Dynamic game difficulty self-adaptive adjustment method and system based on user behavior feedback

Abstract

The application provides a dynamic game difficulty self-adaptive adjustment method and system based on user behavior feedback, and relates to the technical field of game design. The method comprises the following steps: performing dynamic weighted calculation based on the difficulty adaptation index, combining real-time physiological feedback data of a user to generate a dynamic difficulty correction coefficient, wherein the dynamic difficulty correction coefficient comprises a scene complexity adjustment parameter and an interactive response threshold adjustment parameter; inputting the user feedback parameter set into an image feature weight distributor, dynamically adjusting the weight distribution of an image retrieval feature vector according to user attention distribution data and operation delay parameters, and generating an optimized scene element retrieval strategy; performing feature space mapping on the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy, and generating a final game difficulty control instruction set through nonlinear superposition. The application can optimize game design and enhance the adaptability and flexibility of games.

Description

Technical Field

[0001] This invention relates to the field of game design technology, and in particular to a method and system for dynamically adjusting game difficulty based on user behavior feedback. Background Technology

[0002] With the rapid development of the video game industry, players' demands for personalized gaming experiences are increasing. Traditional game difficulty adjustment mechanisms mostly adopt static settings or simple linear adjustments, usually based on single-dimensional data such as player completion time and number of failures to determine difficulty.

[0003] This type of method has some drawbacks, such as:

[0004] First, static threshold settings cannot adapt to the dynamic changes in the operational abilities of different players, which can easily lead to an imbalance in difficulty. Second, the lack of real-time perception of the player's physiological state (such as concentration and stress level) makes it difficult to capture the user's real experience load. Third, the correlation between game scene elements and operation feedback is not effectively utilized, resulting in a lack of coordination between difficulty adjustment and interface information presentation. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and system for dynamic adaptive adjustment of game difficulty based on user behavior feedback, which can improve user experience, optimize game design, and enhance game adaptability and flexibility.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] Firstly, a dynamic game difficulty adaptive adjustment method based on user behavior feedback, the method comprising:

[0008] Step S1: During the game interaction process, collect multi-dimensional behavioral data streams of users in real time, and extract features from the behavioral data streams to obtain game performance image feature sets and user feedback parameter sets;

[0009] Step S2: Based on the game performance image feature set, a difficulty adaptation index is generated by calculating the matching degree between the user operation trajectory and the preset benchmark mode. The difficulty adaptation index represents the matching deviation between the current game difficulty level and the user's ability level.

[0010] Step S3: Perform dynamic weighted calculation based on the difficulty adaptation index, and generate a dynamic difficulty correction coefficient by combining real-time physiological feedback data of the user. The dynamic difficulty correction coefficient includes scene complexity adjustment parameters and interaction response threshold adjustment parameters.

[0011] Step S4: Input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight allocation of the image retrieval feature vector according to the user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy.

[0012] Step S5: Establish a multi-dimensional parameter fusion device, map the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy in the feature space, and generate the final game difficulty control instruction set through nonlinear superposition.

[0013] Step S6: Receive real-time user feedback on the adjusted game difficulty, and iteratively optimize through backpropagation algorithm to form a dynamically adaptive difficulty adjustment loop.

[0014] Secondly, a dynamic game difficulty adaptive adjustment system based on user behavior feedback includes:

[0015] The acquisition module is used to collect multi-dimensional behavioral data streams of users in real time during game interaction, and to extract features from the behavioral data streams to obtain game performance image feature sets and user feedback parameter sets.

[0016] The calculation module is used to generate a difficulty adaptation index by calculating the matching degree between the user's operation trajectory and the preset benchmark mode based on the game performance image feature set. The difficulty adaptation index represents the matching deviation between the current game difficulty level and the user's ability level. Based on the difficulty adaptation index, dynamic weighted calculation is performed, and combined with the user's real-time physiological feedback data, a dynamic difficulty correction coefficient is generated. The dynamic difficulty correction coefficient includes scene complexity adjustment parameters and interaction response threshold adjustment parameters.

[0017] The fusion module is used to input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight allocation of the image retrieval feature vector according to the user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy; and establish a multi-dimensional parameter fusion module to map the dynamic difficulty correction coefficient with the optimized scene element retrieval strategy in the feature space, and generate the final game difficulty control instruction set through nonlinear superposition.

[0018] The adjustment module is used to receive user feedback on the adjusted game difficulty in real time, and iterates through the backpropagation algorithm to form a dynamic and adaptive difficulty adjustment loop.

[0019] Thirdly, a computing device, comprising:

[0020] One or more processors;

[0021] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0022] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0023] The above-described solution of the present invention has at least the following beneficial effects:

[0024] By integrating user operation behavior, physiological feedback, and visual attention distribution data, this approach breaks through the limitations of traditional single-dimensional adjustments, achieving a dynamic balance between user cognitive load and game difficulty, and solving the problem of fragmented experience caused by player operation delays or information overload.

[0025] Based on the matching degree calculation between the user's operation trajectory and the preset benchmark mode, combined with the real-time feedback of physiological stress indicators, a multi-dimensional linkage difficulty correction coefficient is generated, which reduces the difficulty adjustment error and reduces the frustration of players caused by sudden changes in difficulty.

[0026] By dynamically optimizing the scene element retrieval strategy through an image feature weight allocator, and combining user attention heatmaps and operation latency sensitivity, the interface information density and interaction focus are automatically adjusted, thereby improving target recognition efficiency and reducing user error rate.

[0027] By employing feature space mapping and nonlinear superposition techniques, deep coupling of scene complexity parameters, interaction fault tolerance thresholds, and rendering commands is achieved, enabling dynamic elements of the game environment (such as enemy AI behavior and resource refresh logic) to match user operation capabilities in real time, thereby enhancing the game's immersive experience.

[0028] By continuously optimizing the difficulty assessment model and feature weight allocator through backpropagation algorithm and combining cross-modal alignment of user experience evaluation data, a dynamically evolving adjustment loop is formed, which improves the adaptive iteration speed and keeps the difficulty curve synchronized with the player's growth in the long term.

[0029] By utilizing a sliding window mechanism and temporal alignment technology to process multi-source heterogeneous data, computational redundancy in the feature extraction process is reduced, thus lowering the GPU load compared to traditional methods and ensuring stable operation of real-time adjustments on mainstream gaming hardware platforms. Attached Figure Description

[0030] Figure 1 This is a flowchart illustrating the dynamic game difficulty adaptive adjustment method based on user behavior feedback provided in an embodiment of the present invention.

[0031] Figure 2This is a schematic diagram of a dynamic game difficulty adaptive adjustment system based on user behavior feedback provided in an embodiment of the present invention. Detailed Implementation

[0032] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0033] like Figure 1 As shown, embodiments of the present invention propose a dynamic game difficulty adaptive adjustment method based on user behavior feedback, the method comprising the following steps:

[0034] Step S1: During the game interaction process, collect multi-dimensional behavioral data streams of users in real time, and extract features from the behavioral data streams to obtain game performance image feature sets and user feedback parameter sets;

[0035] Step S2: Based on the game performance image feature set, a difficulty adaptation index is generated by calculating the matching degree between the user operation trajectory and the preset benchmark mode.

[0036] Step S3: Perform dynamic weighted calculation based on the difficulty adaptation index, and generate a dynamic difficulty correction coefficient by combining the user's real-time physiological feedback data;

[0037] Step S4: Input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight allocation of the image retrieval feature vector according to the user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy.

