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

Abstract

The invention 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: carrying out dynamic weighting calculation based on a difficulty adaptation index, generating a dynamic difficulty correction coefficient in combination with user real-time physiological feedback data, and adjusting the dynamic game difficulty according to the dynamic difficulty correction coefficient; the dynamic difficulty correction coefficient comprises a scene complexity adjustment parameter and an interaction response threshold adjustment parameter; inputting the user feedback parameter set into an image feature weight distributor, dynamically adjusting weight distribution of image retrieval feature vectors according to user attention distribution data and operation delay parameters, and generating an optimized scene element retrieval strategy; and 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. Game design can be optimized, and game adaptability and flexibility are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of game design, in particular to a dynamic game difficulty self-adaptive adjustment method and system based on user behavior feedback. BACKGROUND

[0002] With the rapid development of the electronic game industry, players' demand for personalized game experience is increasing. Traditional game difficulty adjustment mechanisms mostly use static setting or simple linear adjustment methods, which are usually based on single-dimensional data such as players' completion time and failure times to determine the difficulty.

[0003] Such methods have some defects, for example: Firstly, static threshold setting cannot adapt to the dynamic changes of different players' operation ability, which easily leads to difficulty imbalance; secondly, it lacks real-time perception of players' physiological state (such as attention concentration and stress level), which makes it difficult to capture users' real experience load; thirdly, the relevance of game scene elements and operation feedback is not effectively utilized, which leads to the lack of coordination between difficulty adjustment and interface information presentation. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a dynamic game difficulty self-adaptive adjustment method and system based on user behavior feedback, which can improve user experience, optimize game design, and enhance game adaptability and flexibility.

[0005] To solve the above technical problems, the technical solution of the present application is as follows: In a first aspect, a dynamic game difficulty self-adaptive adjustment method based on user behavior feedback, the method comprising: Step S1, real-time collection of multi-dimensional behavior data stream of users in the game interaction process, feature extraction of the behavior data stream to obtain game performance image feature set and user feedback parameter set; Step S2, generation of difficulty adaptation index by calculating the matching degree of user operation trajectory and preset reference mode according to the game performance image feature set, the difficulty adaptation index representing the matching deviation amount of current game difficulty level and user ability level; Step S3, dynamic weighted calculation based on the difficulty adaptation index, combination of user real-time physiological feedback data to generate dynamic difficulty correction coefficient, the dynamic difficulty correction coefficient including scene complexity adjustment parameter and interactive response threshold adjustment parameter; Step S4, input of the user feedback parameter set into image feature weight distributor, dynamic adjustment of weight distribution of image retrieval feature vector according to user attention distribution data and operation delay parameter to generate optimized scene element retrieval strategy; Step S5, a multi-dimensional parameter fusioner is established to map the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy in a feature space, and a final game difficulty control instruction set is generated through nonlinear superposition; Step S6, real-time user experience evaluation data for the adjusted game difficulty is received, and a dynamic self-adaptive difficulty adjustment loop is formed through iterative optimization of a back propagation algorithm.

[0006] In a second aspect, a dynamic game difficulty self-adaptive adjustment system based on user behavior feedback includes: An acquisition module is configured to collect a multi-dimensional behavior data stream of a user in real time during a game interaction process, and extract features from the behavior data stream to obtain a game performance image feature set and a user feedback parameter set; A calculation module is configured to generate a difficulty adaptation index by calculating a matching degree between a user operation trajectory and a preset reference mode according to the game performance image feature set, wherein the difficulty adaptation index represents a matching deviation amount between a current game difficulty level and a user ability level; and generate a dynamic difficulty correction coefficient by dynamically weighting calculation based on the difficulty adaptation index and combining real-time physiological feedback data of the user, wherein the dynamic difficulty correction coefficient includes a scene complexity adjustment parameter and an interactive response threshold adjustment parameter; A fusion module is configured to input the user feedback parameter set into an image feature weight distributor, dynamically adjust weight distribution of an image retrieval feature vector according to user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy; and establish a multi-dimensional parameter fusioner to map the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy in a feature space, and generate a final game difficulty control instruction set through nonlinear superposition. An adjustment module is configured to receive real-time user experience evaluation data for an adjusted game difficulty, and form a dynamic self-adaptive difficulty adjustment loop through iterative optimization of a back propagation algorithm.

[0007] In a third aspect, a computing device includes: One or more processors; A storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0008] In a fourth aspect, a computer readable storage medium stores a program, when the program is executed by a processor, the method is implemented.

[0009] The above-mentioned scheme of the present application at least has the following beneficial effects: By fusing user operation behavior, physiological feedback and visual attention distribution data, the limitation of traditional single dimension adjustment is broken through, dynamic balance of user cognitive load and game difficulty is realized, and the experience fragmentation problem caused by operation delay or information overload of players is solved.

[0010] Based on matching degree calculation of user operation trajectory and preset reference mode, combined with real-time feedback of physiological stress index, a multi-dimensional linkage difficulty correction coefficient is generated, which reduces the difficulty adjustment error and reduces the frustration of players caused by sudden difficulty.

[0011] Through dynamic optimization of scene element retrieval strategy by image feature weight distributor, combined with user attention heat map and operation delay sensitivity, the interface information density and interaction focus are automatically adjusted, the target recognition efficiency is improved, and the user misoperation rate is reduced.

[0012] Adopting feature space mapping and nonlinear superposition technology, the depth coupling of scene complexity parameters, interaction fault tolerance threshold and rendering instructions is realized, the real-time matching of game environment dynamic elements (such as enemy AI behavior, resource refreshing logic) and user operation ability is realized, and the game immersion is enhanced.

[0013] Through continuous optimization of difficulty evaluation model and feature weight distributor by back propagation algorithm, combined with cross-modal alignment of user experience evaluation data, a dynamic evolution adjustment loop is formed, the adaptive iteration speed is improved, and the difficulty curve and player growth are kept synchronized in the long term.

[0014] Using sliding window mechanism and time domain alignment technology to process multi-source heterogeneous data, reducing the calculation redundancy in feature extraction process, compared with traditional method, reducing the GPU load, ensuring the stable operation of real-time adjustment on mainstream game hardware platform. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 is a flowchart of a dynamic game difficulty self-adaptive adjustment method based on user behavior feedback provided by an embodiment of the present application.

[0016] Figure 2 is a schematic diagram of a dynamic game difficulty self-adaptive adjustment system based on user behavior feedback provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0018] AsFigure 1 As shown, the embodiments of the present application propose a dynamic game difficulty self-adaptive adjustment method based on user behavior feedback, which comprises the following steps: Step S1, real-time collection of multi-dimensional behavior data stream of the user in the game interaction process, feature extraction of the behavior data stream to obtain game performance image feature set and user feedback parameter set; Step S2, generation of difficulty adaptation index by calculating the matching degree of user operation trajectory and preset reference mode according to the game performance image feature set; Step S3, dynamic weighting calculation based on the difficulty adaptation index, combined with real-time physiological feedback data of the user to generate dynamic difficulty correction coefficient; Step S4, input of the user feedback parameter set into image feature weight distributor, dynamic adjustment of weight distribution of image retrieval feature vector according to user attention distribution data and operation delay parameter to generate optimized scene element retrieval strategy; Step S5, feature space mapping of the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy, generation of final game difficulty control instruction set through nonlinear superposition; Step S6, real-time reception of experience evaluation data of the user on the adjusted game difficulty, iterative optimization through back propagation algorithm to form a dynamic self-adaptive difficulty adjustment loop.

