Game parameter generation methods, systems, devices and media
By constructing a personalized emotional response model and optimizing it with Bayesian algorithms, target game plot parameters are generated, solving the problem of existing technologies being unable to accurately perceive player emotions and micro-controls, thus enhancing the personalized and immersive experience of the game plot.
Patent Information
- Application Number
- CN202610400197.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-30
- Publication Date
- 2026-07-03
AI Technical Summary
Existing methods for generating game parameters cannot accurately perceive players' inner emotional states, lack personalized emotional response modeling, cannot proactively guide players' emotional experiences, and are difficult to finely control micro-narrative elements, resulting in homogenized game experiences and discrepancies between emotional design intentions and player feelings.
By acquiring multidimensional behavioral data, a personalized emotional response model is constructed. Combined with a Bayesian algorithm, the model searches within the game's narrative parameter space to generate target game narrative parameters. This allows for matching the player's emotional state characteristic trajectory and enables refined control over micro-narrative parameters such as emotional tone, character relationships, and speech rate.
It achieves accurate perception of players' real-time emotional state, adapts to the unique emotional baseline and response range of different players, enhances the personalized immersive experience of the game's storyline, and ensures that the design intent is accurately conveyed to each player.
Smart Images
Figure CN122333398A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the technical field of game design, and in particular to methods, systems, devices and media for generating game parameters. Background Technology
[0002] With the rapid development of the gaming industry, players' demands for gaming experiences are increasing. They not only focus on the operability and challenge of games, but also pursue deep emotional immersion and personalized narrative experiences. However, current methods for generating game parameters mainly rely on fixed difficulty levels or simple dynamic adjustments based on rules. For example, they linearly adjust the game difficulty coefficient based on behavioral indicators such as player completion time and number of deaths. These methods have many limitations in practical applications.
[0003] Firstly, adjustments to game parameters only focus on players' external behaviors, such as whether they pass a level or how long it takes, failing to perceive their internal emotional state. Different players experience significantly different emotions when faced with the same game storyline. For example, the same tense scene might excite and engage one player while causing frustration for another. This lack of emotional perception leads to a disconnect between game parameter adjustments and players' actual emotional needs, making it difficult to achieve precise emotional alignment.
[0004] Secondly, existing methods lack personalized modeling of players' emotional response characteristics. Each player has a unique emotional baseline, i.e., the baseline level of emotion without stimulation, and a unique emotional response amplitude, i.e., the intensity of fluctuation when faced with stimulation. Using a "one-size-fits-all" adjustment strategy cannot adapt to the emotional response patterns of different players, making the homogenization of the game experience prominent and making it difficult to achieve true personalized immersion in the game's storyline.
[0005] Third, current methods are mostly reactive adjustments, meaning the difficulty is lowered only after poor player performance is detected, lacking the ability to proactively guide the player's emotional experience. The emotional curves carefully designed by game designers, such as the fluctuations of calm-tension-climax-release, are difficult to accurately convey to each player's actual experience under a passive adjustment mode, resulting in a discrepancy between design intent and player perception.
[0006] Fourth, the granularity of game parameter adjustments is usually coarse, often only changing the overall difficulty level or the direction of plot branches, and failing to finely control the micro-narrative elements that affect the emotional experience, such as the emotional tone of dialogue, the dynamic changes in character relationships, and the adjustment of speech speed and rhythm, thus limiting the subtlety of emotional expression in the game's plot.
[0007] How to proactively generate game plot parameters that can guide players' emotional experiences to the expected goals based on their personalized emotional response patterns, thereby enhancing players' personalized immersive experience in the game's plot, has become a pressing technical problem that needs to be solved. Summary of the Invention
[0008] To enhance players' personalized and immersive experience in the game's storyline, this application provides methods, systems, devices, and media for generating game parameters.
[0009] Firstly, this application provides a method for generating game parameters, employing the following technical solution: Game parameter generation methods include: The system acquires multidimensional behavioral data of the target player during the game and performs correlation mining on the multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. Based on the emotional state feature sequence, a personalized emotional response model is constructed for the target player, and a target emotional state feature trajectory is generated for the target player according to the preset emotional change template and the personalized emotional response model. The target emotional state feature trajectory is used to characterize the relative emotional change trend that the target player is expected to achieve in the subsequent game process. Acquire historical game interaction data and construct an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences; The emotional state prediction model is invoked, and a set of target game plot parameters is generated by searching in the game plot parameter space based on the Bayesian algorithm, so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory.
[0010] By employing the aforementioned technical solution, multidimensional behavioral data of the target player during gameplay is acquired, and correlation mining is performed on this multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. Then, based on the emotional state feature sequence, a personalized emotional response model for the target player is constructed. According to a preset emotional change template and the personalized emotional response model, a target emotional state feature trajectory for the target player is generated. This target emotional state feature trajectory represents the expected relative emotional change trend of the target player in subsequent gameplay. Next, historical game interaction data is acquired, and an emotional state prediction model is constructed based on this data. This historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. The emotional state prediction model is then invoked, and a Bayesian algorithm is used to search the game plot parameter space to generate a set of target game plot parameters, ensuring that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory. This invention achieves accurate perception of players' real-time emotional states through correlation mining of multi-dimensional behavioral data; it adapts to the unique emotional baselines and response amplitudes of different players by constructing a personalized emotional response model; it generates personalized target trajectories based on preset emotional change templates, and combines emotional state prediction models with Bayesian optimization algorithms to achieve proactive and refined control of micro-narrative parameters such as emotional tone, character relationships, and speech rate, thereby accurately conveying the designer's emotional intentions to each player's actual experience and enhancing the personalized immersive experience of the game's storyline.
