Difficulty adjustment method, system and device based on player behavior data

Abstract

The application discloses a difficulty adjustment method and system based on player behavior data and a medium, relates to the technical field of mobile game difficulty adjustment, and is specifically aimed at the scenario of simulation management mobile games, improves the method on the basis of existing game difficulty adjustment technology, constructs a player ability model based on multi-dimensional player behavior data, comprehensively considers player strategy planning, execution efficiency and economic investment capacity in the generation of a difficulty parameter matrix, introduces a double-path or multi-path adjustment mechanism and combines a virtual game model to cope with player exploration behavior, considers player behavior randomness and exploration behavior in difficulty adjustment, iteratively updates parameters according to adjustment effects, and provides a personalized, smooth and forward-looking game difficulty adjustment scheme suitable for simulation management mobile games.

Description

Technical Field

[0001] This application relates to the field of difficulty adjustment technology based on player behavior data, and in particular to methods, systems and devices for difficulty adjustment based on player behavior data. Background Technology

[0002] In the process of adjusting the difficulty of simulation management mobile games, fixed difficulty parameters or single-dimensional behavioral feedback are usually used. However, due to significant individual differences among players in terms of strategy planning, execution efficiency, and willingness to invest economically, and because behavioral patterns are random and dynamic, it is difficult to accurately match the difficulty adjustment with the player's real-time ability. In the process of adjusting the difficulty in multiple iterations of the game, problems such as excessive difficulty fluctuations, abrupt adjustments, and lack of foresight will inevitably occur. This makes the existing difficulty adjustment methods have certain shortcomings in adaptability and flexibility. In the case of simulation management mobile games that focus on long-term retention and immersive experience, these shortcomings will seriously affect the overall performance of the game.

[0003] Difficulty adjustment in simulation management mobile games can cover various game scenarios and situations. Its main purpose is to quickly adjust the game difficulty to match player abilities when they perform actions such as placing functional items, synthesizing resources, and making in-app purchases, ensuring the game's challenge and enjoyment. In simulation management mobile games centered on functional item placement and resource management, if the difficulty adjustment is mismatched with player abilities, it will have a serious negative impact on player engagement, affecting game retention, enjoyment, and user stickiness, and may even affect player conversion rates and the normal operation of the game. Therefore, ensuring that the difficulty adjustment of simulation management mobile games is aligned with individual player abilities and achieves smooth and precise adjustments is crucial. Summary of the Invention

[0004] In order to at least overcome the above-mentioned shortcomings in the prior art, the purpose of this application is to provide a method, system and device for difficulty adjustment based on player behavior data to solve the above problems.

[0005] Firstly, this application provides a difficulty adjustment method based on player behavior data, including: In the current cycle, based on the player's first interaction, obtain the first action sequence, the second action sequence, and the third action sequence for the current cycle; Based on the sequence of the first to third lines, obtain the player's ability level model; Based on the player ability level model, the player's first ability deviation is obtained to obtain the difficulty parameter matrix; When entering the next cycle, according to the difficulty parameter matrix, the difficulty is adjusted, and the above steps are repeated to obtain the player's second interactive behavior; Based on the second interaction behavior, a second ability deviation is obtained, and in response to the second ability deviation being greater than the first ability deviation, the difficulty parameter matrix is ​​updated.

[0006] In one possible implementation, obtaining a second ability deviation based on the second interaction behavior, and updating the difficulty parameter matrix in response to the second ability deviation being greater than a first ability deviation, includes: In the current cycle, based on the player's first interaction, obtain the third action sequence for the current cycle; When entering the next cycle, the third action sequence for the next cycle is obtained based on the player's second interaction action; Based on the third row sequence of the current period and the third row sequence of the next period, obtain the change in the third row sequence. When the change in the third row sequence exceeds a preset threshold, the difficulty parameter matrix is ​​updated based on the adjustment strategy function.

[0007] In one possible implementation, obtaining the player's first ability deviation based on the player ability level model includes: Obtain the target capability vector in the current period from the preset target challenge curve; Based on the player ability level model and the target ability vector, calculate the Euclidean distance to obtain the first ability deviation. Based on the first ability deviation, a difficulty parameter matrix is ​​obtained.

[0008] In one possible implementation, obtaining the player's first ability deviation based on the player ability level model includes: Based on the second line sequence, obtain the player cluster of active players; Based on the player ability level model of the active player cluster, a socially comfortable difficulty range is obtained; Based on the social comfort difficulty range and the player ability level model, obtain the player's first ability deviation. Based on the first ability deviation, obtain the difficulty parameter matrix; The step of obtaining the socially comfortable difficulty range based on the player ability level model of the active player cluster includes: Obtain player ability level model data for active players; Active players are divided into at least one player cluster through cluster analysis; For each player cluster, calculate the statistical distribution of the player ability level vector in the current period, and take the range of mean plus or minus standard deviation as the social comfort difficulty range of the player cluster.