[0019] In an embodiment of the present invention, a personalized difficulty adaptation index is generated by real-time analysis of the degree of match between a player's operation trajectory and a preset benchmark pattern (step S2), so that the game difficulty is dynamically matched to the player's current ability, avoiding frustration caused by excessively high difficulty or boredom caused by too low difficulty. A dynamic difficulty correction coefficient is generated by combining real-time physiological data such as heart rate and eye movement (step S3), accurately capturing the player's emotional fluctuations (such as tension and fatigue), further refining the difficulty adjustment dimension, 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, for example, highlighting the key targets of the player's attention or simplifying interference elements, guiding the player to complete the task more efficiently and improving the smoothness of operation. Through feature space mapping and nonlinear superposition (step S5), the difficulty correction is deeply integrated with the scene element retrieval strategy to achieve coordinated adjustment of multi-dimensional parameters (such as enemy strength and environment complexity), making the difficulty change more natural and delicate. The player experience evaluation data (step S6) is used to iteratively optimize the model parameters through the back propagation 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 adaptation in the long term, reduce the workload of manually pre-setting difficulty levels, cover a wider range of player groups (from novices to experts) through automation mechanisms, and improve the universality and life cycle value of the game; integrate multi-source data such as operation trajectories, physiological signals, and attention distribution (steps S1 to S4) to avoid the one-sidedness of single-dimensional judgment and make the basis for difficulty adjustment more comprehensive and reliable; real-time calculation of the entire process (from data collection to instruction generation) ensures that difficulty adjustment is synchronized with player behavior, reduces the sense of delay, and enhances the immediacy and authenticity of interaction.

[0020] In a preferred embodiment of the present invention, step S1 collects a multi-dimensional behavior data stream of a user in real time during game interaction, and performs feature extraction on the behavior data stream to obtain a game performance image feature set and a user feedback parameter set, including: Collect user action sequence data through game client tracking, including action frequency, response time, task completion path, and error triggering events; Capturing a game performance image data stream through a graphics processor interface, performing multi-scale feature extraction on consecutive game image frames, and generating the game performance image feature set including interface element distribution density, character motion vectors, and scene complexity indicators; Synchronously collecting user feedback data through biosensors and an external interactive platform, including eye tracking hotspot distribution, skin conductivity change curves, and subjective difficulty rating inputs, and combining them with in-game operation delay parameters to generate the user feedback parameter set; The feature extraction process adopts a sliding window mechanism to align the time domain of the multi-source heterogeneous data, and a convolutional neural network is used for dynamic object detection and behavior intention classification of the image frames, so as to form a timestamp-synchronized feature matrix.

[0021] In the embodiment of the application, through the preset burying points (such as button clicking, keyboard input and the like interactive nodes) of the game client, the number of operations of the user within a unit time (such as the number of clicks per minute) is recorded in real time; the time interval (such as the time consumption from seeing the enemy to clicking the attack) from receiving the game stimulus (such as the appearance of the enemy, the task prompt) to performing the corresponding operation of the user is tracked; based on the game map coordinate system, the coordinate sequence of the moving path (such as the moving trajectory curve from point A to point B) of the user controlling the role is recorded; the event (such as the skill release error, the number of collision obstacles) of the user triggering the invalid operation or the task failure is identified and recorded, the operation efficiency and the proficiency of the user are quantified, the operation bottleneck (such as the high-frequency error area) is identified, and the basic behavior data for difficulty evaluation is provided, such as the player who reacts slowly may need to reduce the operation complexity.

[0022] By means of the GPU interface, the continuous frames (such as 30 frames per second) of the game picture are intercepted in real time, the original pixel data is extracted, the distribution density of the elements (such as the buttons, the prompt information and the like) in the interface is counted under different resolutions (such as global, local magnification), the position change of the role in the adjacent frames is compared, the motion direction and the speed vector (such as the speed of the role moving to the left is 5 pixels / frame) are generated, the scene visual complexity (such as the complexity of the battle scene is higher than that of the dialogue scene) is quantified through the indexes such as the number of objects and the frequency of light and shadow change, the cognitive load of the user to the game visual information is analyzed, such as the complex scene may cause the distraction of attention, and the data basis for dynamically adjusting the scene elements (such as simplifying the interface, reducing the moving speed of the enemy) is provided.

[0023] By means of the eye tracker, the focus point sequence (such as marking the hotspot if the continuous fixation on a certain area is more than 200 milliseconds) of the user's visual line on the screen is recorded, the change of the skin micro-current is collected in real time through the biological sensor, the tension or excitement degree (such as the sudden increase of the conductivity represents the stress reaction) of the user is mapped, the score (such as 1-5 star difficulty feedback) of the user to the current difficulty is collected through the pop-up window in the game or the external device (such as the handle key), the time difference (such as the skill release delay) of the user's operation instruction (such as the key) to the response of the game picture is calculated, the physiological signal, the subjective score and the operation delay are associated according to the time stamp, and the comprehensive feedback parameter (such as high conductivity + long delay + low score represents that the difficulty is too high) is generated, the emotional fluctuation (such as automatically reducing the strength of the enemy when nervous) of the user is captured in real time, and the experience delicacy is improved.

[0024] The multi-source data such as operation data, image frames, physiological signals are cut according to a fixed time window (such as 1 second), the data in the same window is ensured to correspond to the same game time through timestamp calibration, the key elements (such as enemies, props) in the image frames and their positions are identified, the operation intention (such as moving towards the enemy may indicate the attack intention) is inferred based on the space-time association of the user operation trajectory and the image elements, the features in each dimension (such as operation frequency, eye movement hotspot coordinates, scene complexity) are arranged in time sequence to form a two-dimensional feature matrix with timestamps, the time dislocation of the multi-source data is eliminated, the consistency of the subsequent model input is ensured, the image semantic features are automatically extracted through deep learning, the cost of artificial feature engineering is reduced, and the intention recognition accuracy is improved.

[0025] In a preferred embodiment of the application, in step S2, a difficulty adaptation index is generated by calculating the matching degree of the user operation trajectory and the preset reference mode according to the game performance image feature set, the difficulty adaptation index represents the matching deviation amount of the current game difficulty level and the user ability level, and includes: The behavior data in the game performance image feature set and the user feedback parameter set are standardized and pretreated, and the user ability associated features are extracted, including the average reaction time of the player, the operation accuracy, the task completion time deviation and the task success rate; Based on the preset reference reaction time and the ideal task completion time, the first normalized difference value of the average reaction time and the reference reaction time, and the second normalized difference value of the task completion time deviation and the ideal task completion time are calculated respectively; The operation accuracy and the task success rate are linearly weighted and fused to generate an operation efficiency evaluation component; The first normalized difference value, the second normalized difference value and the operation efficiency evaluation component are coupled in multiple dimensions through a dynamic weight distribution strategy to generate a comprehensive difficulty matching degree index, wherein the weight coefficients are adaptively adjusted according to the current physiological feedback data of the user; The comprehensive difficulty matching degree index is matched with the preset difficulty level threshold interval, if the index is lower than the first threshold, a difficulty degradation instruction is triggered, if the index is higher than the second threshold, a difficulty upgrade instruction is triggered, and the difficulty adaptation index representing the deviation amount of the current difficulty and the user ability is generated.