[0011] Optionally, the step of performing correlation mining on the multidimensional behavioral data to obtain the emotional state feature sequence includes: Each type of data in the multidimensional behavioral data is preprocessed and feature extracted to obtain the corresponding single-modal feature vector; Align each of the single-modal feature vectors along the time dimension to obtain a multimodal feature matrix; The multimodal feature matrix is input into a cross-modal attention network, and the correlation weights between different modal features are calculated through a self-attention mechanism. The multimodal feature matrix is then weighted and fused based on the correlation weights to obtain a fused feature vector sequence. The fused feature vector sequence is input into a temporal coding network, and the dependencies in the time dimension are modeled by a recurrent neural network or a Transformer to output an emotional state feature sequence.
[0012] By adopting the above technical solution, in order to realize the correlation mining of multi-dimensional behavioral data, each type of data in the multi-dimensional behavioral data is preprocessed and feature extracted to obtain the corresponding single-modal feature vector. Then, the single-modal feature vectors are aligned in the time dimension to obtain a multi-modal feature matrix. The multi-modal feature matrix is then input into a cross-modal attention network, and the correlation weights between different modal features are calculated through a self-attention mechanism. Based on the correlation weights, the multi-modal feature matrix is weighted and fused to obtain a fused feature vector sequence. The fused feature vector sequence is then input into a temporal coding network, and the dependencies in the time dimension are modeled through a recurrent neural network or Transformer to output an emotional state feature sequence.
[0013] Optionally, the step of inputting the multimodal feature matrix into a cross-modal attention network, calculating the correlation weights between different modal features through a self-attention mechanism, and weighting and fusing the multimodal feature matrix based on the correlation weights to obtain a fused feature vector sequence includes: The multimodal feature matrices are mapped to query matrices, key matrices, and value matrices, respectively. Calculate the similarity between the query matrix and the key matrix to obtain the attention weight matrix; The attention weight matrix is normalized to obtain a normalized weight matrix; Based on the normalized weight matrix, the value matrix is weighted and summed to obtain the single-head attention output features; The single-head attention output features corresponding to multiple different attention heads are concatenated and then dimensionality is reduced through a linear transformation layer to obtain a fused feature vector sequence.
[0014] By adopting the above technical solution, in order to generate the fused feature vector sequence, the multimodal feature matrix is mapped to a query matrix, a key matrix, and a value matrix, respectively. Then, the similarity between the query matrix and the key matrix is calculated to obtain the attention weight matrix. The attention weight matrix is then normalized to obtain a normalized weight matrix. Based on the normalized weight matrix, the value matrix is weighted and summed to obtain the single-head attention output features. Then, the single-head attention output features corresponding to multiple different attention heads are concatenated and dimensionality is reduced through a linear transformation layer to obtain the fused feature vector sequence.
[0015] Optionally, the step of constructing a personalized emotional response model for the target player based on the emotional state feature sequence includes: Obtain the emotional state feature sequence of the target player within a preset time period prior to the current moment; Statistical analysis is performed on the emotional state feature sequence to extract the emotional baseline feature and emotional response amplitude feature of the target player. The emotional baseline feature is used to characterize the baseline level of the target player's emotional state under no external stimulus or low intensity stimulus, and the emotional response amplitude feature is used to characterize the fluctuation intensity of the target player's emotional state when facing game plot stimuli. Based on the emotional baseline features and the emotional response amplitude features, a personalized emotional response model for the target player is constructed, wherein the personalized emotional response model is used to describe the changing pattern of the target player's emotional state with the stimulation of the game's plot.
[0016] By adopting the above technical solution, in order to construct a personalized emotional response model, the emotional state feature sequence of the target player within a preset time period before the current moment is obtained. Then, the emotional state feature sequence is statistically analyzed to extract the target player's emotional baseline feature and emotional response amplitude feature. The emotional baseline feature is used to characterize the target player's baseline emotional state level under no external stimuli or low-intensity stimuli, and the emotional response amplitude feature is used to characterize the fluctuation intensity of the target player's emotional state when facing game plot stimuli. Then, based on the emotional baseline feature and emotional response amplitude feature, a personalized emotional response model of the target player is constructed. The personalized emotional response model is used to describe the change pattern of the target player's emotional state with game plot stimuli.
[0017] Optionally, the step of generating a target emotional state feature trajectory for the target player based on a preset emotional change template and the personalized emotional response model includes: Obtain a preset emotion change template, wherein the emotion change template is a standardized emotion fluctuation curve associated with the current game plot segment, and the emotion change template is used to characterize the change in the intensity of the emotion that the player is expected to experience in the current game plot segment over time. Based on the personalized emotion response model, the emotion change template is mapped and transformed to obtain an emotion state feature transformation sequence; Based on the emotional state feature transformation sequence, a target emotional state feature trajectory is generated for the target player. The target emotional state feature trajectory retains the fluctuation pattern of the emotional change template, and the corresponding numerical range and fluctuation amplitude match the target player's emotional baseline feature and emotional response amplitude feature.
[0018] By adopting the above technical solution, in order to generate the target emotional state feature trajectory, a preset emotional change template is obtained. The emotional change template is a standardized emotional fluctuation curve associated with the current game plot segment. The emotional change template is used to characterize the change in the intensity of the emotion that the player is expected to experience in the current game plot segment over time. Then, based on the personalized emotional response model, the emotional change template is mapped and transformed to obtain an emotional state feature transformation sequence. Then, based on the emotional state feature transformation sequence, a target emotional state feature trajectory for the target player is generated. The target emotional state feature trajectory retains the fluctuation pattern of the emotional change template, and the corresponding numerical range and fluctuation amplitude match the target player's emotional baseline characteristics and emotional response amplitude characteristics.