[0009] In one possible implementation, based on the player ability level model, a player's first ability deviation is obtained to acquire a difficulty parameter matrix, and the method further includes: Based on the player ability level model, at least one path is selected to obtain the player's first ability deviation. Based on the first ability deviation, obtain the difficulty parameter matrix; When the deviation of the first ability is less than or equal to the preset threshold, the difficulty parameter matrix is ​​obtained based on the preset adjustment strategy function; When the deviation of the first ability exceeds a preset threshold, a difficulty parameter matrix is ​​obtained based on a preset mapping library of in-game events and parameter adjustments.

[0010] In one possible implementation, obtaining the player's first ability deviation based on the player ability level model includes: In the current cycle, based on the player ability level model and a preset baseline difficulty model and / or virtual game model, the player's first ability deviation is obtained, including: Based on the player ability level model, and based on at least two preset baseline difficulty models, the corresponding initial ability deviation is obtained; When the initial ability deviations calculated by different benchmark difficulty models indicate different directions of difficulty adjustment, the game triggering condition is determined to be met; or, When the path conflict intensity value calculated based on the initial capability deviation exceeds the preset conflict threshold, it is determined that the game triggering condition is met. In response to the satisfaction of any of the game triggering conditions, the virtual game model is activated to generate the first capability deviation for the current period; Otherwise, the initial capability deviation generated by the benchmark difficulty model will be used as the first capability deviation for the current cycle; The baseline difficulty model includes at least paths one and two, and the virtual game model includes at least path three, including: The first path is configured as follows: based on the player's ability level model and a preset target challenge curve, obtain the first ability deviation. The second path is configured as follows: based on the player's ability level model and the socially comfortable difficulty range, obtain the first ability deviation. The third path is configured to: obtain the first ability deviation based on the player's ability level model and a virtual game algorithm.

[0011] In one possible implementation, path three is configured to: obtain a first ability deviation based on a player ability level model and a virtual game-based algorithm, including: Obtain at least two initial ability deviations calculated from the baseline difficulty model; When a conflict in adjustment direction or adjustment intensity is detected between the initial capability deviations, it is determined that the game triggering condition is met. In response to the fulfillment of the game triggering condition, a virtual game model is initiated to obtain a first capability deviation corrected for game equilibrium. This first capability deviation includes: Based on the counterfactual reasoning framework, a virtual opponent model is obtained. The virtual opponent model is used to simulate the player's response behavior to the system's difficulty strategy. Its parameters are updated by minimizing the KL divergence between the predicted player behavior distribution and the actual player behavior distribution. Using long-term player engagement metrics as the optimization objective, we solve for the Nash equilibrium point in the system difficulty strategy space to obtain an balanced difficulty strategy. Based on the equilibrium difficulty strategy and the player's current ability level, the first ability deviation corrected by game equilibrium is obtained; The first capability deviation, corrected by game equilibrium, is taken as the first capability deviation.

[0012] In one possible implementation, a player ability level model is obtained based on the first behavior sequence and the second behavior sequence, including: Based on the player's first interaction, a third action sequence is also obtained; Based on the first to third behavior sequences, first to third behavior feature vectors representing strategic planning ability, execution efficiency ability, and willingness to pay are obtained respectively. Based on the first behavioral feature vector, a first ability score representing the player's strategic planning ability is obtained; The first behavioral feature vector is linearly combined with a preset first weight vector to obtain the first ability score, wherein the first weight vector is obtained by dimensionality reduction calculation based on historical player behavior data. Based on the second behavioral feature vector, a second ability score representing the player's execution efficiency is obtained; The components of the second behavioral feature vector are normalized and then weighted and fused based on preset weights to obtain the second capability score. Based on the third behavior feature vector, a third ability score representing the player's willingness to invest economically is obtained; The third row of the feature vector is subjected to a logarithmic transformation at least one component, and then weighted and fused with the remaining components to obtain the third capability score. A player ability level model is obtained based on the first ability score, the second ability score, and the third ability score.

[0013] Secondly, this application provides a difficulty adjustment system based on player behavior data, including a behavior acquisition unit, a player level calculation unit, and a difficulty adjustment unit connected by sequential electrical connections; The behavior acquisition unit is configured to: in the current cycle, based on the player's first interaction behavior, obtain the first behavior sequence and the second behavior sequence of the current cycle; The player level calculation unit is configured to: obtain a player ability level model based on the first and second behavior sequences; The difficulty adjustment unit is configured to: obtain the player's first ability deviation degree according to the player ability level model, so as to obtain a difficulty parameter matrix; The behavior acquisition unit is also configured to: when entering the next cycle, perform difficulty adjustment according to the difficulty parameter matrix, and repeat the above steps to acquire the player's second interactive behavior; The difficulty adjustment unit is further configured to: obtain a second ability deviation degree based on the second interaction behavior, and update the difficulty parameter matrix in response to the second ability deviation degree being greater than the first ability deviation degree.

[0014] Thirdly, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of a difficulty adjustment method based on player behavior data.