[0026] In the embodiments of the present application, the game performance image features (such as scene complexity, character moving speed) and user behavior data (such as operation frequency, reaction time) are normalized to eliminate dimensional differences (for example, converting reaction time from "milliseconds" to a proportional value in the [0, 1] interval); the average response time of the user to game stimuli within a unit time (such as the average time from multiple enemy appearances to attacks) is counted, the proportion of effective operation times to total operation times (such as the number of skill hits on enemies / the total number of skill releases) is calculated, the difference between the actual time taken by the user to complete a task and the system preset standard time (such as a standard time of 60 seconds, a user taking 75 seconds is a deviation of +25%) is compared, and the proportion of the number of successful completions of a user in a specific task (such as 8 out of 10 tasks, a success rate of 80%) is counted. The present application unifies the formats of multiple sources of data, improves the compatibility of model input, accurately quantifies the user ability dimension, and provides comparable benchmark indicators for difficulty assessment.

[0027] Normalized difference calculation: First normalized difference (reaction time deviation): compare the user's average reaction time with the system's preset "baseline reaction time" (such as the average reaction time threshold of a novice player) to calculate the difference proportion. For example, the baseline reaction time is 500 milliseconds, and the user's average reaction time is 600 milliseconds, then the difference is +20%, and the normalized value is 0.2.

[0028] Second normalized difference (task time deviation): compare the user's task completion time deviation with the "ideal task completion time" (such as the designed optimal completion time) to calculate the deviation proportion. For example, the ideal time is 50 seconds, and the user's deviation is +30% (actual time taken is 65 seconds), and the normalized value is 0.3.

[0029] The present application quantifies the gap between the user's reaction speed and task efficiency and the ideal level, intuitively reflects the ability short board, and provides clear numerical basis for difficulty adjustment (such as reducing the enemy appearance frequency if the reaction is slow).

[0030] Operation efficiency evaluation component generation process: Linear weighted fusion, the operation accuracy and task success rate are weighted and summed according to the preset weight (such as operation accuracy accounts for 60%, task success rate accounts for 40%). For example, the operation accuracy is 70%, and the task success rate is 80%, then the evaluation component is: 0.7x0.6+0.8x0.4=0.74 (i.e. 74% operation efficiency level), which comprehensively evaluates the stability of user operation and the quality of task completion, avoids the one-sidedness of single indicator, distinguishes between "fast reaction but many mistakes" and "slow reaction but high accuracy" player types, and supports differentiated difficulty adaptation.

[0031] Multi-dimensional coupling and dynamic weight distribution process: Dynamic weight adjustment, automatically adjust the weight of each index according to the user's real-time physiological data (such as eye movement hotspot concentration, skin conductivity). For example, when detecting that the user is nervous (conductivity increases), reduce the reaction time weight to avoid excessive misjudgment of ability level due to emotional fluctuations.

[0032] Multi-dimensional coupling calculation, superimpose the first normalized difference value, the second normalized difference value, and the operation performance evaluation component according to the dynamic weight to generate a comprehensive difficulty matching index. For example, the reaction time deviation weight is 0.3, the task time deviation weight is 0.3, and the operation performance weight is 0.4. The comprehensive index is: 0.2x0.3+0.3x0.3+0.74x0.4=0.476 (the smaller the value, the more suitable the difficulty), combined with physiological state dynamic adjustment of evaluation focus, improve the accuracy of difficulty judgment, multi-dimensional data cross verification, avoid misjudgment caused by accidental operation failure or scene complexity mutation.

[0033] Difficulty level mapping and instruction generation process: Threshold interval matching: preset difficulty adaptation threshold range (for example: index <0.3: difficulty is too high, need to downgrade; 0.3≤index≤0.7: difficulty is moderate; index>0.7: difficulty is too low, need to upgrade).

[0034] Instruction triggering, generate corresponding instructions according to the interval where the comprehensive index is located. For example, when the index is 0.2, trigger difficulty downgrade (such as reducing the number of enemies), when the index is 0.8, trigger difficulty upgrade (such as increasing the moving speed of the enemy), realize the automatic decision of difficulty adjustment, reduce the cost of artificial intervention, through the explicit threshold rule to ensure the logicality and explainability of difficulty change, avoid the experience fragmentation caused by random adjustment.

[0035] In a preferred embodiment of the present application, step S3, based on the difficulty adaptation index, dynamic weighted calculation, combined with user real-time physiological feedback data to generate dynamic difficulty correction coefficient, the dynamic difficulty correction coefficient contains scene complexity adjustment parameter and interactive response threshold adjustment parameter, including: Obtain the current game level, game base level and historical task completion time sequence of the player, extract the player ability growth trend characteristics; Based on the difficulty adaptation index, calculate the level adaptation component through a preset nonlinear conversion function of player level difference, the conversion function dynamically adjusts the gain amplitude according to the difference between the player level and the game base level; According to the skin conductivity change curve and eye movement tracking hotspot distribution in the user real-time physiological feedback data, extract the user stress load characteristics, dynamically adjust the weight proportion of the level adaptation component and the task time efficiency component in the dynamic weighted calculation; The difficulty adaptation index, the level adaptation component and the task time efficiency component are fused 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; According to the numerical interval of the comprehensive actual difficulty coefficient, scene complexity adjustment parameters and interactive response threshold adjustment parameters are generated, the scene complexity adjustment parameters are used to dynamically control the element density of the game interface and the generation frequency of dynamic objects, and the interactive response threshold adjustment parameters are used to adaptively correct the judgment fault tolerance range of the user operation input.

[0036] In the embodiment of the application, the current game level of the player (such as the character level, experience value), the game basic level (such as the default level of a novice) and the historical task completion time record (such as the completion time sequence of the past 10 tasks) are read.

[0037] Trend analysis: Level difference calculation, comparison of the current level and the basic level, judgment of the player's ability stage (such as entering the skilled stage when the level is 2 levels higher than the basic level); Time sequence analysis, judgment of the player's ability growth speed (such as accelerated growth, stable improvement or stagnation) through the fluctuation trend of the historical task completion time (such as the shortening of the task time for 3 consecutive times), identification of the player's ability development stage, and differentiation between the “novice adaptation period” and the “high-level challenge period”; long-term ability evolution basis is provided for difficulty adjustment to avoid short-term behavior misjudgment (such as accidental failure leading to difficulty misadjustment).