[0038] Step S5: Map the dynamic difficulty correction coefficient to the optimized scene element retrieval strategy in the feature space, and generate the final game difficulty control instruction set through nonlinear superposition.

[0039] Step S6: Receive real-time user feedback on the adjusted game difficulty, and iteratively optimize through backpropagation algorithm to form a dynamically adaptive difficulty adjustment loop.

[0040] In this embodiment of the invention, by analyzing the matching degree between the player's operation trajectory and the preset benchmark mode in real time (step S2), a personalized difficulty adaptation index is generated, so that the game difficulty is dynamically matched with the player's current ability, avoiding frustration due to excessive difficulty or boredom due to excessive difficulty; by combining real-time physiological data such as heart rate and eye movement (step S3), a dynamic difficulty correction coefficient is generated to accurately capture the player's emotional fluctuations (such as tension and fatigue), further refining the difficulty adjustment dimensions and enhancing immersion; based on the player's attention distribution and operation delay (step S4), the feature weights of the scene element retrieval strategy are dynamically adjusted, such as highlighting the key targets that the player is concerned about or simplifying interfering elements, guiding the player to complete tasks more efficiently and improving the smoothness of operation; through feature space mapping and nonlinear superposition (step S5), the difficulty correction and scene element retrieval strategy are deeply integrated to achieve coordinated adjustment of multi-dimensional parameters (such as enemy strength and environmental complexity), making the difficulty change more natural and delicate; by using player experience evaluation data (step S6), the model parameters are iteratively optimized through the backpropagation algorithm to form a self-evolving difficulty adjustment loop. As player behavior data accumulates, the system can continuously "learn" player preferences, maintain the accuracy of difficulty matching over the long term, reduce the workload of manually pre-setting difficulty levels, and cover a wider range of players (from beginners to experts) through automation mechanisms, thereby improving the game's universality and lifecycle value. By integrating multi-source data such as operation trajectories, physiological signals, and attention distribution (steps S1 to S4), the system avoids the one-sidedness of single-dimensional judgments, making the basis for difficulty adjustment more comprehensive and reliable. Real-time calculation throughout the entire process (from data collection to command generation) ensures that difficulty adjustment is synchronized with player behavior, reducing the sense of lag and enhancing the immediacy and realism of the interaction.

[0041] In a preferred embodiment of the present invention, step S1 involves real-time acquisition of multi-dimensional behavioral data streams from the user during game interaction, and feature extraction of the behavioral data streams to obtain a game performance image feature set and a user feedback parameter set, including:

[0042] By embedding data points in the game client, user operation behavior sequence data is collected, including operation frequency, reaction time, task completion path trajectory and error triggering events;

[0043] The game performance image data stream is captured through the graphics processor interface, and multi-scale feature extraction is performed on continuous game image frames to generate the game performance image feature set, which includes the distribution density of interface elements, character motion vectors, and scene complexity indicators.

[0044] User feedback data is collected synchronously through biosensors and an external interactive platform, including eye-tracking hotspot distribution, skin conductivity change curves, and subjective difficulty score input. The user feedback parameter set is generated by combining in-game operation latency parameters.

[0045] The feature extraction process employs a sliding window mechanism to align multi-source heterogeneous data in the temporal domain, and uses a convolutional neural network to perform dynamic object detection and behavioral intent classification on image frames, forming a feature matrix with synchronized timestamps.

[0046] In this embodiment of the invention, the number of user operations per unit time (e.g., clicks per minute) is recorded in real time through preset tracking points (such as button clicks, keyboard inputs, and other interaction nodes) in the game client; the time interval between receiving game stimuli (e.g., enemy appearance, task prompts) and performing the corresponding operation (e.g., the time taken from seeing an enemy to clicking to attack) is tracked; based on the game map coordinate system, the movement path coordinate sequence of the user-controlled character is recorded (e.g., the movement trajectory curve from point A to point B); events that trigger invalid operations or task failures (e.g., skill release errors, number of times colliding with obstacles) are identified and recorded, quantifying the user's operational efficiency and proficiency, identifying operational bottlenecks (e.g., high-frequency error areas), and providing basic behavioral data for difficulty assessment, such as reducing the operational complexity for players with slow reactions.

[0047] By capturing continuous frames of the game screen in real time (e.g., 30 frames per second) through the GPU interface, the raw pixel data is extracted. The distribution density of elements such as buttons and prompts in the interface is statistically analyzed at different resolutions (e.g., global and local zoom). The changes in character position in adjacent frames are compared to generate motion direction and velocity vectors (e.g., the speed of a character moving to the left is 5 pixels / frame). The visual complexity of the scene is quantified by indicators such as the number of objects and the frequency of light and shadow changes (e.g., the complexity of a battle scene is higher than that of a dialogue scene). The cognitive load of the game's visual information on the user is analyzed. For example, complex scenes may cause attention to be distracted. This provides a data basis for dynamically adjusting scene elements (e.g., simplifying the interface and reducing the movement speed of enemies).

[0048] This invention records the user's gaze on the screen using an eye tracker (e.g., continuously staring at a certain area for more than 200 milliseconds is marked as a hotspot), collects changes in skin microcurrents in real time using biosensors to map the user's level of tension or excitement (e.g., a sudden increase in conductivity indicates a stress response), collects the user's rating of the current difficulty through in-game pop-ups or external devices (e.g., 1-5 star difficulty feedback), calculates the time difference between the user's operation command (e.g., button press) and the game screen response (e.g., skill release delay), and correlates physiological signals, subjective ratings, and operation delays by timestamp to generate comprehensive feedback parameters (e.g., high conductivity + long delay + low rating indicates excessive difficulty). This invention also captures the user's emotional fluctuations in real time (e.g., automatically reducing enemy strength when tense), improving the user experience's refinement.

[0049] This invention segments multi-source data, including operation data, image frames, and physiological signals, into fixed time windows (e.g., 1 second). Timestamp calibration ensures that data within the same window corresponds to the same game moment. It identifies key elements (e.g., enemies, items) and their locations in image frames. Based on the spatiotemporal correlation between user operation trajectories and image elements, it infers operation intentions (e.g., moving towards an enemy may indicate an attack intention). It arranges features of various dimensions (e.g., operation frequency, eye-tracking hotspot coordinates, scene complexity) in chronological order to form a two-dimensional feature matrix with timestamps. This invention eliminates temporal misalignment of multi-source data, ensures consistency of subsequent model inputs, and automatically extracts semantic features of images through deep learning, reducing the cost of manual feature engineering and improving the accuracy of intent recognition.

[0050] In a preferred embodiment of the present invention, step S2 involves generating a difficulty adaptation index based on the game performance image feature set by calculating the matching degree between the user's operation trajectory and a preset benchmark mode. The difficulty adaptation index characterizes the matching deviation between the current game difficulty level and the user's ability level, including:

[0051] The behavioral data in the game performance image feature set and user feedback parameter set are standardized and preprocessed to extract user ability-related features, including the player's average reaction time, operation accuracy, task completion time deviation and task success rate.

[0052] Based on the preset benchmark reaction time and ideal task completion time, the first normalized difference between the average reaction time and the benchmark reaction time, and the second normalized difference between the deviation of the task completion time and the ideal task completion time are calculated respectively.

[0053] The operational accuracy rate and task success rate are linearly weighted and fused to generate an operational performance evaluation component.

[0054] A dynamic weight allocation strategy is used to couple the first normalized difference, the second normalized difference, and the operational efficiency evaluation component in multiple dimensions to generate a comprehensive difficulty matching index, wherein the weight coefficients are adaptively adjusted based on the user's current physiological feedback data.