[0019] Optionally, the step of acquiring historical game interaction data and constructing an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences, includes: Historical game interaction data is acquired and preprocessed to obtain a training sample set, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. An initial emotional state prediction model is constructed, comprising an input layer, a hidden layer, and an output layer. The input layer is used to receive the historical game plot parameters, the hidden layer is used to learn the mapping relationship between the historical game plot parameters and the historical emotional state feature sequence, and the output layer is used to output the predicted emotional state feature sequence. The training sample set is divided into a training set and a validation set. The initial sentiment state prediction model is iteratively trained using the training set and validated using the validation set. When the prediction error of the initial sentiment state prediction model on the validation set meets the preset accuracy requirement, training is stopped, and the trained initial sentiment state prediction model is used as the sentiment state prediction model.
[0020] By adopting the above technical solution, in order to construct an emotional state prediction model, historical game interaction data is acquired and preprocessed to obtain a training sample set. The historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. Then, an initial emotional state prediction model is constructed, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive historical game plot parameters, the hidden layer is used to learn the mapping relationship between historical game plot parameters and historical emotional state feature sequences, and the output layer is used to output the predicted emotional state feature sequences. The training sample set is then divided into a training set and a validation set. The initial emotional state prediction model is iteratively trained using the training set and validated using the validation set. When the prediction error of the initial emotional state prediction model on the validation set meets the preset accuracy requirements, training stops, and the trained initial emotional state prediction model is used as the emotional state prediction model.
[0021] Optionally, the step of calling the emotional state prediction model and searching in the game plot parameter space based on the Bayesian algorithm to generate a set of target game plot parameters, so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory, includes: Initialize the Gaussian process proxy model and define the Bayesian optimized acquisition function; Several parameter points are randomly sampled in the game plot parameter space. Each parameter point is input into the emotional state prediction model to obtain the corresponding emotional state feature prediction sequence. The matching degree between the emotional state feature prediction sequence and the target emotional state feature trajectory is calculated. The matching degree is used as the objective function value of each parameter point. If there is a parameter point whose matching degree is less than a preset threshold, then that parameter point is used as the set of target game plot parameters; If there are no parameter points with a matching degree less than a preset threshold, then each parameter point and its corresponding objective function value are used as observation data to update the Gaussian process proxy model. The acquisition function is optimized based on the updated Gaussian process proxy model. The next parameter point to be evaluated is determined in the game plot parameter space. The parameter point is input into the emotion state prediction model to obtain a new objective function value. The Gaussian process proxy model is iteratively updated based on the parameter point and the objective function value. When the preset iteration termination condition is met, the parameter point with the highest matching degree in the observed data is taken as a set of target game plot parameters.
[0022] By adopting the above technical solution, in order to search for game plot parameters, a Gaussian process proxy model is initialized, and a Bayesian optimized acquisition function is defined. Then, several parameter points are randomly sampled in the game plot parameter space, and each parameter point is input into the emotion state prediction model to obtain the corresponding emotion state feature prediction sequence. The matching degree between the emotion state feature prediction sequence and the target emotion state feature trajectory is calculated, and the matching degree is used as the objective function value of each parameter point. If there is a parameter point with a matching degree less than a preset threshold, then the parameter point is used as a set of target game plot parameters. If there is no parameter point with a matching degree less than the preset threshold, then each parameter point and the corresponding objective function value are used as observation data to update the Gaussian process proxy model. Then, based on the updated Gaussian process proxy model, the acquisition function is optimized to determine the next parameter point to be evaluated in the game plot parameter space. This parameter point is input into the emotion state prediction model to obtain a new objective function value, and the Gaussian process proxy model is iteratively updated based on the parameter point and the objective function value. When the preset iteration termination condition is met, the parameter point with the highest matching degree in the observation data is used as a set of target game plot parameters.
[0023] Secondly, this application also provides a game parameter generation system, which adopts the following technical solution: The game parameter generation system includes: The association mining module is used to acquire multidimensional behavioral data of the target player during the game, and to perform association mining on the multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. The personalized emotion modeling module is used to construct a personalized emotion response model for the target player based on the emotion state feature sequence, and generate a target emotion state feature trajectory for the target player according to a preset emotion change template and the personalized emotion response model. The target emotion state feature trajectory is used to characterize the relative emotion change trend that the target player is expected to achieve in the subsequent game process. The model building module is used to acquire historical game interaction data and build an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. The plot parameter optimization module is used to call the emotional state prediction model and search in the game plot parameter space based on the Bayesian algorithm to generate a set of target game plot parameters so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory.
[0024] Thirdly, this application also provides a computer device, which adopts the following technical solution: A computer device includes a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the computer program to implement the method described in the first aspect.
[0025] Fourthly, this application also provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program capable of being loaded by a processor and executing the method described in the first aspect.
[0026] In summary, this application includes at least the following beneficial technical effects: acquiring multidimensional behavioral data of the target player during the game, and performing correlation mining on the multidimensional behavioral data to obtain an emotional state feature sequence, wherein the multidimensional behavioral data includes at least two of physiological signal data, operational behavior data, facial expression data, and voice data; then, based on the emotional state feature sequence, constructing a personalized emotional response model for the target player, and generating a target emotional state feature trajectory for the target player according to a preset emotional change template and the personalized emotional response model, wherein the target emotional state feature trajectory is used to characterize the relative emotional change trend expected by the target player in the subsequent game process; then, acquiring historical game interaction data, and constructing an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences; then, calling the emotional state prediction model, and searching in the game plot parameter space based on the Bayesian algorithm to generate a set of target game plot parameters, so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory. This invention achieves accurate perception of players' real-time emotional states through correlation mining of multi-dimensional behavioral data; it adapts to the unique emotional baselines and response amplitudes of different players by constructing a personalized emotional response model; it generates personalized target trajectories based on preset emotional change templates, and combines emotional state prediction models with Bayesian optimization algorithms to achieve proactive and refined control of micro-narrative parameters such as emotional tone, character relationships, and speech rate, thereby accurately conveying the designer's emotional intentions to each player's actual experience and enhancing the personalized immersive experience of the game's storyline. Attached Figure Description
[0027] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application.