[0038] Level adaptation component calculation process: Nonlinear conversion function application, calculation of the level adaptation component through a preset function (such as an exponential function, a segmented function) according to the difference between the player's level and the basic level (such as +N levels or -N levels). For example, when the player's level is lower than the basic level, the function outputs a negative value (indicating that the difficulty needs to be reduced); when the player's level is higher than the basic level, the function outputs a positive value (indicating that the difficulty can be increased), and the greater the difference, the greater the absolute value of the component (such as the component of a level difference of +3 levels is twice that of a level difference of +1 level). The “stepwise” adjustment of the difficulty based on the level difference meets the player's growth expectation, and the nonlinear characteristic can avoid the loss of challenge for low-level players due to sudden difficulty reduction, or the lack of stimulation for high-level players due to linear adjustment. The formula of the preset function can be: ; wherein, represents the difference between the player's level and the basic level; is a parameter for adjusting the steepness of the curve (recommended value: 0.2-0.5); is a scaling coefficient that controls the overall adjustment amplitude (recommended value: 0.1-0.3); when the player's level is lower than the basic level , using a negative exponential function, output negative values, and the greater the level gap, the greater the absolute value of the negative value but the growth rate slows down; when the player's level is higher than the base level , using a positive exponential function, output positive values, and the greater the level gap, the positive value increases exponentially; The greater the difficulty, the steeper the curve, and the more obvious the difficulty increase for high-level players. The greater the overall adjustment range, the greater the overall adjustment range, and the more dramatic the change in game difficulty. The level difference increases by 1, and the adaptive component changes (such as from +1 to +2, the change is about 0.37, and from +2 to +3, the change is about 0.65), forming a step difficulty increase, and the component of +3 levels (1.476) is about 5.6 times the component of +1 levels (0.263), rather than a simple linear multiple relationship, which meets the requirement of "the greater the difference, the greater the absolute value of the component". This formula cleverly realizes the non-linear mapping between level and difficulty adjustment through an exponential function, ensuring that novice players have enough room to grow, and providing continuous challenges for high-level players.

[0039] Pressure load characteristics and weight dynamic adjustment process: Physiological index analysis: Skin conductivity analysis, detect conductivity peak frequency and duration, judge stress level (such as conductivity > 20% of the reference value and last for 5 minutes, marked as "high stress"), and calculate the proportion of the duration of the player's gaze in non-critical areas (such as blank interface, irrelevant props) to assess the degree of attention dispersion (such as a proportion > 40% indicating inattention).

[0040] Weight adjustment strategy: When the stress level increases, automatically reduce the weight of the task time efficiency component (such as from 50% to 30%), to avoid misjudgment of operation delay as insufficient ability due to nervousness, and when attention is dispersed, increase the weight of the level adaptation component (such as from 30% to 40%), to prioritize difficulty adjustment based on long-term player ability, real-time emotional state perception, dynamic balance between "ability assessment" and "experience protection", and avoid triggering false difficulty adjustment due to short-term physiological fluctuations (such as sudden interference leading to operation delay).

[0041] Comprehensive actual difficulty coefficient generation process: Task time efficiency component calculation, take the inverse of the player's average task completion time (such as an average time of 80 seconds, component of 1 / 80 ≈ 0.0125), to represent the task completion efficiency per unit time, and superimpose the difficulty adaptation index (reflecting the current matching deviation), the level adaptation component (reflecting long-term ability), and the task time efficiency component (reflecting immediate performance) according to the dynamic weight (adjusted proportion based on physiological data). For example: The difficulty adaptation index is 0.4 (moderate), the level adaptation component is +0.2 (higher than the basic level), the task time efficiency component is 0.01, and the weights are 50%, 30% and 20% respectively, so 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), the multi-dimensional data of “current matching degree”, “long-term growth” and “immediate efficiency” are fused to improve the comprehensiveness of difficulty evaluation, and the dynamic weight mechanism enables the system to switch between “conservative adjustment” (when the pressure is high) and “active adjustment” (when the state is stable).

[0042] Adjustment parameter decomposition and application process: Scene complexity adjustment parameter generation: When the comprehensive coefficient is greater than 0.5, the interface element density is reduced (such as hiding non-key buttons), and the dynamic object generation frequency is reduced (such as reducing the enemy refresh rate from 10 seconds / time to 15 seconds / time); when the comprehensive coefficient is less than 0.3, the prompt information prominence is increased (such as enlarging the task target icon), and the prop generation probability is increased (such as increasing the blood pack appearance probability from 20% to 30%).

[0043] Interaction response threshold adjustment parameter generation: If the average player operation delay is greater than 500 milliseconds, the operation judgment tolerance range is expanded (such as extending the skill release button long press judgment time from 200 milliseconds to 300 milliseconds), and if the operation accuracy is greater than 90%, the tolerance range is reduced (such as shortening the button response time, improving the operation precision requirement). Scene adjustment directly affects visual cognitive load, avoiding information overload or scarcity, and interaction threshold adapts to players of different operation levels. Novices can reduce frustration through loose judgment, and experts can obtain a sense of achievement through strict judgment.

[0044] In a preferred embodiment of the present application, step S4, input the user feedback parameter set into the image feature weight allocator, dynamically adjust the weight distribution of the image retrieval feature vector according to the user attention distribution data and the operation delay parameter, and generate an optimized scene element retrieval strategy, including: The eye movement tracking hotspot distribution data in the user feedback parameter set is subjected to time domain convolution processing to generate a user visual attention heat map, and the image feature vector of the high-frequency fixation area is extracted; 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; According to the visual attention heat map and the time sensitivity evaluation result, a feature weight dynamic allocation matrix is established to attenuate the weight of the interference scene element features with a user misrecognition rate higher than a threshold, and to enhance the weight of the key target features with a user correct recognition and response delay sensitivity; An incremental learning mechanism is used to optimize the adjusted feature weights online, and a weight update coefficient with time decay characteristics is generated based on 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, which includes scene element display and control instructions and target highlight rendering parameters, for real-time adjustment of the information density and interactive focus distribution of the game interface.

[0045] In the embodiments of the present application, the hotspot coordinate sequence recorded by eye tracking is arranged in time sequence (such as 100 fixation points recorded per second), and the fixation frequency in the time dimension is analyzed by sliding window convolution operation (such as 3 second window) to identify the area of continuous attention of the user (such as the skill bar in the upper left corner of the screen).

[0046] Heat map construction, mapping the pixel coordinates of high-frequency fixation areas to a two-dimensional heat map, with color depth representing the fixation time proportion (such as red area representing fixation time proportion > 60%), extracting the image feature vector of the area (such as RGB color distribution, edge contour complexity), accurately positioning the user's visual focus, and distinguishing between key operation areas (such as attack buttons) and secondary information areas, providing intuitive visual basis for scene element priority adjustment, and avoiding irrelevant elements from occupying attention resources.

[0047] Time sensitivity evaluation and priority classification process: Operation delay analysis, statistics of user's average response time to different scene elements (such as enemy, prop, prompt information) (such as 400 milliseconds from seeing the enemy to clicking the attack, 200 milliseconds from seeing the prop to picking up).