[0055] The comprehensive difficulty matching index is mapped to a preset difficulty level threshold range. If the index is lower than the first threshold, a difficulty downgrade instruction is triggered; if it is higher than the second threshold, a difficulty upgrade instruction is triggered, thereby generating the difficulty matching index that represents the deviation between the current difficulty and the user's ability.

[0056] In this embodiment of the invention, the game performance image features (such as scene complexity and character movement speed) and user behavior data (such as operation frequency and reaction time) are normalized to eliminate differences in units (e.g., reaction time is converted from "milliseconds" to a proportional value in the range [0, 1]); the average response time of the user to game stimuli per unit time is calculated (e.g., the average time from the appearance of multiple enemies to the attack), and the proportion of effective operation counts to total operation counts is calculated (e.g., the number of times a skill hits an enemy / the total number of skill releases); the difference between the user's actual task completion time and the system's preset standard time is compared (e.g., if the standard time is 60 seconds and the user takes 75 seconds, the deviation is +25%); the percentage of successful completions of a specific task by the user is calculated (e.g., 8 successful completions out of 10 tasks, the success rate is 80%). This invention unifies the multi-source data format, improves the compatibility of model input, accurately quantifies the user's ability dimensions, and provides a comparable benchmark indicator for difficulty assessment.

[0057] Normalized difference calculation:

[0058] First normalized difference (reaction time deviation): This compares the user's average reaction time with the system's preset "baseline reaction time" (such as the average reaction time threshold for novice players) and calculates the percentage difference. For example, if the baseline reaction time is 500 milliseconds and the user's average reaction time is 600 milliseconds, the difference is +20%, which is 0.2 after normalization.

[0059] The second normalized difference (task time deviation): compares the user's task completion time deviation with the "ideal task completion time" (such as the optimal completion time expected by the design) and calculates the deviation ratio. For example, if the ideal time is 50 seconds and the user's deviation is +30% (actual time taken is 65 seconds), the normalized value is 0.3.

[0060] This invention quantifies the gap between the user's reaction speed and task efficiency and the ideal level, intuitively reflects the shortcomings in ability, and provides clear numerical basis for difficulty adjustment (such as reducing the frequency of enemy appearance if the reaction is slow).

[0061] Operational performance evaluation component generation process:

[0062] Linear weighted fusion sums the operation accuracy and task success rate according to preset weights (e.g., operation accuracy accounts for 60%, task success rate accounts for 40%). For example, if the operation accuracy is 70% and the task success rate is 80%, the evaluation component is: 0.7×0.6+0.8×0.4=0.74 (i.e., 74% operation efficiency level). This comprehensively evaluates the stability of user operation and the quality of task completion, avoiding the one-sidedness of a single indicator, distinguishing between player types who are "fast in reaction but make many mistakes" and "slow in reaction but have high accuracy", and supporting differentiated difficulty adaptation.

[0063] Multidimensional coupling and dynamic weight allocation process:

[0064] Dynamic weight adjustment automatically adjusts the weights of each indicator based on the user's real-time physiological data (such as eye-tracking hotspot concentration and skin conductivity). For example, when user tension is detected (elevated conductivity), the weight of reaction time is reduced to avoid misjudging ability levels due to excessive emotional fluctuations.

[0065] Multi-dimensional coupled calculations dynamically weight the first normalized difference, the second normalized difference, and the operational efficiency evaluation component to generate a comprehensive difficulty matching index. For example, with a reaction time deviation weight of 0.3, a task time deviation weight of 0.3, and an operational efficiency weight of 0.4, the comprehensive index is: 0.2×0.3+0.3×0.3+0.74×0.4=0.476 (the smaller the value, the better the difficulty matching). This approach dynamically adjusts the evaluation focus based on physiological state, improving the accuracy of difficulty judgment. Multi-dimensional data cross-validation avoids misjudgments caused by accidental operational errors or sudden changes in scenario complexity.

[0066] Difficulty level mapping and instruction generation process:

[0067] Threshold range matching: Preset threshold range for difficulty adaptation (e.g., index < 0.3: difficulty too high, need to be downgraded; 0.3 ≤ index ≤ 0.7: difficulty moderate; index > 0.7: difficulty too low, need to be upgraded).

[0068] Commands are triggered based on the range of the comprehensive index. For example, a difficulty downgrade (such as reducing the number of enemies) is triggered when the index is 0.2, and a difficulty upgrade (such as increasing the movement speed of enemies) is triggered when the index is 0.8. This achieves automated decision-making for difficulty adjustment, reduces the cost of manual intervention, and ensures the logic and interpretability of difficulty changes through clear threshold rules, avoiding the experience fragmentation caused by random adjustments.

[0069] In a preferred embodiment of the present invention, step S3 involves performing a dynamic weighted calculation based on the difficulty adaptation index, and generating a dynamic difficulty correction coefficient by combining real-time user physiological feedback data. The dynamic difficulty correction coefficient includes scene complexity adjustment parameters and interaction response threshold adjustment parameters, including:

[0070] Obtain the player's current game level, base game level, and historical mission completion time sequence, and extract the characteristics of the player's ability growth trend;

[0071] Based on the difficulty adaptation index, the level adaptation component is calculated through a preset nonlinear conversion function for player level differences. The conversion function dynamically adjusts the gain amplitude according to the difference between the player level and the game's base level.

[0072] Based on the skin conductivity change curve and eye-tracking hotspot distribution in the real-time physiological feedback data of the user, the user stress load characteristics are extracted, and the weight ratio of the level adaptation component and the task time efficiency component in the dynamic weighted calculation is dynamically adjusted.

[0073] The difficulty adaptation index, level adaptation component, and task time efficiency component are fused from multiple sources to generate a comprehensive actual difficulty coefficient, wherein the task time efficiency component is represented by the reciprocal of the average task completion time of the player.

[0074] Based on the numerical range of the comprehensive actual difficulty coefficient, scene complexity adjustment parameters and interaction response threshold adjustment parameters are generated. The scene complexity adjustment parameters are used to dynamically control the density of game interface elements and the frequency of dynamic object generation. The interaction response threshold adjustment parameters are used to adaptively correct the error tolerance range of user operation input.

[0075] In this embodiment of the invention, the player's current game level (such as character level, experience points), basic game level (such as the default level for beginners) and historical task completion time records (such as the completion time sequence of the past 10 tasks) are read.

[0076] Trend Analysis:

[0077] The level difference is calculated by comparing the current level with the base level to determine the player's current stage of ability (e.g., being 2 levels higher than the base level indicates that the player has entered the proficient stage).

[0078] Time series analysis, by observing the fluctuation trend of historical task completion time (such as three consecutive shortened task times), can determine the speed of player ability growth (such as accelerated growth, stable improvement or stagnation), identify the player's ability development stage, and distinguish between the "newbie adaptation period" and the "expert challenge period"; it provides a long-term basis for ability evolution for difficulty adjustment and avoids short-term behavioral misjudgments (such as incorrect difficulty adjustment caused by accidental mistakes).