[0015] In summary, the beneficial effects that this application can achieve are: This application proposes a multi-dimensional player ability modeling method that extracts three types of behavioral sequences to construct a three-dimensional ability vector, adapting to scenarios with no paying players and new players. This addresses the problem of one-sided characterization in traditional modeling and achieves precise quantification of player abilities. The proposed multi-path deviation calculation mechanism, through dynamic path selection and the introduction of a virtual game model, solves the problem of poor adaptability of single paths, achieving precise difficulty matching. The proposed adjustment and iterative optimization strategy, by monitoring the adjustment effect and updating parameters based on paying behavior, solves the problem of no feedback in adjustment, achieving smooth and dynamic adjustment of difficulty. The methods in this application represent significant technological advancements and beneficial effects in improving the difficulty adaptability and player experience of simulation management mobile games. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the method steps in Embodiment 1 of this application; Figure 2 This is a schematic diagram of the method steps in Embodiment 2 of this application; Figure 3This is a flowchart illustrating the parallel selection of two paths in an embodiment of this application. Figure 4 This is a flowchart illustrating the three path selection methods in an embodiment of this application. Figure 5 This is a schematic diagram of the system structure according to an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0018] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0019] Example 1 Please refer to the following: Figure 1 , Figure 1 This is a schematic diagram of the steps of the difficulty adjustment method based on player behavior data provided in the embodiment of the present invention. Further, the difficulty adjustment method based on player behavior data may specifically include the contents described in steps S1-S5.

[0020] Step S1: In the current cycle, based on the player's first interaction behavior, obtain the first action sequence of resource management and the second action sequence of progress advancement for the current cycle; Step S2: Obtain the player ability level model based on the first and second line sequences; Step S3: Based on the player's ability level model, obtain the player's first ability deviation to obtain the difficulty parameter matrix; Step S4: When entering the next cycle, adjust the difficulty according to the difficulty parameter matrix and repeat the above steps to obtain the player's second interactive behavior; Step S5: Based on the second interaction behavior, obtain the second ability deviation. In response to the second ability deviation being greater than the first ability deviation, update the difficulty parameter matrix.

[0021] In the implementation of this application embodiment, all player actions in the game are continuously monitored and recorded within the current game cycle. These actions constitute the player's first interactive behavior. For subsequent analysis, specific behavioral patterns can be extracted from these original interactive behaviors to form a first behavior sequence and a second behavior sequence. For example, rules can be set to categorize and record a series of actions such as clicking the "Build" button, dragging a building model to a designated location, and confirming a building upgrade in the game interface as the first behavior sequence. This sequence reflects the player's strategic placement of functional items. Simultaneously, the player's actions such as clicking the "Craft" button, selecting crafting materials, confirming the crafting operation, and the player's character's stamina decreasing due to action or recovering due to rest can be categorized and recorded as the second behavior sequence. This sequence reflects the player's resource management and execution efficiency.

[0022] After obtaining the first and second behavior sequences for the current period, this behavioral data needs to be processed to construct a player ability level model. One approach is to count the behavioral events in the first behavior sequence, such as the total number of functional items placed by the player within the period and the average placement time, forming a set of indicators reflecting strategic planning ability. Simultaneously, the behavioral events in the second behavior sequence are counted, such as the number of times the player completes resource synthesis within the period and the frequency of stamina consumption and recovery, forming a set of indicators reflecting execution efficiency. Subsequently, a comprehensive player ability score can be obtained based on these indicators, thus yielding the player ability level model.

[0023] After obtaining the player's ability level model, it's necessary to assess the gap between the player's current ability and the game's design goals, i.e., to obtain the first ability deviation. One approach is to preset an ideal target value for the player's ability or a comfortable ability range. The ability score output by the player's ability level model is directly compared with this target value, and the difference or percentage difference is calculated as the first ability deviation. For example, if the player's ability score is higher than the target value, the deviation is positive; if it is lower than the target value, the deviation is negative. Subsequently, based on this first ability deviation, a preset mapping table or a simple linear function can be consulted to generate a difficulty parameter matrix.

[0024] Once the difficulty parameter matrix is ​​determined, the game environment and challenges will be adjusted based on the parameters in the matrix when the game enters the next cycle. Under the new difficulty setting, all player actions in the game will continue to be monitored, and these actions will be recorded as the player's secondary interactions. These secondary interactions will serve as input data for the next round of difficulty adjustments, enabling continuous and dynamic difficulty adjustments.

[0025] At the end of the next cycle, based on the player's second interaction, a method similar to obtaining the first ability deviation will be used to recalculate the player's ability level and obtain a second ability deviation. For example, the player's functional item placement behavior, resource synthesis operations, and stamina status changes during the new cycle can be statistically analyzed again, and a new ability score can be calculated based on this data. This score is then compared with the preset target ability value to obtain the second ability deviation. Subsequently, the second ability deviation will be compared with the previous first ability deviation. If the second ability deviation is found to be greater than the first ability deviation, it may mean that the player has encountered a greater challenge at the new difficulty level, or that the gap between their ability and the target has widened. In this case, the difficulty parameter matrix will be updated responsively. For example, by fine-tuning certain parameters in the difficulty parameter matrix through preset rules, it is intended to better adapt to the player's current ability status.