[0028] Figure 2 This is a schematic diagram of the system structure of this application.
[0029] Figure 3This is a structural block diagram of the computer device described in this application. Detailed Implementation
[0030] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0031] This application discloses a method for generating game parameters.
[0032] Reference Figure 1 Game parameter generation methods include: Step S11: Obtain multidimensional behavioral data of the target player during the game, and perform correlation mining on the multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data.
[0033] It should be noted that in step S11, the multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. These data reflect the player's real-time emotional changes during the game from different dimensions. Since single-type data is often difficult to comprehensively and accurately depict emotional states, this step uses association mining technology to fuse multi-source heterogeneous data, explore the intrinsic relationships between different modalities, and generate a quantitative feature sequence that can comprehensively characterize the player's real-time emotional experience level, providing accurate emotional input for subsequent personalized modeling.
[0034] Step S12: Based on the emotional state feature sequence, construct a personalized emotional response model for the target player, and generate a target emotional state feature trajectory for the target player according to the preset emotional change template and the personalized emotional response model. The target emotional state feature trajectory is used to characterize the relative emotional change trend that the target player is expected to achieve in the subsequent game process.
[0035] It should be noted that in step S13, due to the differences in emotional response patterns among players (e.g., different sensitivities and baseline levels to the same stimuli), this step first utilizes the player's own emotional state feature sequence to construct a personalized emotional response model that describes the player's emotional change patterns. Simultaneously, the game designers pre-define a standardized emotional change template associated with the current game story segment; this template represents an idealized emotional fluctuation rhythm. The emotional change template is input into the personalized emotional response model for adaptation and transformation, generating a target emotional state feature trajectory that retains the intended fluctuation pattern while conforming to the player's emotional response characteristics. This trajectory serves as the target for subsequent parameter optimization, guiding the player to achieve the desired emotional experience in subsequent game progression.
[0036] Step S13: Obtain historical game interaction data and construct an emotional state prediction model based on the historical game interaction data. The historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences.
[0037] It should be noted that in step S13, the role of the emotional state prediction model is to learn the mapping relationship between game plot parameters and player emotional state. That is, given a set of plot parameters (such as dialogue emotion, character relationship, speech rate, etc.), the model can predict the emotional state feature sequence that the player may generate under the influence of these parameters. By using a large amount of historical game data to train the model, the model can be equipped with the ability to accurately evaluate the emotional effects caused by any candidate plot parameters, providing an evaluation basis for subsequent optimization search.
[0038] Step S14: Invoke the emotional state prediction model and search the game plot parameter space based on the Bayesian algorithm to generate a set of target game plot parameters so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory.
[0039] It should be noted that in step S14, the game plot parameter space refers to a multi-dimensional space composed of all adjustable plot parameters (such as emotional tone, character relationship tags, dialogue speed, etc.) and their value ranges. Each point in the space represents a specific combination of parameters. In practical applications, this parameter space is usually high-dimensional and non-convex, making direct exhaustive search or random search inefficient. This step employs a Bayesian optimization algorithm to perform an efficient global search in the parameter space, gradually approaching the optimal parameter combination by balancing exploration and utilization. The final target game plot parameters ensure that the predicted emotional state sequence under the influence of these parameters matches the target trajectory generated in step S12, thereby achieving the technical effect of proactively guiding the player's emotional experience to the expected goal based on the player's personalized emotional response patterns.
[0040] In the above implementation, multidimensional behavioral data of the target player during the game is acquired, and correlation mining is performed on the multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. Then, based on the emotional state feature sequence, a personalized emotional response model for the target player is constructed. According to the preset emotional change template and the personalized emotional response model, a target emotional state feature trajectory for the target player is generated. The target emotional state feature trajectory is used to characterize the relative emotional change trend that the target player is expected to achieve in the subsequent game process. Then, historical game interaction data is acquired, and an emotional state prediction model is constructed based on the historical game interaction data. The historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. Then, the emotional state prediction model is called, and a set of target game plot parameters is generated by searching in the game plot parameter space based on the Bayesian algorithm, so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory. This invention achieves accurate perception of players' real-time emotional states through correlation mining of multi-dimensional behavioral data; it adapts to the unique emotional baselines and response amplitudes of different players by constructing a personalized emotional response model; it generates personalized target trajectories based on preset emotional change templates, and combines emotional state prediction models with Bayesian optimization algorithms to achieve proactive and refined control of micro-narrative parameters such as emotional tone, character relationships, and speech rate, thereby accurately conveying the designer's emotional intentions to each player's actual experience and enhancing the personalized immersive experience of the game's storyline.
[0041] As a further implementation of the method, the step of performing correlation mining on multidimensional behavioral data to obtain a sequence of emotional state features includes: Step S21: Preprocess and extract features for each type of data in the multidimensional behavioral data to obtain the corresponding single-modal feature vector.
[0042] Step S22: Align each single-modal feature vector along the time dimension to obtain a multimodal feature matrix.
[0043] Step S23: Input the multimodal feature matrix into the cross-modal attention network, calculate the correlation weights between different modal features through the self-attention mechanism, and perform weighted fusion of the multimodal feature matrix based on the correlation weights to obtain the fused feature vector sequence.
[0044] Step S24: Input the fused feature vector sequence into the temporal coding network, and model the time-dimensional dependency through a recurrent neural network or Transformer to output the emotional state feature sequence.