[0079] Level adaptation component calculation process:

[0080] The application of nonlinear transformation functions calculates level adaptation components based on the difference between the player's level and the base level (e.g., +N or -N levels) using preset functions (e.g., exponential or piecewise functions). For example, when the player's level is lower than the base level, the function outputs a negative value (indicating a need to reduce difficulty); when the player's level is higher than the base level, the function outputs a positive value (indicating a need to increase difficulty). The larger the difference, the larger the absolute value of the component (e.g., a +3 level difference has twice the component of a +1 level difference). This "step-by-step" adjustment of difficulty based on level differences aligns with player progression expectations. The nonlinear characteristic prevents lower-level players from losing challenge due to a sudden drop in difficulty, or higher-level players from lacking stimulation due to linear adjustments. The formula for the preset function can be:

[0081] ;

[0082] in, This represents the difference between the player's level and the base level. This is a parameter for adjusting the steepness of the curve (recommended value: 0.2-0.5). This is the scaling factor, which controls the overall adjustment range (recommended value: 0.1-0.3); when the player's level is lower than the base level... Using a negative exponential function, the output is negative, and the larger the level difference, the larger the absolute value of the negative value, but the slower the growth rate; when the player's level is higher than the base level... Using a positive exponential function, it outputs positive values, and the greater the level difference, the more exponentially the positive values ​​increase. The larger the value, the steeper the curve, and the more significant the increase in difficulty for advanced players. The larger the difference, the greater the overall adjustment and the more drastic the change in game difficulty. Each increase of 1 in the level difference results in a different change in the adaptive component (e.g., the change from +1 to +2 is about 0.37, and from +2 to +3 is about 0.65), forming a step-like increase in difficulty. The component of a +3 level difference (1.476) is approximately 5.6 times that of a +1 level difference (0.263), rather than a simple linear multiple relationship. This conforms to the requirement that "the larger the difference, the larger the absolute value of the component." This formula cleverly achieves a non-linear mapping between level and difficulty adjustment through an exponential function, ensuring that novice players have enough room to grow while providing continuous challenges for advanced players.

[0083] Pressure load characteristics and dynamic weight adjustment process:

[0084] Analysis of physiological indicators:

[0085] Skin conductivity analysis detects the peak frequency and duration of conductivity to determine the pressure level (e.g., conductivity > 20% of the baseline value and lasts for 5 minutes is marked as "high pressure"). It also counts the percentage of time the gaze lingers on non-critical areas (e.g., blank interfaces, irrelevant props) to assess the degree of attentional distraction (e.g., percentage > 40% indicates inattention).

[0086] Weighting adjustment strategy:

[0087] When the stress level increases, the weight of the task time efficiency component is automatically reduced (e.g., from 50% to 30%) to avoid operational delays caused by tension being misjudged as insufficient ability. When attention is distracted, the weight of the level adaptation component is increased (e.g., from 30% to 40%). The difficulty is adjusted based on the player's long-term ability first. The player's emotional state is sensed in real time, and the "ability assessment" and "experience protection" are dynamically balanced to avoid triggering incorrect difficulty adjustments due to short-term physiological fluctuations (e.g., operational delays caused by sudden interference).

[0088] The process of generating the overall difficulty level:

[0089] The task time efficiency component is calculated by taking the reciprocal of the player's average task completion time (e.g., if the average time is 80 seconds, the component is 1 / 80 ≈ 0.0125), representing the task completion efficiency per unit time. The difficulty adaptation index (reflecting current matching deviation), level adaptation component (reflecting long-term ability), and task time efficiency component (reflecting immediate performance) are then dynamically weighted (adjusted for by physiological data). For example:

[0090] With a difficulty matching index of 0.4 (moderate), a level matching component of +0.2 (higher than the basic level), and a task time efficiency component of 0.01, and weights of 50%, 30%, and 20% respectively, the comprehensive coefficient is: 0.4×0.5+0.2×0.3+0.01×0.2=0.262 (the larger the value, the higher the actual difficulty). By integrating multi-dimensional data such as "current matching degree", "long-term growth" and "immediate efficiency", the comprehensiveness of difficulty assessment is improved. The dynamic weighting mechanism allows the system to switch between "conservative adjustment" (when the pressure is high) and "active adjustment" (when the state is stable) modes.

[0091] Adjustment of parameter decomposition and application process:

[0092] Scene complexity adjustment parameter generation:

[0093] When the overall coefficient is greater than 0.5, reduce the density of interface elements (such as hiding non-critical buttons) and reduce the frequency of dynamic object generation (such as reducing the enemy refresh rate from 10 seconds / time to 15 seconds / time). When the overall coefficient is less than 0.3, increase the prominence of prompt information (such as enlarging the mission target icon) and increase the probability of item generation (such as increasing the probability of health packs from 20% to 30%).

[0094] Interactive response threshold adjustment parameter generation:

[0095] If the average player operation latency is greater than 500 milliseconds, the tolerance range for operation judgment is expanded (e.g., the judgment time for long-pressing skill release buttons is extended from 200 milliseconds to 300 milliseconds). If the operation accuracy is greater than 90%, the tolerance range is reduced (e.g., the button response time is shortened, and the operation precision requirements are increased). Scene adjustment directly affects visual cognitive load, avoiding information overload or scarcity. The interaction threshold is adapted to players of different operation levels. Novices can reduce frustration through lenient judgment, while experts can gain a sense of challenge and accomplishment through strict judgment.

[0096] In a preferred embodiment of the present invention, step S4 involves inputting the user feedback parameter set into an image feature weight allocator, dynamically adjusting the weight allocation of the image retrieval feature vector based on user attention distribution data and operation delay parameters, and generating an optimized scene element retrieval strategy, including:

[0097] Temporal convolution processing is performed on the eye-tracking hotspot distribution data in the user feedback parameter set to generate a user visual attention heatmap and extract image feature vectors of high-frequency gaze regions.

[0098] Based on the operation delay parameter, a time sensitivity evaluation function is constructed to prioritize the real-time response requirements of scene element retrieval feature vectors.

[0099] Based on the visual attention heatmap and time sensitivity assessment results, a dynamic feature weight allocation matrix is ​​established. The weights of interfering scene elements with a user misidentification rate higher than the threshold are attenuated, while the weights of key target features that are correctly identified by the user and sensitive to response delay are enhanced.

[0100] An incremental learning mechanism is used to optimize the adjusted feature weights online, and weight update coefficients with time decay characteristics are generated by combining historical operation accuracy data.

[0101] The optimized feature weights are mapped to the game scene database through a weighted feature space reconstruction algorithm to generate a dynamic retrieval strategy based on user behavior preferences. The retrieval strategy includes scene element visibility control instructions and target highlight rendering parameters, which are used to adjust the information density and interactive focus distribution of the game interface in real time.

[0102] In this embodiment of the invention, the hot spot coordinate sequence recorded by eye tracking is arranged in chronological order (e.g., 100 gaze points are recorded per second), and the gaze frequency in the time dimension is analyzed by sliding window convolution operation (e.g., 3-second window) to identify the area that the user is continuously focusing on (e.g., the skill bar in the upper left corner of the screen).

[0103] Heatmap construction maps the pixel coordinates of high-frequency gaze areas to a two-dimensional heatmap, with color intensity representing the gaze duration percentage (e.g., red areas indicate gaze duration > 60%). Image feature vectors of these areas are extracted (e.g., RGB color distribution, edge contour complexity) to accurately locate the user's visual focus, distinguish key operation areas (e.g., attack buttons) from secondary information areas, and provide an intuitive visual basis for prioritizing scene elements, preventing irrelevant elements from competing for attention resources.

[0104] Time sensitivity assessment and prioritization process:

[0105] Operation latency analysis calculates the average response time of users to different scene elements (such as enemies, items, and prompts). For example, the latency from seeing an enemy to clicking to attack is 400 milliseconds, and the latency from seeing an item to picking it up is 200 milliseconds.