[0026] Therefore, this application collects data on players' functional item placement behavior, resource synthesis operations, and stamina status changes through multiple iterative cycles, and constructs a player ability level model based on this data to assess players' strategic planning ability and execution efficiency. By quantifying the deviation between player ability and goal and dynamically generating a difficulty parameter matrix, it achieves forward-looking adjustment of the difficulty of simulation management mobile games. In subsequent cycles, player behavior is continuously monitored, and the difficulty parameters are adaptively updated according to the changing trend of ability deviation. This effectively solves the problems of inaccurate difficulty adjustment, large fluctuations, and lack of foresight in traditional methods, ensuring that the game difficulty accurately matches the player's real-time ability, and improving the player's long-term retention and immersive experience.

[0027] Example 2 Based on Example 1, please refer to the following: Figure 2 , Figure 2 This is a schematic diagram of the steps of the difficulty adjustment method based on player behavior data provided in the embodiment of the present invention. Further, the difficulty adjustment method based on player behavior data may specifically include the following.

[0028] This invention provides a difficulty adjustment method based on player behavior data. This method constructs a multi-dimensional player ability model and, depending on the actual situation, can introduce dual-path or multi-path adjustment mechanisms. In simulation management mobile games, this allows for more player-centric, personalized, smooth, and forward-looking adjustments to the difficulty parameters for the next game cycle. The technical solution of this invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] In a specific implementation scenario, this method is deployed in a simulation management game, primarily focused on the placement of functional items and resource management. The game is divided into multiple consecutive game cycles. Within each cycle, players can perform actions such as placing functional items, synthesizing resources, and making purchases within a pre-defined, limited space. At the end of each cycle, player behavior data is collected, and a difficulty parameter matrix for the next cycle is generated accordingly, thereby configuring and adjusting the game's difficulty.

[0030] Step S1: In the current cycle, based on the player's first interaction behavior, obtain the first behavior sequence, the second behavior sequence, and the third behavior sequence of the current cycle.

[0031] When implementing the embodiments of this application, the player's first interaction behavior is first obtained within the current cycle; Then, the player's functional item placement behavior data is obtained through the space layout module to generate the first behavior sequence; Simultaneously, the resource management module acquires the player's resource synthesis operations and stamina status change data to generate a second action sequence. Among them, the resource synthesis operation refers to the player's synthesis processing of game resources based on the preset synthesis recipe, which is a conversion operation that transforms at least two game resources and stamina into the target game resource. The stamina status change data refers to the numerical change of a parameter named stamina in the game. Stamina is a consumable game resource that is used to limit the frequency of continuous operations of the player. At the same time, the system obtains players' paid transaction data through the paid behavior module and generates a third behavior sequence. The first to third rows are stored in chronological order.

[0032] In this embodiment, the first interactive behavior is obtained by three independent but logically related behavior modules, namely the space layout module, the resource management module, and the payment behavior module. The first interactive behavior of the player's original operation is extracted and stored in the local cache of the corresponding module. In order to save costs, this embodiment is processed at the end of the cycle.

[0033] Although the behaviors acquired by these three modules are essentially independent, they are logically related because they are from the same player and are extracted according to a time sequence.

[0034] In one possible implementation, the game cycle is preset to a fixed period of one day. The game cycle begins at the start of a level and ends when a level is completed. The next level marks the start of a new cycle.

[0035] In practice, the cycle is usually divided according to levels in the early stages and according to a fixed duration in the later stages. However, this embodiment divides the cycle according to levels, which will be explained in detail here.

[0036] The space arrangement module is used to acquire and store players' placement behavior of functional props; among them, functional props refer to configurable game resources that can be placed in different game scenes to change the state of the game scene or provide gameplay benefits; placement behavior refers to the positioning and configuration operations of functional props in different game scenes, including confirmation placement operation and undo or redo operation.

[0037] Whenever a player drags a functional item to the target coordinates and confirms the placement operation, record the type of the event, the timestamp of the occurrence, the target coordinates, and the functional item type identifier; If a player performs an undo or redo operation, the type of the event and the timestamp of the event are recorded simultaneously; After each spatial layout adjustment is completed (meaning placement and undo / redo operations are finished), the space utilization rate is obtained. Space utilization rate is based on the ratio of the number of grid cells occupied by functional items to the total usable area of ​​the currently controllable game scene, outputting a value between 0 and 1. .

[0038] All the behaviors obtained by the spatial arrangement module from the first interaction behavior constitute the first behavior sequence.

[0039] The resource management module acquires and stores the player's resource crafting trajectory and stamina status changes during crafting operations. Stamina is a game parameter that decreases with each player action and recovers to a preset value at the end of each game cycle.

[0040] The crafting operation is a game operation that responds to the player's crafting command, selects game resources, consumes a set amount of stamina, and executes the crafting logic to produce new game resources.