[0045] It should be noted that from steps S21 to S24, through multimodal feature extraction, alignment, fusion, and temporal modeling of multidimensional behavioral data, effective mining of emotional state feature sequences from raw behavioral data is achieved. Specifically, firstly, various types of behavioral data are preprocessed and feature extracted to obtain corresponding single-modal feature vectors; then, a unified multimodal feature matrix is constructed through temporal alignment; next, the multimodal feature matrix is input into a cross-modal attention network, and the correlation weights between different modal features are calculated using a self-attention mechanism and then weighted and fused to obtain a fused feature vector sequence; finally, the fused feature sequence is input into a temporal coding network (such as a recurrent neural network or a Transformer) to model the dependencies in the temporal dimension, outputting an emotional state feature sequence that comprehensively reflects the dynamic changes in the player's emotions. This process fully integrates multi-source heterogeneous data, enhancing the accuracy and robustness of emotional state estimation.
[0046] In the above implementation, in order to realize the association mining of multi-dimensional behavioral data, each type of data in the multi-dimensional behavioral data is preprocessed and feature extracted to obtain the corresponding single-modal feature vector. Then, the single-modal feature vectors are aligned in the time dimension to obtain a multi-modal feature matrix. The multi-modal feature matrix is then input into a cross-modal attention network. The association weights between different modal features are calculated through a self-attention mechanism. The multi-modal feature matrix is then weighted and fused based on the association weights to obtain a fused feature vector sequence. The fused feature vector sequence is then input into a temporal coding network. The dependencies in the time dimension are modeled through a recurrent neural network or a Transformer to output an emotional state feature sequence.
[0047] As a further implementation of the method, the steps of inputting the multimodal feature matrix into a cross-modal attention network, calculating the correlation weights between different modal features through a self-attention mechanism, and weighting and fusing the multimodal feature matrix based on the correlation weights to obtain a fused feature vector sequence include: Step S31: Map the multimodal feature matrix into a query matrix, a key matrix, and a value matrix, respectively.
[0048] Step S32: Calculate the similarity between the query matrix and the key matrix to obtain the attention weight matrix.
[0049] Step S33: Normalize the attention weight matrix to obtain the normalized weight matrix.
[0050] Step S34: Based on the normalized weight matrix, the value matrix is weighted and summed to obtain the single-head attention output features.
[0051] Step S35: Concatenate the single-head attention output features corresponding to multiple different attention heads, and reduce the dimensionality through a linear transformation layer to obtain a fused feature vector sequence.
[0052] It should be noted that from steps S31 to S35, by introducing a multi-head self-attention mechanism in the cross-modal attention network, dynamic modeling and adaptive fusion of deep correlations between features of different modalities are achieved. Specifically, firstly, the multimodal feature matrix is mapped to a query matrix, a key matrix, and a value matrix through a linear transformation. Then, the similarity between the query matrix and the key matrix is calculated to obtain the attention weight matrix, which quantifies the degree of mutual influence between features of different modalities. Next, the attention weight matrix is normalized to ensure the stability and interpretability of the weights. Then, the value matrix is weighted and summed based on the normalized weight matrix to generate a single-head attention output feature that integrates cross-modal information. Finally, the output features of multiple attention heads are concatenated and dimensionality is reduced through a linear transformation layer to obtain the final fused feature vector sequence. This process not only effectively captures the complex interaction relationships between multimodal data but also enhances the expressive power and robustness of the model through the multi-head mechanism, providing high-quality feature input for subsequent temporal modeling of sentiment states.
[0053] In the above implementation, in order to generate the fused feature vector sequence, the multimodal feature matrix is mapped to a query matrix, a key matrix, and a value matrix, respectively. Then, the similarity between the query matrix and the key matrix is calculated to obtain the attention weight matrix. The attention weight matrix is then normalized to obtain a normalized weight matrix. Based on the normalized weight matrix, the value matrix is weighted and summed to obtain the single-head attention output features. Then, the single-head attention output features corresponding to multiple different attention heads are concatenated and dimensionality is reduced through a linear transformation layer to obtain the fused feature vector sequence.
[0054] As a further implementation of the method, the step of constructing a personalized emotional response model for the target player based on the emotional state feature sequence includes: Step S41: Obtain the emotional state feature sequence of the target player within a preset time period before the current moment.
[0055] Step S42: Perform statistical analysis on the emotional state feature sequence to extract the target player's emotional baseline features and emotional response amplitude features. The emotional baseline features are used to characterize the target player's baseline emotional state level under conditions of no external stimuli or low-intensity stimuli, while the emotional response amplitude features are used to characterize the fluctuation intensity of the target player's emotional state when facing game plot stimuli.
[0056] Step S43: Based on the emotional baseline characteristics and emotional response amplitude characteristics, construct a personalized emotional response model for the target player. The personalized emotional response model is used to describe the changing pattern of the target player's emotional state with the stimulation of the game's plot.
[0057] It should be noted that from steps S41 to S43, through statistical analysis of the target player's recent emotional state feature sequence, personalized emotional baseline features and emotional response amplitude features are extracted, thereby constructing a personalized emotional response model that can characterize the dynamic changes in the player's emotions. Specifically, firstly, the emotional state feature sequence within a preset time period before the current moment is obtained to ensure that the data used reflects the player's recent stable emotional behavior pattern; then, statistical analysis is performed on this sequence to extract emotional baseline features (representing the player's baseline emotional level under no or low-intensity stimuli) and emotional response amplitude features (representing the intensity of the player's emotional fluctuations when facing game plot stimuli). These two features together constitute the core personalized parameters of the player's emotional response; finally, a personalized emotional response model is constructed based on the extracted features. This model can accurately describe how the target player's emotional state evolves with changes in game plot stimuli, providing support for subsequently generating a target emotional trajectory that meets the player's personalized emotional needs.