[0048] Sensitivity function application, priority level division according to delay threshold: High sensitivity: delay < 300 milliseconds (such as enemy movement in real-time combat), real-time priority retrieval is required; Medium sensitivity: 300 milliseconds ≤ delay ≤ 500 milliseconds (such as task prompt text), which can be processed slightly slower; Low sensitivity: delay > 500 milliseconds (such as background decoration elements), non-urgent retrieval object.

[0049] Ensure the retrieval priority of high-timeliness elements (such as dynamic enemies), improve the operation response speed, allocate computing resources according to the actual processing capacity of the user, and avoid excessive consumption of performance by low-priority elements.

[0050] Feature weight dynamic allocation and interference decay process: Misrecognition rate statistics, analyze the number of times of user click / interaction errors on scene elements (such as the frequency of mis-touching the advertisement button), mark elements with misrecognition rate > 20% as "interference items" (such as non-functional icons in the corner of the interface), multiply the feature vector weight of interference items by a decay coefficient (such as 0.5), reduce their matching priority in image retrieval, multiply the weight of correctly recognized and high sensitivity elements (such as task target arrows) by an enhancement coefficient (such as 1.5), and ensure priority rendering and retrieval. The present application reduces the misdirection of interface interference elements to users, reduces the operation failure rate, highlights the visual features of key targets (such as high-light color, enlarged size), and guides attention to quickly locate.

[0051] Incremental learning and weight update process: Historical data integration records the accuracy of the user's past N operations (such as the number of hits in the past 100 skill releases), generates a decay sequence in chronological order (such as the most recent operation weight proportion 10%, 7 days ago operation proportion 1%), and compares the current operation accuracy with the historical average after adjusting the feature weight each time. If it improves by > 5%, the weight adjustment is retained; if it drops by > 3%, it is rolled back to the previous version weight and the adjustment amplitude is reduced, avoiding misadjustment caused by single operation fluctuations. Smooth weight changes through historical data to continuously learn changes in user operation habits (such as attention shift from novice to master), and realize adaptive evolution of weight distribution.

[0052] Dynamic retrieval strategy generation process: Feature space mapping, similarity matching of the adjusted weight vector with the element features in the game scene database (such as enemy models, prop icons), generating a retrieval priority list (such as enemy feature matching degree > 90% when retrieving first), setting transparency > 70% or delayed loading for low-weight elements (such as background decorations), forcing high-weight elements (such as health bars) to remain visible, adding lighting effects or outline strokes to key target features (such as mission end flags), and dynamically adjusting parameters according to weight (such as weight increases by 0.1, brightness increases by 15%). Dynamically balance the information density of the interface to avoid "information bombing" or "buried key information", reduce the cognitive cost of user target search through visual interaction focus guidance, improve task completion efficiency, and respond to user attention changes in real time, such as automatically adjusting the retrieval strategy when switching from combat to puzzle solving, enhancing the smoothness of scene switching.

[0053] In a preferred embodiment of the present application, step S5, a multi-dimensional parameter fusioner is established to 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, including: The scene complexity adjustment parameter, the interactive response threshold adjustment parameter and the optimized scene element retrieval strategy are orthogonally decomposed, nonlinear coupling characteristics among multi-dimensional parameters are extracted, and a multi-modal characteristic space containing a spatial layout weight, a dynamic element density and an interactive fault tolerance threshold is established. The parameters in the multi-modal characteristic space are analyzed in cross-dimension association through an attention mechanism, and an implicit mapping relationship between the scene complexity parameter and the user attention distribution characteristic is identified. The interactive response threshold adjustment parameter in the dynamic difficulty correction coefficient is nonlinearly transformed by using a hyperbolic tangent function, and a sensitivity adjustment component matched with the current operation delay characteristic of the user is generated. The scene element display and hiding control instruction and the target highlight rendering parameter are spatially encoded, feature channel fusion is performed on the sensitivity adjustment component through a convolutional neural network, and an enhanced rendering instruction containing a dynamic light and shadow effect intensity and an interface element refresh rate is generated. Based on a preset game physics engine interface protocol, the fusion result of the multi-modal characteristic space and the enhanced rendering instruction are instruction set encapsulated, a final game difficulty control instruction set containing an enemy AI behavior mode, a resource refresh frequency and a physics engine parameter is generated, and real-time adaptation of game environment dynamic elements and user operation ability is realized.

[0054] In the embodiment of the application, scene complexity parameters (such as element density, dynamic object generation frequency), interactive response thresholds (such as operation fault tolerance range), retrieval strategies (such as element display and hiding control, highlight parameter) are dimensionally split (such as spatial dimension, time dimension, visual dimension), independent characteristic components (such as “information density” and “dynamic interference” in scene complexity) among the parameters are extracted through principal component analysis (PCA) and the like, each parameter component is mapped to a three-dimensional space (x-axis: spatial layout weight, y-axis: dynamic element density, z-axis: interactive fault tolerance threshold), and a multi-modal characteristic space is formed.

[0055] The application decouples the redundant association among the parameters, avoids difficulty imbalance caused by repeated adjustment, constructs a unified characteristic space for cross-dimension analysis, and finds, for example, a conflict combination of “high information density” and “low fault tolerance threshold”.

[0056] Based on the user eye movement hotspot distribution (such as gaze duration, saccade path), the attention weight of each parameter dimension (such as high attention to the battle area, giving "enemy AI complexity" higher weight) is calculated, and the hidden dependence between parameters (such as "interface element density increase" will lead to "operation accuracy decrease", but only when "time pressure is high") is found by association rule mining (such as Apriori algorithm), the invention realizes the "intelligent focus" of parameter adjustment, and preferentially optimizes the difficulty experience of the user attention area, and finds the parameter combination effect that is difficult for artificial to intuitively perceive (such as "low fault tolerance + high visual interference" will significantly reduce the game experience).

[0057] Nonlinear transformation and sensitivity adjustment process: Threshold nonlinear transformation, apply hyperbolic tangent function (tanh) to interaction response threshold parameter, map original threshold range (such as 0-100ms) to [-1, 1] interval, generate sensitivity adjustment component, for example: when the user operation delay average is 400ms, the function output is about 0.8, indicating that the system response sensitivity needs to be improved, according to the sensitivity component, the operation judgment logic is adjusted in real time (such as extending the skill release judgment time, expanding the click area), the invention smooths the difficulty adjustment curve, avoids the "difficulty cliff" caused by linear adjustment (such as when the operation delay changes from 300ms to 400ms, the sensitivity change amount changes from 0.2 to 0.5), and adapts to players with different operation habits (such as the response needs of touch screen users and keyboard and mouse users are different).

[0058] Space encoding and rendering instruction enhancement process: Convert the show and hide control instructions of scene elements (such as "hide background decoration") and highlight parameters (such as "task target brightness + 50%") into encoding vectors in three-dimensional space (such as [x, y, z] represent position, size, brightness respectively).

