaction dataset D scene Perform preprocessing to extract the player's scene preference feature set P and arrange them in chronological order to form a time-series data stream;

[0127] S83. Clustering model using multimodal features Perform cluster analysis on the scene preference feature set P in the time series data stream to obtain the main scene interaction category set in different time periods

[0128] S84. Clustering results Adopting adaptive adjustment algorithm Analyze the players’ preference strength in different interaction scenarios and generate a custom scenario interaction configuration list C config ,The custom scene interaction configuration list contains the player’s preferred scene elements and adjustment parameters;

[0129] S85. Provide players with customized scene interaction options Allow players to dynamically adjust scene elements according to personal preferences and update the custom scene interaction configuration list C config , optimized to match the players’ latest preferences;

[0130] S86. Configure list C according to custom scene interaction config , dynamically adjust the interaction logic of the game scene Optimize players' customized preferences to be applied to game interactions;

[0131] S87. When players interact with the scene, monitor the interaction effect and the player's emotional feedback in real time to generate a feedback data set D feedback , and optimize custom configurations based on players' emotional changes and interactive effects;

[0132] S88. Through the cross-platform data synchronization method, the player's custom scene interaction configuration list C config With the feedback dataset D feedback Synchronize to different devices or platforms of players, and optimize the scene interaction effects that players can maintain customized on different devices.

[0133] Example

[0134] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a mobile phone game called "Brave Continent" for detailed explanation. The game is an open world role-playing game in which players explore the vast continent, complete tasks, fight monsters and level up, and interact with other players. During the game, players need to make many decisions, including choosing tasks, equipment, skills, and cooperation and confrontation with other players.

[0135] In Brave Continent, the player's gaming experience is affected by many factors, including mission difficulty, enemy strength, and resource allocation. In order to improve the player's immersion and personalized experience, the game development team decided to apply the scene interaction intelligent optimization method based on generative adversarial network of the present invention.

[0136] This embodiment will demonstrate how to analyze the player's behavior in real time through intelligent algorithms, and dynamically adjust the game scene according to their personalized characteristics, so as to improve the player's participation and satisfaction. In the game, the system will record the player's operation data in real time, including the character's movement path, attack method, skill usage frequency, and task completion. The system monitors that a player often chooses to use remote skills when performing the "killing monster" task, and actively looks for the next target after completing the task. By analyzing the collected behavioral data, the system uses a deep learning algorithm to generate a player's behavioral feature model. The player has the following characteristics in the game: the frequency of using remote attack skills is high at 75%, the survival time in enemy-dense areas is short, usually 60 seconds, the frequency of collaboration with teammates is low, the proportion of completing tasks alone is 80%, and activities are often carried out at night in the game, and single-player tasks are preferred.

[0137] Based on the generative adversarial network, the system generates multiple scene interaction examples. In the "Dark Forest" area, the system can increase the density of monsters and provide hiding places and high ground environmental features for players to use. At this time, the system will dynamically adjust the difficulty of the monsters based on the player's behavioral characteristics to ensure that the player can get the challenge but not feel frustrated.

[0138] As players play the game, the system monitors their performance in real time. When the system finds that the player's death rate increases in the new scene, and the mortality rate increases from 20% to 40%, the system quickly evaluates the generated scene interaction examples and makes adjustments. Including reducing the strength of monsters in the area, increasing the resources available to players, such as the appearance rate of medical kits and magic potions, adjusting the AI ​​of monsters to make them easier to be defeated by players, and reducing the blow to novice players. The system allows players to personalize according to their own game style. Players can choose the "Teamwork" mode, and the system will recommend teammates who match them in the game, and provide collaboration information in the task to enhance the teamwork experience. At the same time, the system will generate personalized task recommendations based on the player's historical behavior. For players who prefer single-player tasks, the system will provide relevant challenge tasks.

[0139] In order to verify the effectiveness of the present invention, the game development team conducted an experiment from August to October 2024, with a total of 2,000 players participating in the experiment. The experiment was divided into a control group and an experimental group. The experimental group applied the optimization method of the present invention, while the control group maintained the traditional game settings. The data came from the real-time monitoring system inside the game and player feedback.

[0140] Table 1. Results display

[0141]

[0142]

[0143] Data analysis shows that the average game time of players in the experimental group increased significantly, indicating that their participation in the game has increased significantly. At the same time, the reduction in the average length of a single game means that players experience higher efficiency and fun in the game. The increase in win rate and satisfaction score further verifies the effectiveness of personalized scene interaction in improving player experience. In addition, the significant increase in the frequency of props use in the experimental group shows that players are more actively using game resources, and the increase in the number of player-customized scenes shows the success of the system in personalized settings.

[0144] In the questionnaire survey after the experiment, 76% of the players said they were satisfied with the new features and believed that this personalized optimization significantly enhanced their gaming experience and fun. Players particularly mentioned that the personalized recommendations and environmental changes provided by the system made them feel fresh every time they played the game, and no longer felt monotonous and boring.