[0058] In the above implementation, in order to construct a personalized emotional response model, the emotional state feature sequence of the target player within a preset time period before the current moment is obtained. Then, the emotional state feature sequence is statistically analyzed to extract the target player's emotional baseline feature and emotional response amplitude feature. The emotional baseline feature is used to characterize the target player's baseline emotional state level under no external stimuli or low-intensity stimuli, and the emotional response amplitude feature is used to characterize the fluctuation intensity of the target player's emotional state when facing game plot stimuli. Then, based on the emotional baseline feature and emotional response amplitude feature, a personalized emotional response model of the target player is constructed. The personalized emotional response model is used to describe the change pattern of the target player's emotional state with game plot stimuli.
[0059] As a further implementation of the method, the step of generating a target emotional state feature trajectory for the target player based on a preset emotional change template and a personalized emotional response model includes: Step S51: Obtain a preset emotion change template, wherein the emotion change template is a standardized emotion fluctuation curve associated with the current game plot segment, and the emotion change template is used to characterize the change in the intensity of the emotion that the player is expected to experience in the current game plot segment over time.
[0060] Step S52: Based on the personalized emotion response model, the emotion change template is mapped and transformed to obtain the emotion state feature transformation sequence.
[0061] Step S53: Based on the emotional state feature transformation sequence, generate a target emotional state feature trajectory for the target player. The target emotional state feature trajectory retains the fluctuation pattern of the emotional change template, and the corresponding numerical range and fluctuation amplitude match the target player's emotional baseline features and emotional response amplitude features.
[0062] It should be noted that from steps S51 to S53, by combining the preset standardized emotional change template with the player's personalized emotional response model, a precise conversion from general emotional design to personalized emotional target trajectory is achieved. Specifically, firstly, an emotional change template associated with the current game plot segment is obtained. This template, as a standardized emotional fluctuation curve, defines the basic form of the expected change in the player's emotional intensity over time during the plot, reflecting the game designer's emotional guidance intention. Then, based on the personalized emotional response model constructed in steps S41 to S43, the template is mapped and transformed. That is, according to the player's emotional baseline characteristics and response amplitude characteristics, the emotional intensity values in the template are adjusted to a suitable range and fluctuation amplitude for the player's actual values, thereby generating an emotional state feature transformation sequence. Finally, based on this transformation sequence, a target emotional state feature trajectory for the target player is formed. This trajectory retains the fluctuation pattern of the original template (i.e., the key rhythm and turning points of emotional change) and matches the player's personalized characteristics in terms of numerical level and fluctuation intensity. This ensures that the game's emotional experience not only follows the original design intention but also matches the player's real emotional reaction pattern, providing a clear emotional guidance target for subsequent plot parameter optimization.
[0063] In the above implementation, in order to generate the target emotional state feature trajectory, a preset emotional change template is obtained. The emotional change template is a standardized emotional fluctuation curve associated with the current game plot segment. The emotional change template is used to characterize the change in the intensity of the emotion that the player is expected to experience in the current game plot segment over time. Then, based on the personalized emotional response model, the emotional change template is mapped and transformed to obtain an emotional state feature transformation sequence. Then, based on the emotional state feature transformation sequence, a target emotional state feature trajectory for the target player is generated. The target emotional state feature trajectory retains the fluctuation pattern of the emotional change template, and the corresponding numerical range and fluctuation amplitude match the target player's emotional baseline characteristics and emotional response amplitude characteristics.
[0064] As a further implementation of the method, the steps of acquiring historical game interaction data and constructing an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences, include: Step S61: Obtain historical game interaction data and preprocess the historical game interaction data to obtain a training sample set. The historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences.
[0065] Step S62: Construct an initial emotional state prediction model, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive historical game plot parameters, the hidden layer is used to learn the mapping relationship between historical game plot parameters and historical emotional state feature sequences, and the output layer is used to output the predicted emotional state feature sequence.
[0066] Step S63: Divide the training sample set into a training set and a validation set, and iteratively train the initial sentiment state prediction model using the training set and validate the initial sentiment state prediction model using the validation set.
[0067] Step S64: When the prediction error of the initial sentiment state prediction model on the validation set meets the preset accuracy requirement, training is stopped, and the initial sentiment state prediction model obtained from the training is used as the sentiment state prediction model.
[0068] In the above implementation, in order to construct the emotion state prediction model, historical game interaction data is acquired and preprocessed to obtain a training sample set. The historical game interaction data includes historical game plot parameters and corresponding historical emotion state feature sequences. Then, an initial emotion state prediction model is constructed, which includes an input layer, a hidden layer, and an output layer. The input layer is used to receive historical game plot parameters, the hidden layer is used to learn the mapping relationship between historical game plot parameters and historical emotion state feature sequences, and the output layer is used to output the predicted emotion state feature sequences. The training sample set is then divided into a training set and a validation set. The initial emotion state prediction model is iteratively trained using the training set and validated using the validation set. When the prediction error of the initial emotion state prediction model on the validation set meets the preset accuracy requirement, training is stopped, and the trained initial emotion state prediction model is used as the emotion state prediction model.
[0069] As a further implementation of the method, the step of invoking an emotion state prediction model and searching the game plot parameter space based on a Bayesian algorithm to generate a set of target game plot parameters, so that the emotion state feature prediction sequence corresponding to the target game plot parameters matches the target emotion state feature trajectory, includes: Step S71: Initialize the Gaussian process surrogate model and define the Bayesian optimized acquisition function.
[0070] Step S72: Randomly sample several parameter points in the game plot parameter space, input each parameter point into the emotional state prediction model to obtain the corresponding emotional state feature prediction sequence, and calculate the matching degree between the emotional state feature prediction sequence and the target emotional state feature trajectory, and use the matching degree as the objective function value of each parameter point.
[0071] Step S73: If there are parameter points with a matching degree less than a preset threshold, then the parameter point is used as a set of target game plot parameters.