[0059] CNN feature fusion: Process the encoding vector through convolutional neural network (CNN), and perform channel-level fusion with sensitivity adjustment component (such as multiply "enemy highlight intensity" and "operation delay" feature map), output enhanced parameters (such as dynamic light intensity decreases with operation delay increases, interface refresh rate increases with attention dispersion), the invention realizes the coordinated change of visual performance and operation difficulty, for example, automatically enhance target prompt when operation error is frequent, use CNN to capture the complex relationship of spatial parameters (such as the visual interference effect of multiple highlight elements).

[0060] Instruction set packaging and physical engine adaptation process: Convert the fusion results of the multi-modal feature space (such as spatial layout weights, interaction thresholds) into parameters recognizable by the game physics engine (such as enemy movement speed, collision volume), encapsulate the instruction set according to the physical engine interface protocol (such as Unity's API), including: AI behavior mode adjustment (such as enemy attack frequency, pursuit distance); resource refresh mechanism (such as prop generation time interval, rarity); physical parameters (such as character jumping height, weapon recoil).

[0061] The present application realizes a complete closed loop from the "perception layer" (user behavior) to the "execution layer" (game engine), supports cross-platform deployment, and through a unified interface protocol adapts different game engines (such as Unity, Unreal), dynamic physical parameter adjustment provides more delicate difficulty changes (such as reducing character gravity when operation delay is high, reducing jumping difficulty).

[0062] In a preferred embodiment of the present application, step S6, real-time receiving user experience evaluation data of the adjusted game difficulty, iterative optimization through back propagation algorithm, forming a dynamic self-adaptive difficulty adjustment loop, including: Synchronize user real-time experience evaluation data through in-game log interface and biosensor, including subjective difficulty score adjustment amount, physiological stress index change rate and operation interruption frequency; Design a hierarchical convolutional neural network architecture containing a time convolutional layer and an attention mechanism module for multi-dimensional feature extraction of the execution effect of the final game difficulty control instruction set, generating an error feature vector containing enemy AI behavior deviation, resource refresh abnormality rate and interface interaction delay; Align the error feature vector and the user real-time experience evaluation data through a loss function, dynamically calculate the parameter gradient of the difficulty evaluation model and the feature offset of the image feature weight allocator; Jointly update the reference pattern matching parameters in the difficulty evaluation model and the attention response coefficient of the image feature weight allocator, and the optimization direction satisfies the operation trajectory fitting accuracy and the user physiological load balance constraint at the same time; According to the multi-modal data fusion gateway and the optimized parameter configuration, generate a difficulty control compensation instruction containing a dynamic balance factor, and form a spatio-temporal coordinated closed loop adjustment loop with the scene element retrieval strategy.

[0063] In the embodiment of the present application, the difficulty evaluation model construction process is as follows: Model architecture design: Multi-input branch design Behavior data branch: process time series data such as operation delay and accuracy, use LSTM network to capture operation mode changes; Physiological data branch: Analyze biological signals such as skin conductivity and heart rate variability, use 1D-CNN to extract physiological features; Scene data branch: Analyze in-game logs (such as enemy positions, item refreshes), model scene complexity through graph neural networks (GNN); Hierarchical feature fusion mechanism: Low-level feature layer: Each branch independently extracts features (such as behavior branch extracting hit interval features).

[0064] Intermediate fusion layer: Weighted fusion of multi-source features through attention mechanism (such as assigning higher weights to physiological features under high stress).

[0065] Decision output layer: Use fully connected layer to output difficulty adaptation index (0-1 interval, representing the matching degree of current difficulty and user ability).

[0066] Benchmark pattern library construction: User ability classification system: Based on operation speed, accuracy, strategy complexity and other dimensions, players are divided into 5 ability levels (such as novice, advanced, skilled, expert, master).

[0067] Establish benchmark behavior patterns for each level (such as expert player's reaction time threshold is 200ms, operation accuracy ≥85%).

[0068] Difficulty scene template: Design 20 typical game scene templates (such as BOSS battle, resource collection, puzzle level), preset difficulty parameter combinations for each scene template (such as puzzle scene prompt frequency, enemy number in battle scene) Dynamic matching algorithm: Behavior pattern matching engine, calculate the cosine similarity between user real-time behavior sequence and benchmark pattern, use dynamic time warping (DTW) algorithm to process time series data alignment problem (such as different players' skill release rhythm difference) Multi-dimensional scoring mechanism: Reaction ability score: Calculate based on the deviation of operation delay and benchmark threshold; Strategy depth score: Analyze the complexity of user path selection through decision tree; Stress tolerance score: Evaluate by combining physiological index fluctuation amplitude and duration.

[0069] Adaptive adjustment: Difficulty elastic interval setting: Set a ±15% difficulty adjustment elasticity interval for each skill level (e.g., a novice level can float within 85%-100% of the baseline difficulty), dynamically adjust the interval width based on user status (e.g., expand the lower limit to 70% when the user is tired), and simultaneously adjust the enemy attack frequency and damage value when detecting a decrease in operation accuracy. When the physiological stress indicator rises, prioritize reducing visual complexity rather than directly reducing the number of enemies.

[0070] Model training process: Mixed data training: Supervised learning phase: Use labeled player data (e.g., expert-level player operations at different difficulties) to train the base model.

[0071] Reinforcement learning phase: Optimize the difficulty adjustment strategy using the A3C algorithm, with user retention rate and experience satisfaction as the reward function.

[0072] Continuous evolution mechanism: Trigger incremental training every 100,000 new data collected, protect user privacy using federated learning technology, and complete model update calculations on local devices.

[0073] Verification and evaluation system: Offline verification indicators: Difficulty prediction accuracy: Evaluate the matching degree of the model's predicted difficulty level and the player's actual performance.

[0074] Adjustment response speed: Measure the delay time from user skill changes to the effectiveness of difficulty adjustment.

[0075] Online A / B testing: Randomly allocate 10% of users to the model experiment group, compare core indicators such as retention rate and daily active time, evaluate the effect of model iteration every week, eliminate poorly performing adjustment strategies, and achieve millimeter-level evaluation of player skills through multi-dimensional feature analysis. The whole process from identifying skill changes to completing difficulty adjustment has a delay of less than 10 seconds.

[0076] Collect user ratings adjustments (e.g., from 3 stars to 4 stars) through in-game pop-ups (e.g., "Is the current difficulty appropriate?") or handle button presses, synchronize biosensor data, calculate stress indicators such as skin conductance change rate (e.g., fluctuation amplitude per minute) and heart rate variability (HRV), and count the number of abnormal operation interruptions (e.g., the frequency of sudden stops lasting more than 5 seconds) within a fixed time. This invention cross- validates user experience through multi-source data (e.g., high ratings but high heart rate may indicate "too challenging"), captures experience fluctuations in real-time (e.g., rapidly triggers adjustments when operation interruptions surge), and avoids negative experience accumulation.

[0077] Hierarchical CNN and execution effect feature extraction process: Temporal convolutional layer processing, game instruction execution data (such as enemy movement trajectory, prop refresh time) are input into CNN in time sequence to extract time sequence features (such as periodic fluctuations in enemy attack frequency).