[0106] Sensitivity function application, prioritizing based on delay threshold:

[0107] High sensitivity: latency <300 milliseconds (e.g., enemy movements in real-time combat), requires real-time priority retrieval;

[0108] Medium sensitivity: 300 milliseconds ≤ delay ≤ 500 milliseconds (such as task prompt text), can be processed slightly slower;

[0109] Low sensitivity: latency > 500 milliseconds (e.g., background decorative elements), non-urgent retrieval objects.

[0110] Ensure the retrieval priority of high-time-sensitivity elements (such as dynamic enemies) to improve operation response speed, allocate computing resources according to the user's actual processing capacity, and avoid low-priority elements consuming too much performance.

[0111] Dynamic allocation of feature weights and interference attenuation process:

[0112] The system analyzes the number of user clicks / interaction errors on scene elements (such as the frequency of accidentally touching advertising buttons) by statistically analyzing the false recognition rate. Elements with a false recognition rate >20% are marked as "interference items" (such as non-functional icons in the corner of the interface). The feature vector weights of interference items are multiplied by a decay coefficient (such as 0.5) to reduce their matching priority in image retrieval. For correctly identified and highly sensitive elements (such as task target arrows), the weights are multiplied by an enhancement coefficient (such as 1.5) to ensure priority rendering and retrieval. This invention reduces the misleading effect of interface interference elements on users, lowers the operation error rate, highlights the visual features of key targets (such as highlighted colors and enlarged sizes), and guides attention to quickly locate them.

[0113] Incremental learning and weight update process:

[0114] Historical data is integrated to record the accuracy of a user's past N operations (e.g., the number of hits in the past 100 skill releases). A decay sequence is generated in chronological order (e.g., the most recent operation has a weight of 10%, and the operation 7 days ago has a weight of 1%). After each adjustment of feature weights, the current operation accuracy is compared with the historical average. If the improvement is greater than 5%, the weight adjustment is retained; if the decrease is greater than 3%, the weight is reverted to the previous version and the adjustment range is reduced to avoid erroneous adjustments caused by fluctuations in a single operation. By smoothing weight changes through historical data, the system continuously learns changes in user operation habits (e.g., the shift in attention from novice to expert), achieving adaptive evolution of weight allocation.

[0115] Dynamic retrieval strategy generation process:

[0116] Feature space mapping matches the adjusted weight vectors with element features (such as enemy models and item icons) in the game scene database based on similarity, generating a search priority list (e.g., prioritizing searches when enemy feature matching is >90%). Low-weight elements (such as background decorations) are set to have transparency >70% or delayed loading, while high-weight elements (such as health bars) are forced to remain visible. Key target features (such as mission endpoint flags) are given glowing effects or outlines, with parameters dynamically adjusted based on weight (e.g., brightness increases by 15% for every 0.1 increase in weight). This dynamically balances the interface information density, avoiding "information bombardment" or "burying key information." Visual interaction focuses guide the user, reducing the cognitive cost of searching for the target and improving task completion efficiency. It also responds in real-time to changes in user attention, such as automatically adjusting the search strategy when transitioning from combat to puzzle-solving, enhancing the smoothness of scene transitions.

[0117] In a preferred embodiment of the present invention, step S5 involves establishing a multi-dimensional parameter fusion processor, mapping the dynamic difficulty correction coefficient to the optimized scene element retrieval strategy in the feature space, and generating a final game difficulty control instruction set through nonlinear superposition, including:

[0118] The scene complexity adjustment parameters, interaction response threshold adjustment parameters, and optimized scene element retrieval strategy are orthogonally decomposed to extract the nonlinear coupling features between multidimensional parameters and establish a multimodal feature space including spatial layout weight, dynamic element density, and interaction fault tolerance threshold.

[0119] By using an attention mechanism, cross-dimensional correlation analysis is performed on the parameters in the multimodal feature space to identify the implicit mapping relationship between scene complexity parameters and user attention distribution characteristics.

[0120] The interaction response threshold adjustment parameter in the dynamic difficulty correction coefficient is nonlinearly transformed using the hyperbolic tangent function to generate a sensitivity adjustment component that matches the user's current operation latency characteristics.

[0121] The scene element visibility control command and the target highlight rendering parameters are spatially encoded, and the feature channels are fused with the sensitivity adjustment component through a convolutional neural network to generate an enhanced rendering command that includes dynamic light and shadow effect intensity and interface element refresh rate.

[0122] Based on the preset game physics engine interface protocol, the fusion results of the multimodal feature space and the enhanced rendering instructions are encapsulated into an instruction set to generate the final game difficulty control instruction set, which includes enemy AI behavior patterns, resource refresh frequency and physics engine parameters, so as to realize the real-time adaptation of dynamic elements of the game environment and user operation capabilities.

[0123] In this embodiment of the invention, scene complexity parameters (such as element density and dynamic object generation frequency), interaction response thresholds (such as operation fault tolerance range), and retrieval strategies (such as element visibility control and highlight parameters) are split by dimensions (such as spatial dimension, temporal dimension, and visual dimension). Independent feature components between parameters (such as "information density" and "dynamic interference" in scene complexity) are extracted by methods such as principal component analysis (PCA). Each parameter component is mapped to a three-dimensional space (x-axis: spatial layout weight, y-axis: dynamic element density, z-axis: interaction fault tolerance threshold) to form a multimodal feature space.

[0124] This invention decouples redundant correlations between parameters, avoids imbalances in difficulty caused by repeated adjustments, and constructs a unified feature space to facilitate cross-dimensional analysis, such as discovering conflicting combinations of "high information density" and "low fault tolerance threshold".

[0125] Based on the distribution of user eye movement hotspots (such as gaze duration and saccade path), attention weights for each parameter dimension are calculated (e.g., high attention to the battle area gives higher weight to "enemy AI complexity"). Through association rule mining (such as the Apriori algorithm), hidden dependencies between parameters are discovered (e.g., "increased density of interface elements" leads to "decreased operation accuracy", but this is only significant when "time pressure is high"). This invention achieves "intelligent focusing" of parameter adjustment, prioritizes optimizing the difficulty experience of the user's attention area, and discovers parameter combination effects that are difficult for humans to perceive intuitively (e.g., "low fault tolerance + high visual interference" will significantly reduce the game experience).

[0126] Nonlinear transformation and sensitivity adjustment process:

[0127] The threshold nonlinear transformation applies a hyperbolic tangent function (tanh) to the interaction response threshold parameter, mapping the original threshold range (e.g., 0-100ms) to the [-1, 1] interval, generating a sensitivity adjustment component. For example, when the average user operation delay is 400ms, the function output is approximately 0.8, indicating that the system response sensitivity needs to be increased. The operation judgment logic is adjusted in real time based on the sensitivity component (e.g., extending the skill release judgment time, expanding the click area). This invention smooths the difficulty adjustment curve, avoiding the "difficulty cliff" caused by linear adjustment (e.g., when the operation delay increases from 300ms to 400ms, the sensitivity change changes from 0.2 to 0.5), and adapts to players with different operation habits (e.g., the difference in response needs between touch screen users and keyboard and mouse users).

[0128] Spatial encoding and rendering instruction enhancement process:

[0129] The scene element visibility control commands (such as "hide background decoration") and highlight parameters (such as "task target brightness +50%) are converted into encoded vectors in three-dimensional space (such as [x, y, z] representing position, size, and brightness respectively).