[0145] Through the specific application and data analysis of this embodiment, the beneficial effects of the intelligent optimization method for mobile game scene interaction based on generative adversarial networks in improving player experience, enhancing game interactivity and improving game satisfaction are fully demonstrated. This method not only provides players with personalized and rich game experience, but also provides game developers with innovative ideas and promotes the development of the game industry.

[0146] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for intelligent optimization of mobile game scene interaction based on generative adversarial networks, characterized in that: The steps include: S1. Collect players’ behavior data and interaction information in mobile games and generate a preliminary behavior feature model; S2. Construct a generative adversarial network model, including a generator and a discriminator, to train the preliminary behavior feature model. The generator adopts a nonlinear projection method based on hypersphere mapping, and the discriminator adopts a dynamic network structure based on graph theory. S3, generate diverse game scene interaction examples based on the trained generator; S4. Use the discriminator to evaluate the generated scene interaction examples and optimize the quality to meet the player experience requirements; S5, receiving real-time feedback from players, updating the player's behavior feature model, and constructing a behavior feature matrix to describe the multi-dimensional features of user behavior; S6. Based on the updated behavior feature model, dynamically adjust the generation strategy to optimize the scene interaction effect; S7. Integrate environmental data, optimize scene interaction in real time, and optimize player immersion; S8. Provide players with customized scene interaction options, allowing players to adjust according to personal preferences. The customized scene interaction configuration list contains the scene elements and adjustment parameters preferred by players.

2. A method for intelligent optimization of mobile game scene interaction based on generative adversarial network according to claim 1, characterized in that: The S1 specifically includes: S11, the behavior data set is D = {d1, d2, ..., d n }, a behavior data processing method QR(D) based on quantum state reconstruction is introduced to process the behavior data by constructing the coherent superposition form of the quantum state ψ; S12, using a method based on topological data analysis to build a preliminary behavioral feature model, and analyzing topological features by calculating the persistent homology of the data; S13, the feature set is F = {f1, f2, ..., f m }, and adopt a feature optimization method based on spectral graph theory to evaluate the relationship between features and improve them through the Laplacian operator of the features; S14. Apply a behavior pattern recognition algorithm based on chaotic system to identify the player’s behavior pattern P = {p1, p2, …, p k }: Among them, P represents the set of identified player behavior patterns. By analyzing and clustering player behaviors, the patterns obtained can help optimize the gaming experience. C represents the cluster set. The algorithm divides the behavior data into multiple clusters. C is the set of all possible clusters. k is the number of clusters, and L is the number of possible clusters. Lyapunov (C j ) represents cluster C j The Lyapunov index is used to measure the stability and chaos of a dynamic system. The larger the value, the more unstable the system is and the faster the behavior pattern changes. Lyapunov (C j ) is used to evaluate the dynamic characteristics and diversity of clustering, Entropy (C j ) represents cluster C j The entropy is an indicator of information uncertainty or complexity. The higher the entropy, the more complex and diverse the behavior patterns in the clustering. The δ adjustment parameter is used to balance the weight between the Lyapunov exponent and the entropy.

3. A method for intelligent optimization of mobile game scene interaction based on generative adversarial network according to claim 1, characterized in that: The S2 specifically includes: S21. Construct a generative adversarial network model, including a generator G and a discriminator D. The generator adopts a nonlinear projection method based on hypersphere mapping: Among them, z is random noise, c i is the weight coefficient associated with the sample, |d i > is a high-dimensional feature vector, φ i The output of the G(z) generator is the corresponding angle parameter, which represents the generated game scene interaction sample. represents a nonlinear mapping or transformation function, which maps the internal representation of the generator to the generated sample space. n represents the dimension or number of feature vectors, that is, the number of basic features used for generation. c i With each sample feature |d i >The relevant weight coefficient controls the influence of the feature on the generated sample, sin(φ i ) is an angle parameter related to the feature, which indicates the role of the feature in the generation process. The sine function of the angle is used to introduce periodic changes and enhance the diversity of the generated samples. i >High-dimensional feature vector, representing each feature in the preliminary behavioral feature model; S22, the discriminator adopts a dynamic network structure based on graph theory and outputs the credibility p: in, is the topological feature map of the sample, n j is a node in the network, W j is the weight matrix, is the similarity kernel function, β is the dynamic adjustment factor, and D(y) is the output of the discriminator, which indicates the credibility of the input sample y. represents a normalization or standardization function used to map features to a standard range. The topological feature map of sample y represents the topological structure or position of the sample in the feature space. The dynamic adjustment factor β is used to adjust the weight of similarity calculation. The number of nodes in the m network represents the complexity of the discriminator network structure. Similarity kernel function, used to calculate the similarity between the input sample y and the discriminator network node n j The similarity between them reflects the proximity of samples in the feature space, W j Weight matrix, sigmoid(W j ) function is used to map the weight value to the interval [0,1], generate probability values, and optimize the nonlinear characteristics of the discriminator output; S23. The loss function of the generator is constructed by introducing a coherence measure based on information theory: in, The overall loss function represents the joint loss of the generator G and the discriminator D, guiding the training of the generator. Expected symbol, indicating the calculation of the average value under a certain distribution, y~p data (y) represents the distribution of real data p data The real sample y sampled in the logD(y) discriminator is the logarithm of the predicted probability of the real sample y, which reflects the discriminator's ability to judge the real sample. The higher the value, the more confident the discriminator is that y is a real sample. z (z) represents the random noise distribution p z The noise vector z sampled in, G(z) is the generated output of the generator G for the noise z, which represents the generated fake sample, and log(1-D(G(z))) is the logarithm of the predicted probability of the discriminator for the generated sample G(z), which reflects the discriminator's ability to judge the generated samples; S24, the loss function of the discriminator is improved by introducing an adaptive adjustment mechanism based on fuzzy logic; S25. Introduce feedback adjustment mechanism based on chaos theory during training; S26. Through models and mechanisms, efficient training of the generator and discriminator is achieved, highly diverse and realistic game scene interaction samples are generated, and the player's immersion and interactive experience are optimized.