[0078] Attention mechanism application, feature map weighting highlights key error items (such as assigning higher weights when resource refresh anomaly rate > 15%), generating error vectors containing AI behavior deviation (such as enemy not moving along preset path), interaction delay (such as skill release animation stutter).

[0079] The invention captures subtle anomalies in instruction execution (such as difficulty imbalance caused by AI path deviation), attention mechanism focuses on core issues, reduces irrelevant data interference, and improves feature extraction efficiency.

[0080] Cross-modal alignment and parameter gradient calculation process: Loss function design, simultaneously calculating subjective score error (such as the difference between predicted score and actual score) and physiological indicator error (such as the deviation between predicted stress value and actual conductivity).

[0081] Cross-modal fusion, error feature vector and experience data are mapped to the same dimensional space (such as normalized values in the [0, 1] interval), and joint loss value is calculated.

[0082] Gradient backpropagation, through the backpropagation algorithm to calculate the adjustment direction of the benchmark matching parameters (such as reaction time threshold) in the difficulty assessment model and the attention coefficients (such as increasing the weight of "operation accuracy") in the weight distributor, the invention ensures that the model optimization meets the dual goals of "subjective experience" and "physiological load", cross-modal alignment avoids single data misdirection (such as relying only on subjective score may ignore users' hidden stress).

[0083] Joint iterative update and constraint optimization process: Parameter group update: Difficulty assessment model, adjust benchmark mode matching parameters (such as increasing the novice benchmark reaction time from 500ms to 600ms to adapt to slow reaction players); Weight distributor, update attention response coefficient (such as increasing the influence factor of "eye movement hotspot concentration" on scene element weight).

[0084] Constraint condition application: Operation trajectory fitting accuracy ≥ 85% (ensuring that difficulty adjustment does not deviate from the actual ability of the player), physiological load balance (such as skin conductance fluctuation controlled within the benchmark value ± 15% range).

[0085] The invention avoids model overfitting to short-term data (such as a one-time high score leading to difficulty rising), balances "ability matching" and "experience health", and prevents players from being exhausted due to excessive challenge.

[0086] Closed-loop regulation loop and compensation instruction generation process: Compensation instruction generation, according to the optimized parameters, generate dynamic balance factor (such as difficulty adaptation index buffer coefficient automatically adjust enemy strength when fluctuation).

[0087] Space-time coordination regulation: Time dimension, smooth adjustment of difficulty according to historical optimization period (such as every 10 minutes), avoid mutation; Spatial dimension, combined with scene element retrieval strategy (such as temporarily enhancing map marker weight in maze scene), realize "instruction-scene" linkage, the dynamic balance factor of the invention prevents over-regulation (such as difficulty rise and fall amplitude is limited to ± 20% of the current level), space-time coordination ensures that the adjustment conforms to the scene logic (such as the enemy AI is optimized in the battle scene, and the hint density is adjusted in the puzzle scene).

[0088] As shown in Figure 2 The embodiment of the present application also provides a dynamic game difficulty self-adaptive adjustment system based on user behavior feedback, comprising: An acquisition module is configured to collect multi-dimensional behavior data stream of a user in real time during game interaction, and extract features from the behavior data stream to obtain a game performance image feature set and a user feedback parameter set; A calculation module is configured to generate a difficulty adaptation index by calculating the matching degree of a user operation trajectory and a preset reference mode according to the game performance image feature set, wherein the difficulty adaptation index represents the matching deviation amount of the current game difficulty level and the user's ability level; and generate a dynamic difficulty correction coefficient by dynamically weighting calculation based on the difficulty adaptation index, in combination with real-time physiological feedback data of the user, wherein the dynamic difficulty correction coefficient includes a scene complexity adjustment parameter and an interactive response threshold adjustment parameter; A fusion module is configured to input the user feedback parameter set into an image feature weight distributor, dynamically adjust the weight distribution of the image retrieval feature vector according to user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy; and establish a multi-dimensional parameter fusion device to perform feature space mapping on the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy, and generate a final game difficulty control instruction set through nonlinear superposition; An adjustment module is configured to receive experience evaluation data of the user on the adjusted game difficulty in real time, and form a dynamic self-adaptive difficulty regulation loop through iterative optimization by a back propagation algorithm.

[0089] The above is the preferred embodiment of the present application. It should be noted that for ordinary skilled persons in the technical field, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A method for adaptively adjusting the difficulty of a dynamic game based on user behavior feedback, characterized in that: The method comprises: Step S1: collecting multi-dimensional behavior data streams of users in real time during game interaction, and performing feature extraction on the behavior data streams to obtain a game performance image feature set and a user feedback parameter set; Step S2, generating a difficulty adaptation index by calculating the degree of matching between the user operation trajectory and a preset reference pattern based on the game performance image feature set; Step S3, performing dynamic weighted calculation based on the difficulty adaptation index and generating a dynamic difficulty correction coefficient in combination with the user's real-time physiological feedback data; Step S4, inputting the user feedback parameter set into an image feature weight allocator, dynamically adjusting the weight distribution of the image retrieval feature vector according to the user attention distribution data and the operation delay parameter, and generating an optimized scene element retrieval strategy; Step S5, mapping the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy into a feature space, and generating a final game difficulty control instruction set through nonlinear superposition; Step S6: Receive user experience evaluation data on the adjusted game difficulty in real time, and iteratively optimize through the back-propagation algorithm to form a dynamic and adaptive difficulty adjustment loop.

2. The method for adaptively adjusting the difficulty of a dynamic game based on user behavior feedback according to claim 1, characterized in that: Step S1, collecting the user's multi-dimensional behavior data stream in real time during the game interaction process, and performing feature extraction on the behavior data stream to obtain a game performance image feature set and a user feedback parameter set, including: Collect user action sequence data through game client tracking, including action frequency, response time, task completion path, and error triggering events; Capturing a game performance image data stream through a graphics processor interface, performing multi-scale feature extraction on consecutive game image frames, and generating the game performance image feature set including interface element distribution density, character motion vectors, and scene complexity indicators; Synchronously collecting user feedback data through biosensors and an external interactive platform, including eye tracking hotspot distribution, skin conductivity change curves, and subjective difficulty rating inputs, and combining them with in-game operation delay parameters to generate the user feedback parameter set; The feature extraction process uses a sliding window mechanism to perform time domain alignment on multi-source heterogeneous data, and uses a convolutional neural network to perform dynamic object detection and behavioral intention classification on image frames to form a feature matrix with timestamp synchronization.

3. The method for adaptively adjusting the difficulty of a dynamic game 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 degree of matching between the user's operation trajectory and a preset reference pattern. The difficulty adaptation index represents the matching deviation between the current game difficulty level and the user's ability level, including: Performing standardized preprocessing on the behavioral data in the game performance image feature set and the user feedback parameter set 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 the ideal task completion time, respectively calculating a first normalized difference between the average reaction time and the benchmark reaction time, and a second normalized difference between the task completion time deviation and the ideal task completion time; Performing linear weighted fusion of the operation accuracy and the task success rate to generate an operation effectiveness evaluation component; The first normalized difference, the second normalized difference, and the operation efficiency evaluation component are multi-dimensionally coupled through a dynamic weight allocation strategy to generate a comprehensive difficulty matching index, wherein the weight coefficient is adaptively adjusted according to the user's current physiological feedback data; The comprehensive difficulty matching index is mapped to the preset difficulty level threshold interval. 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 to generate the difficulty adaptation index that represents the deviation between the current difficulty and the user's ability.