[0130] CNN Feature Fusion:

[0131] By processing the encoded vector through a convolutional neural network (CNN) and performing channel-level fusion with the sensitivity adjustment component (such as multiplying the "enemy highlight intensity" and "operation delay" feature maps), the invention outputs enhanced parameters (such as dynamic light and shadow intensity decreasing with increasing operation delay and interface refresh rate increasing with increased attention distraction). This invention achieves coordinated changes in visual performance and operational difficulty. For example, it automatically enhances target cues when operational errors are frequent, and uses CNN to capture complex relationships of spatial parameters (such as the visual interference effect of multiple highlighted elements).

[0132] Instruction set encapsulation and physics engine adaptation process:

[0133] The fusion results of the multimodal feature space (such as spatial layout weights and interaction thresholds) are converted into parameters that the game physics engine can recognize (such as enemy movement speed and collision volume). The instruction set is encapsulated according to the physics engine interface protocol (such as Unity's API), including: AI behavior pattern adjustment (such as enemy attack frequency and pursuit distance); resource refresh mechanism (such as item generation time interval and rarity); physical parameters (such as character jump height and weapon recoil).

[0134] This invention achieves a complete closed loop from the "perception layer" (user behavior) to the "execution layer" (game engine), supports cross-platform deployment, adapts to different game engines (such as Unity and Unreal) through a unified interface protocol, and provides more subtle difficulty changes through dynamic physical parameter adjustment (such as reducing character gravity and lowering jumping difficulty when operation latency is high).

[0135] In a preferred embodiment of the present invention, step S6 involves receiving real-time user feedback on the adjusted game difficulty, iteratively optimizing it using a backpropagation algorithm to form a dynamically adaptive difficulty adjustment loop, including:

[0136] The game log interface is synchronized with biosensors to capture real-time user experience evaluation data, including the amount of subjective difficulty score adjustment, the rate of change of physiological stress indicators, and the frequency of operation interruption.

[0137] Design a hierarchical convolutional neural network architecture, including a temporal convolutional layer and an attention mechanism module, to extract multi-dimensional features from the execution effect of the final game difficulty control instruction set, and generate an error feature vector that includes enemy AI behavior deviation, resource refresh anomaly rate and interface interaction delay.

[0138] The error feature vector is aligned with the real-time user experience evaluation data across modalities using a loss function, and the parameter gradient of the difficulty assessment model and the feature offset of the image feature weight allocator are dynamically calculated.

[0139] The baseline pattern matching parameters in the difficulty assessment model and the attention response coefficients of the image feature weight allocator are jointly iteratively updated, and the optimization direction simultaneously satisfies the constraints of operation trajectory fitting accuracy and user physiological load balance.

[0140] Based on the multimodal data fusion gateway and the optimized parameter configuration, a difficulty control compensation instruction containing a dynamic balance factor is generated, and a spatiotemporally coordinated closed-loop adjustment loop is formed with the scene element retrieval strategy.

[0141] In this embodiment of the invention, the process of constructing the above-mentioned difficulty assessment model is as follows:

[0142] Model architecture design:

[0143] Multi-input branch design

[0144] Behavioral data branch: Processes time-series data such as operation latency and accuracy, and uses an LSTM network to capture changes in operation patterns;

[0145] Physiological data branch: Analyze biological signals such as skin conductivity and heart rate variability, and extract physiological features using 1D-CNN;

[0146] Scene data branch: Analyze in-game logs (such as enemy positions and item spawns) and model scene complexity using graph neural networks (GNNs);

[0147] Hierarchical feature fusion mechanism:

[0148] Low-level feature layer: Each branch extracts features independently (e.g., the behavior branch extracts the combo interval feature).

[0149] Intermediate fusion layer: Multi-source features are fused in a weighted manner through an attention mechanism (e.g., assigning higher weights to physiological features under high stress).

[0150] Decision output layer: The fully connected layer outputs a difficulty adaptation index (0-1 range, representing the degree of matching between the current difficulty and the user's ability).

[0151] Baseline pattern library construction:

[0152] User capability grading system:

[0153] Players are divided into 5 ability levels (such as novice, intermediate, proficient, expert, and master) based on factors such as operation speed, accuracy, and strategy complexity.

[0154] Establish benchmark behavior patterns for each level (e.g., the reaction time threshold for expert players is 200ms, and the operation accuracy is ≥85%).

[0155] Difficulty Scene Template:

[0156] Design 20 typical game scene templates (such as BOSS battles, resource collection, and puzzle levels), and preset difficulty parameter combinations for each scene template (such as the hint frequency in puzzle scenes and the number of enemies in battle scenes).

[0157] Dynamic matching algorithm:

[0158] The behavior pattern matching engine calculates the cosine similarity between a user's real-time behavior sequence and a baseline pattern, and uses the Dynamic Time Warping (DTW) algorithm to handle temporal data alignment issues (such as differences in skill release rhythms among different players).

[0159] Multi-dimensional scoring mechanism:

[0160] Responsiveness score: calculated based on the deviation of operational delay from a baseline threshold;

[0161] Strategy depth scoring: Analyzes the complexity of user path selection using decision trees;

[0162] Stress tolerance score: assessed by combining the amplitude and duration of fluctuations in physiological indicators.

[0163] Adaptive adjustment:

[0164] Difficulty range setting:

[0165] Set a difficulty adjustment range of ±15% for each ability level (e.g., the beginner level can fluctuate within the range of 85%-100% of the base difficulty). Dynamically adjust the range width according to the user's status (e.g., expand the lower limit to 70% when the user is fatigued). When a decrease in operation accuracy is detected, adjust the enemy attack frequency and damage value simultaneously. When physiological stress indicators rise, prioritize reducing visual complexity rather than directly reducing the number of enemies.

[0166] Model training process:

[0167] Mixed data training:

[0168] Supervised learning phase: The base model is trained using labeled player data (such as the actions of expert players at different difficulty levels).

[0169] Reinforcement learning phase: The difficulty adjustment strategy is optimized through the A3C algorithm, with user retention rate and experience satisfaction as the reward function.

[0170] Continuous evolution mechanism: Incremental training is triggered every 100,000 new data points collected, using federated learning technology to protect user privacy, and model update calculations are completed on local devices.

[0171] Verification and evaluation system:

[0172] Offline verification metrics:

[0173] Difficulty prediction accuracy: Evaluates the degree of match between the difficulty level predicted by the model and the player's actual performance.

[0174] Adjust response speed: Measure the delay time from the change in user ability to the difficulty adjustment taking effect.

[0175] Online A / B testing:

[0176] 10% of users are randomly assigned to the model experiment group, and core indicators such as retention rate and daily active time are compared. The model iteration effect is evaluated weekly, and poorly performing adjustment strategies are eliminated. Through multi-dimensional feature analysis, millimeter-level assessment of players' abilities is achieved, with a latency of less than 10 seconds from recognizing changes in ability to completing difficulty adjustments.

[0177] By collecting user feedback on difficulty levels (e.g., changing from 3 stars to 4 stars) through in-game pop-ups (e.g., "Is the current difficulty appropriate?") or controller buttons, and synchronizing biosensor data, this invention calculates stress indicators such as skin conductivity change rate (e.g., fluctuation range per minute) and heart rate variability (HRV), and counts the number of abnormal operation interruptions by the user within a fixed time period (e.g., the frequency of suddenly stopping movement for more than 5 seconds). This invention uses multi-source data to cross-verify the user's real experience (e.g., a high score but a high heart rate may indicate "too challenging"), and captures experience fluctuations in real time (e.g., quickly triggering adjustments when operation interruptions surge), thus avoiding the accumulation of negative experiences.