4. A method for intelligent optimization of mobile game scene interaction based on generative adversarial network according to claim 1, characterized in that: The S5 specifically includes: S51, generate a multidimensional matrix of behavioral characteristics, and construct a behavioral characteristic matrix X t , used to describe the multidimensional characteristics of user behavior, let t be the time step: in, represents the i-th behavior feature, N is the total number of features, and the matrix is ​​updated with the time step t; S52, Generation of feedback topology model and feedback flow mapping, using feedback topology generation model The model generates the topological structure of the feedback path according to the dynamic changes of user behavior characteristics. The momentum evolution operator included in the model It is used to generate dynamic feedback at different feedback time scales τ and optimize each feedback flow J τ With the feature matrix X t Corresponding: in, is a hypercomplex mapping of the behavioral feature matrix, X t Convert to a higher-dimensional feature expression space to enhance the accuracy and personalization of feedback. represents the feedback path generating function, is the momentum evolution operator, which represents the dynamic response of feedback on different time scales, The behavior feature matrix X t Mapping to hypercomplex space } operators, making the feedback mechanism multi-dimensional; S53, Generation and adjustment of behavior feedback coupling chain. In the feedback generation process, the coupling relationship between behavior characteristics and feedback chain is constructed. The coupling chain can be operated by tensor product T ij accomplish: in, Represents the coupling relationship matrix, which is used to represent the relationship between different feedback paths. ij (Δt) represents the feedback coupling strength between the i-th path and the j-th path at the time difference Δt, T ij is the intermediate feedback relationship calculated in the local coupling chain by the tensor product of the behavior characteristic matrix and the feedback matrix, is a local feedback matrix used to generate feedback for the behavior feature matrix X at each stage t Make updates; S54, Optimization selection of quaternary hypercomplex feedback path, optimizing the feedback path to the quaternary hypercomplex space The optimal path set in Path Set Satisfies the minimum norm criterion: Here, ∥·∥ represents the norm of the feedback path, and the optimization is achieved through the complex path combination of the quaternary hypercomplex space. The quaternary hypercomplex space representing the feedback path enhances the feedback mechanism by increasing the complex dimension of the path. It is the optimal choice of feedback path, adopting the minimum norm criterion to optimize the efficiency of behavioral feedback; S55, non-commutative feedback objective function construction, non-commutative feedback objective function L nc , the dynamic combination of multiple feedback paths solves the optimal feedback solution: Among them, ◇ represents the hypercomplex product of the feedback path, ⊕ represents the non-commutative superposition, L nc As the optimal objective function of the feedback mechanism, the personalized performance of feedback is improved through the non-commutative operation of each path. The optimal path set, is the momentum evolution operator.

5. A method for intelligent optimization of mobile game scene interaction based on generative adversarial network according to claim 1, characterized in that: The S8 specifically includes: S81. Collect player scene interaction dataset D scene , contains the player's interaction records with the scene in the game, including multimodal data features of time, interaction type, interaction frequency and emotional state; S82: the scene interaction dataset D scene Perform preprocessing to extract the player's scene preference feature set P and arrange them in chronological order to form a time-series data stream; S83. Clustering model using multimodal features Perform cluster analysis on the scene preference feature set P in the time series data stream to obtain the main scene interaction category set in different time periods S84. Clustering results Adopting adaptive adjustment algorithm Analyze the players’ preference strength in different interaction scenarios and generate a custom scenario interaction configuration list C config ,The custom scene interaction configuration list contains the player’s preferred scene elements and adjustment parameters; S85. Provide players with customized scene interaction options Allow players to dynamically adjust scene elements according to personal preferences and update the custom scene interaction configuration list C config , optimized to match the players’ latest preferences; S86. Configure list C according to custom scene interaction config , dynamically adjust the interaction logic of the game scene Optimize players' customized preferences to be applied to game interactions; S87. When players interact with the scene, monitor the interaction effect and the player's emotional feedback in real time to generate a feedback data set D feedback , and optimize custom configurations based on players' emotional changes and interactive effects; S88. Through the cross-platform data synchronization method, the player's custom scene interaction configuration list C config With the feedback dataset D feedback Synchronize to different devices or platforms of players to optimize the scene interaction effects that players can maintain customization on different devices.

Images