4. The method for adaptively adjusting the difficulty of a dynamic game based on user behavior feedback according to claim 3, characterized in that: Step S3, performing dynamic weighted calculation based on the difficulty adaptation index and combining the user's real-time physiological feedback data to generate a dynamic difficulty correction coefficient, wherein the dynamic difficulty correction coefficient includes a scene complexity adjustment parameter and an interactive response threshold adjustment parameter, including: Obtain the player's current game level, basic game level, and historical task completion time series to extract the player's ability growth trend characteristics; Based on the difficulty adaptation index, a level adaptation component is calculated using a preset player level difference nonlinear conversion function, wherein the conversion function dynamically adjusts the gain amplitude according to the difference between the player level and the game base level; Extracting user stress load characteristics based on the skin conductivity change curve and eye tracking hotspot distribution in the user's real-time physiological feedback data, and dynamically adjusting the weight ratio of the level adaptation component and the task time efficiency component in the dynamic weighted calculation; The difficulty adaptation index, level adaptation component and task time efficiency component are multi-source integrated to generate a comprehensive actual difficulty coefficient, wherein the task time efficiency component is represented by the inverse of the player's average task completion time; According to the numerical range of the comprehensive actual difficulty coefficient, the scene complexity adjustment parameter and the interactive response threshold adjustment parameter are decomposed and generated. The scene complexity adjustment parameter is used to dynamically control the density of game interface elements and the frequency of dynamic object generation. The interactive response threshold adjustment parameter is used to adaptively correct the judgment tolerance range of user operation input.

5. The method for adaptively adjusting the difficulty of a dynamic game 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 distribution of the image retrieval feature vector according to the user attention distribution data and the operation delay parameter, and generate an optimized scene element retrieval strategy, including: Performing time-domain convolution processing on the eye tracking hotspot distribution data in the user feedback parameter set to generate a user visual attention heat map and extract image feature vectors of high-frequency gaze areas; Constructing a time sensitivity evaluation function based on the operation delay parameter to prioritize the real-time response requirements of the scene element retrieval feature vector; Based on the visual attention heat map and time sensitivity evaluation results, a dynamic feature weight allocation matrix is ​​established to attenuate the weights of interfering scene element features whose user misrecognition rate is higher than a threshold, and to enhance the weights of key target features that are correctly recognized by the user and are sensitive to response delays; An incremental learning mechanism is used to perform online optimization on the adjusted feature weights, and historical operation accuracy data is combined to generate weight update coefficients with time decay characteristics. 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 game interface information density and interaction focus distribution in real time.

6. The method for adaptively adjusting the difficulty of a dynamic game based on user behavior feedback according to claim 5, characterized in that: Step S5: Establish a multi-dimensional parameter aggregator, perform feature space mapping on the dynamic difficulty correction coefficient and the optimized scene element retrieval strategy, and generate a final game difficulty control instruction set through nonlinear superposition, including: orthogonal decomposition is performed on the scene complexity adjustment parameter, the interaction response threshold adjustment parameter, and the optimized scene element retrieval strategy to extract nonlinear coupling characteristics between multidimensional parameters and establish a multimodal feature space including spatial layout weight, dynamic element density, and interaction fault tolerance threshold; Performing cross-dimensional correlation analysis on parameters in the multimodal feature space through an attention mechanism to identify implicit mapping relationships between scene complexity parameters and user attention distribution characteristics; Using a hyperbolic tangent function to perform a nonlinear transformation on the interactive response threshold adjustment parameter in the dynamic difficulty correction coefficient to generate a sensitivity adjustment component that matches the user's current operation delay characteristics; The scene element display and hide control instructions and the target highlight rendering parameters are spatially encoded, and feature channel fusion is performed with the sensitivity adjustment component through a convolutional neural network to generate enhanced rendering instructions that include 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 a final game difficulty control instruction set including enemy AI behavior patterns, resource refresh frequency and physics engine parameters, thereby achieving real-time adaptation of the dynamic elements of the game environment and the user's operating capabilities.

7. The method for adaptively adjusting the difficulty of a dynamic game based on user behavior feedback according to claim 6, characterized in that: Step S6: receiving user experience evaluation data on the adjusted game difficulty in real time, and iteratively optimizing through a back-propagation algorithm to form a dynamic and adaptive difficulty adjustment loop, including: The in-game log interface and biometric sensors are used to synchronously capture real-time user experience evaluation data, including the adjustment of subjective difficulty ratings, the rate of change of physiological stress indicators, and the frequency of operation interruptions. Design a hierarchical convolutional neural network architecture, including temporal convolution layers and attention mechanism modules, to extract multi-dimensional features of the execution effect of the final game difficulty control instruction set, generating error feature vectors that include enemy AI behavior deviations, resource refresh abnormality rates, and interface interaction delays; The error feature vector is cross-modally aligned with the user's real-time experience evaluation data through 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 benchmark pattern matching parameters in the difficulty assessment model and the attention response coefficient of the image feature weight allocator are jointly iteratively updated to optimize the direction while satisfying the constraints of operation trajectory fitting accuracy and user physiological load balance. According to the multimodal data fusion gateway and the optimized parameter configuration, difficulty control compensation instructions containing dynamic balance factors are generated, and a spatiotemporal coordinated closed-loop regulation 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 execute the method according to any one of claims 1 to 7, characterized in that: include: An acquisition module is used to collect the user's multi-dimensional behavior data stream in real time during the game interaction process, and perform feature extraction on the behavior data stream to obtain a game performance image feature set and a user feedback parameter set; a calculation module for generating a difficulty adaptation index based on the game performance image feature set by calculating the degree of match between the user's operation trajectory and a preset benchmark pattern, wherein the difficulty adaptation index represents the amount of match deviation between the current game difficulty level and the user's ability level; performing a dynamic weighted calculation based on the difficulty adaptation index and combining it with the user's real-time physiological feedback data to generate a dynamic difficulty correction coefficient, wherein the dynamic difficulty correction coefficient includes a scene complexity adjustment parameter and an interactive response threshold adjustment parameter; a fusion module, configured to input the user feedback parameter set into an image feature weight allocator, dynamically adjust the weight distribution of image retrieval feature vectors according to user attention distribution data and operation delay parameters, and generate an optimized scene element retrieval strategy; Establishing a multi-dimensional parameter aggregator, mapping the dynamic difficulty correction coefficient with the optimized scene element retrieval strategy into a feature space, and generating a final game difficulty control instruction set through nonlinear superposition; The adjustment module is used to receive real-time user experience evaluation data on the adjusted game difficulty, and iteratively optimize through the back-propagation 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, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

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

Images