[0178] Hierarchical CNN and its performance feature extraction process:

[0179] Temporal convolutional layer processing inputs game command execution data (such as enemy movement trajectory and item refresh time) into the CNN in time sequence to extract temporal features (such as periodic fluctuations in enemy attack frequency).

[0180] The attention mechanism is applied to weight the feature map, highlighting key error terms (such as giving higher weight when the resource refresh anomaly rate is >15%), and generating an error vector that includes AI behavior deviations (such as enemies not moving along the preset path) and interaction delays (such as skill release animation stuttering).

[0181] This invention captures subtle anomalies in instruction execution (such as difficulty imbalance caused by AI path deviation), focuses attention mechanisms on core issues, reduces irrelevant data interference, and improves feature extraction efficiency.

[0182] Cross-modal alignment and parameter gradient calculation process:

[0183] The loss function is designed to simultaneously calculate subjective rating errors (such as the difference between predicted and actual ratings) and physiological index errors (such as the deviation between predicted pressure values ​​and measured conductivity).

[0184] Cross-modal fusion maps the error feature vector and experience data to the same dimensional space (e.g., both are converted to normalized values ​​in the [0,1] interval) and calculates the joint loss value.

[0185] Gradient backpropagation calculates the adjustment direction of the baseline matching parameters (such as reaction time threshold) and the attention coefficient in the weight allocator in the difficulty assessment model through the backpropagation algorithm (such as increasing the weight of "operational accuracy"). This invention ensures that the model optimization simultaneously meets the dual goals of "subjective experience" and "physiological load". Cross-modal alignment avoids misleading data from a single source (such as relying solely on subjective ratings may ignore the hidden pressure of users).

[0186] Joint iterative update and constraint optimization process:

[0187] Parameter grouping update:

[0188] Difficulty assessment model, adjusting benchmark mode matching parameters (e.g., increasing the novice benchmark reaction time from 500ms to 600ms to accommodate slow-reaction players);

[0189] The weight allocator updates the attention response coefficients (e.g., increasing the influence factor of "eye-tracking hotspot concentration" on the weights of scene elements).

[0190] Application of constraints:

[0191] The accuracy of the operation trajectory fitting is ≥85% (ensuring that the difficulty adjustment does not deviate from the player's actual ability), and the physiological load is balanced (such as the fluctuation of skin conductivity is controlled within ±15% of the benchmark value).

[0192] This invention avoids model overfitting to short-term data (such as a sudden increase in difficulty due to a single high score), balances "ability matching" and "experience health", and prevents players from becoming fatigued due to excessive challenges.

[0193] Closed-loop control circuit and compensation command generation process:

[0194] Compensation instructions are generated, and dynamic balancing factors (such as buffer coefficients that automatically adjust enemy strength when the difficulty adaptation index fluctuates) are generated based on the optimized parameters.

[0195] Spatiotemporal Coordination and Regulation:

[0196] In terms of time, the difficulty is smoothly adjusted according to historical optimization cycles (e.g., every 10 minutes) to avoid sudden changes;

[0197] In the spatial dimension, combined with scene element retrieval strategies (such as temporarily enhancing map marker weight in maze scenes), the "command-scene" linkage is realized. The dynamic balancing factor of this invention prevents over-adjustment (such as limiting the difficulty increase or decrease to ±20% of the current level), and the spatiotemporal coordination ensures that the adjustment conforms to the scene logic (such as prioritizing the optimization of enemy AI in combat scenes and prioritizing the adjustment of hint density in puzzle-solving scenes).

[0198] like Figure 2 As shown, embodiments of the present invention also provide a dynamic game difficulty adaptive adjustment system based on user behavior feedback, including:

[0199] The acquisition module is used to collect multi-dimensional behavioral data streams of users in real time during game interaction, and to extract features from the behavioral data streams to obtain game performance image feature sets and user feedback parameter sets.

[0200] The calculation module is used to generate a difficulty adaptation index by calculating the matching degree between the user's operation trajectory and the preset benchmark mode based on the game performance image feature set. The difficulty adaptation index represents the matching deviation between the current game difficulty level and the user's ability level. Based on the difficulty adaptation index, dynamic weighted calculation is performed, and combined with the user's real-time physiological feedback data, a dynamic difficulty correction coefficient is generated. The dynamic difficulty correction coefficient includes scene complexity adjustment parameters and interaction response threshold adjustment parameters.

[0201] The fusion module is used to input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight allocation of the image retrieval feature vector according to the user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy; and establish a multi-dimensional parameter fusion module to map the dynamic difficulty correction coefficient with the optimized scene element retrieval strategy in the feature space, and generate the final game difficulty control instruction set through nonlinear superposition.

[0202] The adjustment module is used to receive user feedback on the adjusted game difficulty in real time, and iterates through the backpropagation algorithm to form a dynamic and adaptive difficulty adjustment loop.

[0203] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for dynamically adjusting game difficulty based on user behavior feedback, characterized in that, The method includes: Step S1: During the game interaction process, collect multi-dimensional behavioral data streams of users in real time, and extract features from the behavioral data streams to obtain game performance image feature sets and user feedback parameter sets; Step S2: Based on the game performance image feature set, a difficulty adaptation index is generated by calculating the matching degree between the user operation trajectory and the preset benchmark mode. Step S3: Perform dynamic weighted calculation based on the difficulty adaptation index, and generate a dynamic difficulty correction coefficient by combining the user's real-time physiological feedback data; Step S4: Input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight allocation of the image retrieval feature vector according to the user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy. Step S5: Map the dynamic difficulty correction coefficient to the optimized scene element retrieval strategy in the feature space, and generate the final game difficulty control instruction set through nonlinear superposition. Step S6: Receive real-time user feedback on the adjusted game difficulty, and iteratively optimize through backpropagation algorithm to form a dynamically adaptive difficulty adjustment loop.

2. The method for adaptive adjustment of dynamic game difficulty based on user behavior feedback according to claim 1, characterized in that, Step S1: During game interaction, collect multi-dimensional behavioral data streams of the user in real time, and extract features from the behavioral data streams to obtain a game performance image feature set and a user feedback parameter set, including: By embedding data points in the game client, user operation behavior sequence data is collected, including operation frequency, reaction time, task completion path trajectory and error triggering events; The game performance image data stream is captured through the graphics processor interface, and multi-scale feature extraction is performed on continuous game image frames to generate the game performance image feature set, which includes the distribution density of interface elements, character motion vectors, and scene complexity indicators. User feedback data is collected synchronously through biosensors and an external interactive platform, including eye-tracking hotspot distribution, skin conductivity change curves, and subjective difficulty score input. The user feedback parameter set is generated by combining in-game operation latency parameters. The feature extraction process employs a sliding window mechanism to align multi-source heterogeneous data in the temporal domain, and uses a convolutional neural network to perform dynamic object detection and behavioral intent classification on image frames, forming a feature matrix with synchronized timestamps.

3. The method for adaptive adjustment of dynamic game difficulty based on user behavior feedback according to claim 2, characterized in that, Step S2: Based on the game performance image feature set, a difficulty adaptation index is generated by calculating the matching degree between the user's operation trajectory and a preset benchmark mode. The difficulty adaptation index represents the matching deviation between the current game difficulty level and the user's ability level, including: The behavioral data in the game performance image feature set and user feedback parameter set are standardized and preprocessed to extract user ability-related features, including the player's average reaction time, operation accuracy, task completion time deviation and task success rate. Based on the preset benchmark reaction time and ideal task completion time, the first normalized difference between the average reaction time and the benchmark reaction time, and the second normalized difference between the deviation of the task completion time and the ideal task completion time are calculated respectively. The operational accuracy rate and task success rate are linearly weighted and fused to generate an operational performance evaluation component. A dynamic weight allocation strategy is used to couple the first normalized difference, the second normalized difference, and the operational efficiency evaluation component in multiple dimensions to generate a comprehensive difficulty matching index, wherein the weight coefficients are adaptively adjusted based on the user's current physiological feedback data. The comprehensive difficulty matching index is mapped to a preset difficulty level threshold range. If the index is lower than the first threshold, a difficulty downgrade instruction is triggered; if it is higher than the second threshold, a difficulty upgrade instruction is triggered, thereby generating the difficulty matching index that represents the deviation between the current difficulty and the user's ability.