[0041] Stamina is a consumable game resource used to limit the frequency of consecutive operations. Stamina is a consumable game resource that recovers naturally over time and is consumed in operations such as crafting and acquiring resources. Stamina status change data is a sequence of stamina consumption and recovery, which can represent the change of stamina reserves over time. It also represents the amount of stamina change triggered by crafting operations and the use of recovery items, along with their corresponding timestamps.

[0042] Whenever a player selects a set of raw materials and triggers a synthesis operation, the system records the trigger time, the identifier of the selected material combination, and the synthesis result status, which includes only success or failure. Simultaneously, the system continuously monitors the player's stamina, recording the corresponding stamina change and the event time when stamina decreases due to synthesis or increases due to the natural recovery mechanism.

[0043] If a player clicks on a recovery item to immediately replenish their stamina, the frequency and timing of this click event are recorded.

[0044] The resource management module obtains all the above behaviors from the first interaction behavior to form the second behavior sequence.

[0045] The paid behavior module obtains all paid interaction behaviors of the player in the current period, including: the time and amount of currency recharge events, the time, item ID and price of purchase behavior in the item shop. All paid related events are arranged in chronological order. All the above behaviors obtained by the paid behavior module from the first interaction behavior constitute the third behavior sequence.

[0046] Step S2: Obtain the player ability level model based on the first row sequence to the second row sequence. In this embodiment, the player level can be obtained in two ways.

[0047] When implementing the embodiments of this application, the first row feature vector is obtained based on the first row sequence, including: the number of times functional props are adjusted per unit time, the average time interval between adjacent adjustments, the average space utilization rate, the proportion of undo and redo operations, and the usage density of functional props; The second-line feature vector is obtained based on the second-line sequence, including: synthesis success rate, raw material combination diversity index, physical exertion rate, accelerated recovery frequency, and physical recovery response delay. The third-line feature vector is obtained based on the third-line sequence, including: total payment amount within the period, highest single payment amount, payment frequency, and the correlation strength between payment behavior and key progress nodes. Based on the first to third rows of feature vectors, a three-dimensional ability vector is constructed as the player's ability level model. Therefore, the player's ability level model includes the strategic planning ability dimension, the execution efficiency ability dimension, and the willingness to invest economic resources dimension.

[0048] In one possible implementation, a player ability level model is obtained based on a first action sequence and a second action sequence, including: The first ability score is obtained based on the first row as a feature vector; The first ability score is obtained by linearly combining the first behavior feature vector with a preset first weight vector, wherein the first weight vector is obtained by dimensionality reduction calculation based on historical player behavior data. The second ability score is obtained based on the feature vector of the second row; The components of the second row feature vector are normalized and then weighted and fused based on preset weights to obtain the second capability score. A player's ability level model is obtained based on the first ability score and the second ability score.

[0049] In one possible implementation, a player ability level model is obtained based on a first action sequence and a second action sequence, including: Based on the player's first interaction, a third action sequence is also obtained; Based on the first to third line sequences, the first to third line feature vectors representing strategic planning ability, execution efficiency ability, and willingness to pay are obtained respectively. Based on the first row feature vector, obtain the first ability score that represents the player's strategic planning ability; The first ability score is obtained by linearly combining the first behavior feature vector with a preset first weight vector, wherein the first weight vector is obtained by dimensionality reduction calculation based on historical player behavior data. Based on the second behavior feature vector, a second ability score representing the player's execution efficiency is obtained; The components of the second row feature vector are normalized and then weighted and fused based on preset weights to obtain the second capability score. Based on the third-row feature vector, a third ability score representing the player's willingness to invest economically is obtained; The third row of the feature vector is logarithmically transformed at least one component and then weighted and fused with the remaining components to obtain the third ability score. A player's ability level model is obtained based on the first ability score, the second ability score, and the third ability score.

[0050] In the implementation of this application embodiment, the data of the behavior is transformed into a specific feature vector according to the sequence of the first to third behaviors.

[0051] A first-beginning feature vector is constructed based on the first-beginning sequence to characterize strategy planning capability, including five components: the number of times functional items are adjusted per unit time. Average adjacent adjustment time interval Average space utilization The percentage of undo and redo operations and the density of use of functional props .

[0052] Among them, the number of times functional props can be adjusted. The calculation formula is: ,in, This represents the total number of furniture drags during the current period. The current cycle duration is specified. For a fixed cycle, it is a preset day. Alternatively, if the cycle ends due to a level, the duration is from the time you enter the current level to the end time of the current level, in hours. Average adjacent adjustment time interval Defined as ,in This is the timestamp of the i-th layout adjustment. To effectively adjust the number of times, if ,but It is assigned the preset maximum value; Average space utilization pass The calculation involves obtaining a space utilization value after each spatial layout adjustment. , This represents the total number of snapshots taken for spatial layout. percentage of undo / redo operations Depend on It is concluded that, among them The number of times an operation can be undone or redone. This represents the total number of operations, including dragging, rotating, deleting, etc. Defined as ,in The number of high-level furniture items with a grade no lower than a preset threshold. This represents the total number of furniture items. This feature vector comprehensively characterizes the player's strategic depth, operational stability, and resource optimization tendencies in spatial planning.