[0072] Step S74: If there are no parameter points with a matching degree less than a preset threshold, then each parameter point and its corresponding objective function value are used as observation data to update the Gaussian process surrogate model.
[0073] Step S75: Optimize the acquisition function based on the updated Gaussian process surrogate model, determine the next parameter point to be evaluated in the game plot parameter space, input the parameter point into the emotion state prediction model to obtain a new objective function value, and iteratively update the Gaussian process surrogate model based on the parameter point and the objective function value.
[0074] Step S76: When the preset iteration termination condition is met, the parameter point with the highest matching degree in the observation data is taken as a set of target game plot parameters.
[0075] It should be noted that from step S71 to step S76, by introducing the Bayesian optimization algorithm, the optimal combination of parameters that can match the predicted emotional sequence with the target emotional trajectory is efficiently searched in the game plot parameter space. Specifically, the Gaussian process surrogate model is first initialized and a data acquisition function is defined. Then, several initial parameter points are randomly sampled in the parameter space. The corresponding emotional state feature prediction sequence is obtained through the emotional state prediction model, and the matching degree between the prediction sequence and the target emotional state feature trajectory is calculated as the objective function value. If there are parameter points that meet the preset matching degree threshold, they are directly output as target game plot parameters. If not, the currently evaluated parameter points and their objective function values are used as observation data to update the Gaussian process surrogate model, making it more accurately approximate the true objective function distribution. Next, the data acquisition function is optimized based on the updated surrogate model, achieving a balance between exploring unknown regions and utilizing the current optimal region. The next parameter point most likely to improve the matching degree is determined and input into the emotional state prediction model to obtain a new objective function value. This process is repeated iteratively to continuously expand the observation data and optimize the surrogate model. When the preset iteration termination condition (such as the maximum number of iterations or the convergence threshold) is reached, the parameter point with the highest matching degree is selected from all evaluated observation data as the final set of target game plot parameters. This process achieves efficient global optimization of high-dimensional, nonlinear parameter space, ensuring that the generated plot parameters can accurately guide the player's emotional state to develop along the desired target trajectory.
[0076] In the above implementation, to achieve the search for game plot parameters, a Gaussian process proxy model is initialized, and a Bayesian optimized acquisition function is defined. Then, several parameter points are randomly sampled in the game plot parameter space, and each parameter point is input into the emotion state prediction model to obtain the corresponding emotion state feature prediction sequence. The matching degree between the emotion state feature prediction sequence and the target emotion state feature trajectory is calculated, and the matching degree is used as the objective function value of each parameter point. If there is a parameter point with a matching degree less than a preset threshold, then the parameter point is used as a set of target game plot parameters. If there is no parameter point with a matching degree less than the preset threshold, then each parameter point and the corresponding objective function value are used as observation data to update the Gaussian process proxy model. Then, based on the updated Gaussian process proxy model, the acquisition function is optimized, and the next parameter point to be evaluated is determined in the game plot parameter space. This parameter point is input into the emotion state prediction model to obtain a new objective function value, and the Gaussian process proxy model is iteratively updated based on the parameter point and the objective function value. When the preset iteration termination condition is met, the parameter point with the highest matching degree in the observation data is used as a set of target game plot parameters.
[0077] This application also discloses a game parameter generation system.
[0078] refer to Figure 2 The game parameter generation system includes: The association mining module is used to acquire multidimensional behavioral data of the target player during the game, and to perform association mining on the multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. The personalized emotion modeling module is used to construct a personalized emotion response model for the target player based on the emotion state feature sequence, and generate a target emotion state feature trajectory for the target player according to the preset emotion change template and the personalized emotion response model. The target emotion state feature trajectory is used to represent the relative emotion change trend that the target player is expected to achieve in the subsequent game process. The model building module is used to acquire historical game interaction data and build an emotional state prediction model based on the historical game interaction data. The historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. The plot parameter optimization module is used to call the emotion state prediction model and search in the game plot parameter space based on the Bayesian algorithm to generate a set of target game plot parameters so that the emotion state feature prediction sequence corresponding to the target game plot parameters matches the target emotion state feature trajectory.
[0079] The game parameter generation system of the present invention can implement any of the game parameter generation methods, and the specific working process of the game parameter generation system of the present invention can refer to the corresponding process in the above-mentioned game parameter generation methods.
[0080] This application also discloses a computer device.
[0081] refer to Figure 3 A computer device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement any of the above-described methods for generating game parameters.
[0082] This application also discloses a computer-readable storage medium.
[0083] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed by any of the above-described methods for generating game parameters.
[0084] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0085] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.
Claims
1. A game parameter generation method characterized by, include: The system acquires multidimensional behavioral data of the target player during the game and performs correlation mining on the multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. Based on the emotional state feature sequence, a personalized emotional response model is constructed for the target player, and a target emotional state feature trajectory is generated for the target player according to the preset emotional change template and the personalized emotional response model. The target emotional state feature trajectory is used to characterize the relative emotional change trend that the target player is expected to achieve in the subsequent game process. Acquire historical game interaction data and construct an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences; The emotional state prediction model is invoked, and a set of target game plot parameters is generated by searching in the game plot parameter space based on the Bayesian algorithm, so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory.
2. The game parameter generation method according to claim 1, characterized in that, The step of performing correlation mining on the multidimensional behavioral data to obtain the emotional state feature sequence includes: Each type of data in the multidimensional behavioral data is preprocessed and feature extracted to obtain the corresponding single-modal feature vector; Align each of the single-modal feature vectors along the time dimension to obtain a multimodal feature matrix; The multimodal feature matrix is input into a cross-modal attention network, and the correlation weights between different modal features are calculated through a self-attention mechanism. The multimodal feature matrix is then weighted and fused based on the correlation weights to obtain a fused feature vector sequence. The fused feature vector sequence is input into a temporal coding network, and the dependencies in the time dimension are modeled by a recurrent neural network or a Transformer to output an emotional state feature sequence.