4. The method for adaptive adjustment of dynamic game difficulty based on user behavior feedback according to claim 3, characterized in that, Step S3: Based on the difficulty adaptation index, perform dynamic weighted calculation and combine it with real-time user physiological feedback data to generate a dynamic difficulty correction coefficient. The dynamic difficulty correction coefficient includes scene complexity adjustment parameters and interaction response threshold adjustment parameters, including: Obtain the player's current game level, base game level, and historical mission completion time sequence, and extract the characteristics of the player's ability growth trend; Based on the difficulty adaptation index, the level adaptation component is calculated through a preset nonlinear conversion function for player level differences. The conversion function dynamically adjusts the gain amplitude according to the difference between the player level and the game's base level. Based on the skin conductivity change curve and eye-tracking hotspot distribution in the real-time physiological feedback data of the user, the user stress load characteristics are extracted, and the weight ratio of the level adaptation component and the task time efficiency component in the dynamic weighted calculation is dynamically adjusted. The difficulty adaptation index, level adaptation component, and task time efficiency component are fused from multiple sources to generate a comprehensive actual difficulty coefficient, wherein the task time efficiency component is represented by the reciprocal of the average task completion time of the player. Based on the numerical range of the comprehensive actual difficulty coefficient, scene complexity adjustment parameters and interaction response threshold adjustment parameters are generated. The scene complexity adjustment parameters are used to dynamically control the density of game interface elements and the frequency of dynamic object generation. The interaction response threshold adjustment parameters are used to adaptively correct the error tolerance range of user operation input.

5. The dynamic game difficulty adaptive adjustment method based on user behavior feedback according to claim 4, characterized in that, Step S4: Input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight allocation of the image retrieval feature vector according to the user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy, including: Temporal convolution processing is performed on the eye-tracking hotspot distribution data in the user feedback parameter set to generate a user visual attention heatmap and extract image feature vectors of high-frequency gaze regions. Based on the operation delay parameter, a time sensitivity evaluation function is constructed to prioritize the real-time response requirements of scene element retrieval feature vectors. Based on the visual attention heatmap and time sensitivity assessment results, a dynamic feature weight allocation matrix is ​​established. The weights of interfering scene elements with a user misidentification rate higher than the threshold are attenuated, while the weights of key target features that are correctly identified by the user and sensitive to response delay are enhanced. An incremental learning mechanism is used to optimize the adjusted feature weights online, and weight update coefficients with time decay characteristics are generated by combining historical operation accuracy data. The optimized feature weights are mapped to the game scene database through a weighted feature space reconstruction algorithm to generate a dynamic retrieval strategy based on user behavior preferences. The retrieval strategy includes scene element visibility control instructions and target highlight rendering parameters, which are used to adjust the information density and interactive focus distribution of the game interface in real time.

6. The method for adaptive adjustment of dynamic game difficulty based on user behavior feedback according to claim 5, characterized in that, Step S5: Establish a multi-dimensional parameter fusion processor, map the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy into the feature space, and generate the final game difficulty control instruction set through nonlinear superposition, including: The scene complexity adjustment parameters, interaction response threshold adjustment parameters, and optimized scene element retrieval strategy are orthogonally decomposed to extract the nonlinear coupling features between multidimensional parameters and establish a multimodal feature space including spatial layout weight, dynamic element density, and interaction fault tolerance threshold. By using an attention mechanism, cross-dimensional correlation analysis is performed on the parameters in the multimodal feature space to identify the implicit mapping relationship between scene complexity parameters and user attention distribution characteristics. The interaction response threshold adjustment parameter in the dynamic difficulty correction coefficient is nonlinearly transformed using the hyperbolic tangent function to generate a sensitivity adjustment component that matches the user's current operation latency characteristics. The scene element visibility control command and the target highlight rendering parameters are spatially encoded, and the feature channels are fused with the sensitivity adjustment component through a convolutional neural network to generate an enhanced rendering command that includes dynamic light and shadow effect intensity and interface element refresh rate. Based on the preset game physics engine interface protocol, the fusion results of the multimodal feature space and the enhanced rendering instructions are encapsulated into an instruction set to generate the final game difficulty control instruction set, which includes enemy AI behavior patterns, resource refresh frequency and physics engine parameters, so as to realize the real-time adaptation of dynamic elements of the game environment and user operation capabilities.

7. The method for adaptive adjustment of dynamic game difficulty based on user behavior feedback according to claim 6, characterized in that, Step S6: Receive real-time user feedback on the adjusted game difficulty, iteratively optimize using a backpropagation algorithm, and form a dynamically adaptive difficulty adjustment loop, including: The game log interface is synchronized with biosensors to capture real-time user experience evaluation data, including the amount of subjective difficulty score adjustment, the rate of change of physiological stress indicators, and the frequency of operation interruption. Design a hierarchical convolutional neural network architecture, including a temporal convolutional layer and an attention mechanism module, to extract multi-dimensional features from the execution effect of the final game difficulty control instruction set, and generate an error feature vector that includes enemy AI behavior deviation, resource refresh anomaly rate and interface interaction delay. The error feature vector is aligned with the real-time user experience evaluation data across modalities using a loss function, and the parameter gradient of the difficulty assessment model and the feature offset of the image feature weight allocator are dynamically calculated. The baseline pattern matching parameters in the difficulty assessment model and the attention response coefficients of the image feature weight allocator are jointly iteratively updated, and the optimization direction simultaneously satisfies the constraints of operation trajectory fitting accuracy and user physiological load balance. Based on the multimodal data fusion gateway and the optimized parameter configuration, a difficulty control compensation instruction containing a dynamic balance factor is generated, and a spatiotemporally coordinated closed-loop adjustment loop is formed with the scene element retrieval strategy.

8. A dynamic game difficulty adaptive adjustment system based on user behavior feedback, the system being used to perform the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to collect multi-dimensional behavioral data streams of users in real time during game interaction, and to extract features from the behavioral data streams to obtain game performance image feature sets and user feedback parameter sets. The calculation module is used to generate a difficulty adaptation index by calculating the matching degree between the user's operation trajectory and the preset benchmark mode based on the game performance image feature set. The difficulty adaptation index represents the matching deviation between the current game difficulty level and the user's ability level. Based on the difficulty adaptation index, dynamic weighted calculation is performed, and combined with the user's real-time physiological feedback data, a dynamic difficulty correction coefficient is generated. The dynamic difficulty correction coefficient includes scene complexity adjustment parameters and interaction response threshold adjustment parameters. The fusion module is used to input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight allocation of the image retrieval feature vector according to the user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy. A multi-dimensional parameter fusion device is established to map the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy into the feature space, and the final game difficulty control instruction set is generated through nonlinear superposition. The adjustment module is used to receive user feedback on the adjusted game difficulty in real time, and iterates through the backpropagation algorithm to form a dynamic and adaptive difficulty adjustment loop.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images