[0053] Simultaneously, a second-line feature vector is constructed based on the second-line sequence to characterize execution efficiency, comprising five components: synthesis success rate. Raw material combination diversity index rate of physical exertion Accelerate recovery frequency and delayed physical recovery response constitute.

[0054] in, , The number of successful synthesis attempts. This represents the total number of attempts. ,in The formula represents the number of times the k-th ingredient combination is used, where K is the total number of combinations. This formula uses a normalized variant of the Gini-Simpson diversity index to quantify the breadth of a player's exploration of synthesis strategies. ,in The total stamina consumption within the cycle is calculated by summing the stamina costs of all synthesis operations. ,in To increase the number of clicks on the restore button; ,in This refers to the time point when physical strength falls below a preset threshold (which can be set to a typical threshold, such as 30%) for the mth time. This represents the time point when the acceleration button is clicked for the first time; if it is not clicked, the sample is discarded. This feature vector reflects the player's overall performance in terms of task execution efficiency, risk-taking willingness, and resource allocation rhythm.

[0055] At this point, a player's ability level model can be obtained based on the first ability score and the second ability score.

[0056] Furthermore, in this embodiment, economic factors are also considered. Therefore, the third behavior sequence obtained from the user's first interaction behavior is used to jointly construct the player's ability level model. This embodiment also constructs a third behavior feature vector based on the third behavior sequence to represent the willingness to pay, including the total amount paid within the period. Maximum single payment amount Frequency of payment And the strength of the correlation between payment behavior and key progress milestones. .in, The total amount of all top-ups and purchases; This represents the highest amount in a single transaction. ,in Total number of paid events; The correlation coefficient was obtained by calculating the Pearson correlation coefficient between the time series of paid events and the time series of new content unlocking events.

[0057] set up For a collection of paid time points, For a set of new level or feature unlock timestamps, map both onto the same timeline and construct a binary indicator function. and Let and represent whether a payment or unlocking event occurs at time t, respectively. Then, calculate the Pearson correlation coefficient between these two discrete signals: ; Where T is the total duration of the cycle, expressed in days or in the time taken to complete a level. and These are the means of the two signals. This indicator effectively measures whether player spending behavior is concentrated during key bottleneck periods in game progress, thus reflecting their willingness to invest in content advancement.

[0058] After obtaining the three behavioral feature vectors above, a model that can reflect the player's skill level can be obtained based on the first to third behavioral feature vectors.

[0059] In practice, the player ability level model is a model built based on multi-period historical data. Each period is represented by a current period player ability level vector. Based on the player ability level vector of each period, the player ability level model is constructed.

[0060] After obtaining the above three behavioral feature vectors, the player's ability level vector for the current period is calculated to construct a player ability level model.

[0061] Player Ability Level Vector Essentially, it's a three-dimensional vector used in each cycle. ,in That is, each dimension corresponds to a behavioral feature vector, which is compressed and calculated to obtain a capability score, thus forming a three-dimensional vector. .

[0062] First ability score The ability to represent strategy planning is obtained by performing dimensionality reduction calculations using principal component analysis on the first row of feature vectors.

[0063] Beforehand, a large amount of historical layout behavior data of players is collected offline to build a training set, and the first principal component direction is extracted through principal component analysis to obtain the first weight vector. .

[0064] During the online phase, the first action of the current player is used as a feature vector. With the first weight vector Perform inner product operation to obtain the first ability score. The higher the value, the stronger the player's strategic thinking and resource utilization efficiency in spatial planning.

[0065] Second ability score The performance efficiency is represented by a normalized weighted fusion of the second row of feature vectors. The five components of the second row of feature vectors are each normalized to the [0,1] interval, and then preset weights are applied. Weighted summation can be expressed as: ; in, For the i-th component after normalization, the weight is... The values ​​are set by the operations team based on the game's balance requirements, such as giving higher weight to the success rate of synthesis and the rate of stamina consumption. These values ​​are intended to comprehensively reflect the player's stability and resource utilization efficiency during task execution.

[0066] Third Ability Score The representation of economic willingness to invest is mainly to avoid excessive disturbance to the model caused by high single payments, while retaining the sensitivity to players' continuous small-amount payment behavior. Therefore, it is formed by the third behavior feature vector after logarithmic transformation and standardization.

[0067] Third dimension , can be represented as: ; in For example, take a fixed coefficient. In fact, to consider the dominant economic role of the amount and highlight the leading role of the total payment amount, therefore... Logarithmic transformation is performed, which also takes into account the correlation between payment frequency and timing.

[0068] In this step, the existence of players who have no intention of making economic investment is also taken into consideration during actual implementation. In this case, only the first and second behavior sequences are obtained to obtain the player ability level model.

[0069] In this step, the situation of the player entering the current game for the first time is also taken into consideration during actual implementation.