3. The game parameter generation method according to claim 2, characterized in that, The step of inputting the multimodal feature matrix into a cross-modal attention network, calculating the correlation weights between different modal features through a self-attention mechanism, and weighting and fusing the multimodal feature matrix based on the correlation weights to obtain a fused feature vector sequence includes: The multimodal feature matrices are mapped to query matrices, key matrices, and value matrices, respectively. Calculate the similarity between the query matrix and the key matrix to obtain the attention weight matrix; The attention weight matrix is normalized to obtain a normalized weight matrix; Based on the normalized weight matrix, the value matrix is weighted and summed to obtain the single-head attention output features; The single-head attention output features corresponding to multiple different attention heads are concatenated and then dimensionality is reduced through a linear transformation layer to obtain a fused feature vector sequence.
4. The game parameter generation method according to claim 1, characterized in that, The step of constructing a personalized emotional response model for the target player based on the emotional state feature sequence includes: Obtain the emotional state feature sequence of the target player within a preset time period prior to the current moment; Statistical analysis is performed on the emotional state feature sequence to extract the emotional baseline feature and emotional response amplitude feature of the target player. The emotional baseline feature is used to characterize the baseline level of the target player's emotional state under no external stimulus or low intensity stimulus, and the emotional response amplitude feature is used to characterize the fluctuation intensity of the target player's emotional state when facing game plot stimuli. Based on the emotional baseline features and the emotional response amplitude features, a personalized emotional response model for the target player is constructed, wherein the personalized emotional response model is used to describe the changing pattern of the target player's emotional state with the stimulation of the game's plot.
5. The game parameter generation method according to claim 4, characterized in that, The step of generating a target emotional state feature trajectory for the target player based on a preset emotional change template and the personalized emotional response model includes: Obtain a preset emotion change template, wherein the emotion change template is a standardized emotion fluctuation curve associated with the current game plot segment, and the emotion change template is used to characterize the change in the intensity of the emotion that the player is expected to experience in the current game plot segment over time. Based on the personalized emotion response model, the emotion change template is mapped and transformed to obtain an emotion state feature transformation sequence; Based on the emotional state feature transformation sequence, a target emotional state feature trajectory is generated for the target player. The target emotional state feature trajectory retains the fluctuation pattern of the emotional change template, and the corresponding numerical range and fluctuation amplitude match the target player's emotional baseline feature and emotional response amplitude feature.
6. The game parameter generation method according to claim 1, characterized in that, The step of acquiring historical game interaction data and constructing an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences, includes: Historical game interaction data is acquired and preprocessed to obtain a training sample set, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. An initial emotional state prediction model is constructed, comprising an input layer, a hidden layer, and an output layer. The input layer is used to receive the historical game plot parameters, the hidden layer is used to learn the mapping relationship between the historical game plot parameters and the historical emotional state feature sequence, and the output layer is used to output the predicted emotional state feature sequence. The training sample set is divided into a training set and a validation set. The initial sentiment state prediction model is iteratively trained using the training set and validated using the validation set. When the prediction error of the initial sentiment state prediction model on the validation set meets the preset accuracy requirement, training is stopped, and the trained initial sentiment state prediction model is used as the sentiment state prediction model.
7. The game parameter generation method according to claim 1, characterized in that, The step of calling the emotional state prediction model and searching in the game plot parameter space based on the Bayesian algorithm to generate a set of target game plot parameters, so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory, includes: Initialize the Gaussian process proxy model and define the Bayesian optimized acquisition function; Several parameter points are randomly sampled in the game plot parameter space. Each parameter point is input into the emotional state prediction model to obtain the corresponding emotional state feature prediction sequence. The matching degree between the emotional state feature prediction sequence and the target emotional state feature trajectory is calculated. The matching degree is used as the objective function value of each parameter point. If there is a parameter point whose matching degree is less than a preset threshold, then that parameter point is used as the set of target game plot parameters; If there are no parameter points with a matching degree less than a preset threshold, then each parameter point and its corresponding objective function value are used as observation data to update the Gaussian process proxy model. The acquisition function is optimized based on the updated Gaussian process proxy model. The next parameter point to be evaluated is determined in the game plot parameter space. The parameter point is input into the emotion state prediction model to obtain a new objective function value. The Gaussian process proxy model is iteratively updated based on the parameter point and the objective function value. When the preset iteration termination condition is met, the parameter point with the highest matching degree in the observed data is taken as a set of target game plot parameters.
8. A game parameter generation system, characterized in that, include: The association mining module is used to acquire multidimensional behavioral data of the target player during the game, and to perform association mining on the multidimensional behavioral data to obtain an emotional state feature sequence. The multidimensional behavioral data includes at least two of the following: physiological signal data, operational behavior data, facial expression data, and voice data. The personalized emotion modeling module is used to construct a personalized emotion response model for the target player based on the emotion state feature sequence, and generate a target emotion state feature trajectory for the target player according to a preset emotion change template and the personalized emotion response model. The target emotion state feature trajectory is used to characterize the relative emotion change trend that the target player is expected to achieve in the subsequent game process. The model building module is used to acquire historical game interaction data and build an emotional state prediction model based on the historical game interaction data, wherein the historical game interaction data includes historical game plot parameters and corresponding historical emotional state feature sequences. The plot parameter optimization module is used to call the emotional state prediction model and search in the game plot parameter space based on the Bayesian algorithm to generate a set of target game plot parameters so that the emotional state feature prediction sequence corresponding to the target game plot parameters matches the target emotional state feature trajectory.
9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and execute the method of any one of claims 1 to 7.