[0070] When a player first enters the current game, it also includes: Retrieve cross-game player strategy profiles generated by this player in other games on the same platform; By using a transfer learning model, cross-game player strategy profiles are mapped to initial prior values ​​of the current game's player ability level model; Based on the initial prior values ​​and the first cycle of the current game, data interaction is performed to obtain the initial player ability level vector, so as to construct a player ability level model.

[0071] Step S3: Based on the player's ability level model, obtain the player's first ability deviation to obtain the difficulty parameter matrix.

[0072] The first capability deviation can be obtained through two alternative or parallel computational paths; Path 1 is based on the calculation of the absolute deviation from the target challenge curve; Path 2 is based on the calculation of relative deviation from the social comfort difficulty range.

[0073] You can choose at least one of the paths, or you can choose a path based on preset conditions.

[0074] In one possible implementation, the player's first ability deviation is obtained based on the player's ability level model, including: Obtain the target capability vector in the current period from the preset target challenge curve; Calculate the Euclidean distance between the player's ability level model and the target ability vector to obtain the first ability deviation. Based on the deviation of the first ability, obtain the difficulty parameter matrix.

[0075] In one possible implementation, the player's first ability deviation is obtained based on the player's ability level model, including: Based on the second line sequence, obtain the player cluster of active players; Based on the player ability level model of active player clusters, a socially comfortable difficulty range is obtained; Based on the social comfort difficulty range and the player's ability level model, the player's first ability deviation is obtained; Based on the deviation from the first ability, obtain the difficulty parameter matrix; Based on a player ability level model of active player clusters, a socially comfortable difficulty range is obtained, including: Obtain player ability level model data for active players; Active players are divided into at least one player cluster through cluster analysis; For each player cluster, calculate the statistical distribution of the player ability level vector in the current period, and take the range of mean plus or minus standard deviation as the social comfort difficulty range of the player cluster.

[0076] When implementing the embodiments of this application, please refer to the following: Figure 3 , Figure 3 The flowchart for parallel selection of two paths provided in the embodiments of the present invention allows for the selection of at least one or more paths based on actual circumstances.

[0077] Path 1 and Path 2 can run in parallel or be interchanged, or they can run in parallel.

[0078] If both path one and path two are chosen to run in parallel, the rules for selecting the computation path include: If a player's registration period is less than the preset number of days, then Path 1 will be selected. Path 1 calculates the deviation of the first ability based on the design team's preset target challenge curve. If a player's registration time exceeds the preset number of days and the standard deviation of their historical first ability deviation is less than the threshold, then Path 2 is selected. Path 2 calculates the first ability deviation based on the social comfort difficulty range.

[0079] In other cases, the information entropy is calculated based on the player's behavior data, that is, for the first to third behavior sequences respectively. Then, when the information entropy of either the first or second behavior sequence is higher than the threshold, path one is selected, and the calculation path based on the target challenge curve is adopted. Otherwise, choose path two, which uses a calculation path based on the social comfort difficulty range.

[0080] In this embodiment, if the information entropy of either the first or second behavior sequence is higher than the threshold, path one can be selected directly based on the player's behavior data; otherwise, path two can be selected.

[0081] In the implementation of this application embodiment, firstly, based on the current period player ability level vector... Calculate the player's first ability deviation. Then, in this embodiment, two parallel paths are used to obtain the first capability deviation through two alternative calculation paths: Path 1 is based on the calculation of the absolute deviation from the target challenge curve; Path 2 is based on the calculation of relative deviation from the social comfort difficulty range; The system automatically selects one of the paths based on preset conditions.

[0082] Choose a path according to the following rules: If a player has been registered for less than seven days, path one will be forced to be used. If the registration period is more than 30 days and the standard deviation of the capability vector over the past five periods is less than 0.1, then use path two; In other cases, calculate the information entropy of the first and second behavior sequences of the current period. If any entropy value is higher than the threshold, for example, 0.8, it indicates that the behavior is highly random, so use path one; otherwise, use path two.

[0083] In Path 1, a target challenge curve is pre-defined. The target challenge curve is a vector of the player's expected ability level at each cycle, pre-defined by the design team under ideal conditions. This vector represents a function curve characterizing the ability level, fitted by the design team before the game's launch based on strategic planning ability, execution efficiency, and retention goals. It is typically a monotonically increasing smooth curve, achieved through methods such as cubic spline interpolation or exponential growth functions. The current cycle number is then used to define this curve. Substitute the target challenge curve The target capability vector for the current period can be represented as: ; Then, the calculated Euclidean distance is used as the first capability deviation. This can be represented as:

[0084] In actual implementation, we introduce Euclidean distance mainly because it can directly measure the absolute gap between a player's current ability and the ideal progress. It is suitable for new users or scenarios where the behavior pattern is not yet stable. Therefore, when selecting multiple paths to calculate the deviation of the player's first ability, this path can be used for new users who have been registered for less than seven days, or unstable users whose historical first ability deviation standard deviation in the previous few periods is greater than the preset threshold.

[0085] In path two, a sample set of ability distribution is extracted from the player group at the same stage. Calculate the average ability of this group. with standard deviation The social comfort difficulty range is defined as follows: If the player's ability If it falls within this interval, then Otherwise, calculate its distance to the nearest interval boundary:

[0086] in, It is a vector composed of the mean and standard deviation of each dimension, and so on. It is a vector composed of the mean and standard deviation of each dimension. This path is applicable to mature users who have been registered for more than 30 days and whose behavioral entropy value is lower than the threshold.

[0087] Then, this socially comfortable difficulty range is used to calculate the distance between the player and this range, reflecting the degree of abnormality of the player relative to the player group at the same stage. It is also applicable to mature user groups and can be further refined. This can better increase player stickiness and prevent the difference in player ability level models among individuals in the same player group from being too large.

[0088] Based on the player's registration duration, the stability of the historical first ability deviation, and the information entropy of the current cycle's behavior sequence, either path one or path two is dynamically selected.

[0089] Once you get It immediately determines whether it matches the preset threshold. The size relationship.

[0090] The threshold is determined using the idea of ​​A / B testing. And specifically set values ​​to ensure that approximately 80% of players are within the steady-state adjustment range; then when When, the difficulty parameter matrix is ​​generated using an adjustment strategy function; when At that time, an event and parameter mapping library is used to generate a difficulty parameter matrix.

[0091] In one possible implementation, based on the player's ability level model, the player's first ability deviation is obtained to acquire the difficulty parameter matrix, and the following are also included: Based on the player's ability level model, select at least one path to obtain the player's first ability deviation. Based on the deviation from the first ability, obtain the difficulty parameter matrix; When the deviation of the first ability is less than or equal to the preset threshold, the difficulty parameter matrix is ​​obtained based on the preset adjustment strategy function; When the deviation of the first ability exceeds the preset threshold, the difficulty parameter matrix is ​​obtained based on the player ability level vector of the current period in the player ability level model and the preset mapping library of in-game events and parameter adjustments.

[0092] The adjustment policy function is a multi-layer neural network model. The input layer receives the player's ability level model, the hidden layer uses the ReLU activation function, and the output layer is constrained to a preset range by the Sigmoid function and mapped to the actual parameter adjustment range through a linear transformation. The regulation policy function is a differentiable mapping In this embodiment, This indicates four categories of adjustable game parameters, including: daily mission objectives, stamina recovery rate, crafting recipe difficulty, and spatial layout restrictions. This indicates the strength of the effect of each type of adjustable game parameter on strategy, execution, and investment.

[0093] The policy function is then adjusted using a three-layer fully connected neural network, with the input layer receiving the player's current ability level vector. The hidden layer uses the ReLU activation function, and the output layer is constrained to the [0,1] interval by the Sigmoid function, and then mapped to the actual parameter range through a linear transformation.

[0094] Then, during the initial training of the model, the weights of the neural network are optimized through a reward function, which is defined as the weighted sum of player retention rate and payment conversion rate over multiple future periods. The retention rate is defined by the first and second behavior sequences of the player, and the payment conversion rate is defined by the third behavior sequence.

[0095] This allows for more continuous output and smoother gradients, enabling subtle adjustments to difficulty and preventing abrupt changes in player perception.

[0096] The mapping library for in-game events and parameter adjustments includes at least one predefined rule, and each rule includes the triggering event conditions and the corresponding parameter increment matrix; When the mapping library for event and parameter adjustment is called, the matching degree between the current player behavior characteristics and the triggering event conditions of each rule is calculated, and the parameter increment matrix corresponding to the best matching rule is selected and superimposed on the basic difficulty matrix to generate the final difficulty parameter matrix.

[0097] The mapping library for in-game events and parameter adjustments is a predefined set of rules. ,in Each rule includes a triggering event condition. and the corresponding parameter increment matrix .

[0098] Each rule is set according to the actual situation, for example: Rule 1 is set as follows: If a player's crafting failure rate has been higher than 80% for three consecutive periods, then... Applying a negative increment to the synthesis formulation dimension represents reducing the formulation difficulty; Rule 2 is set as follows: the player's space utilization rate has been below 0.3 for two consecutive periods and the density of high-level furniture is zero. Relax the initial available area in terms of spatial constraints.

[0099] exist At that time, the current behavioral characteristics will be compared with all Matching and similarity calculation are performed. This implementation directly measures the structural similarity of the features of two vectors and selects the best match. This serves as the trigger condition for the player's current cycle event, and its corresponding parameter increment matrix is ​​used. Superimposed on the basic difficulty matrix Basic Difficulty Matrix The final difficulty parameter matrix is ​​generated by the game design team's default configuration. .

[0100] In this step, difficulty adjustment and player behavior are actually treated as a repeated game. Therefore, after paths one and two, a preset path three can be introduced. At this time, paths one and two are used as the baseline difficulty model, and path three is used as the virtual game model. The baseline difficulty model includes at least paths one and two, and the virtual game model includes at least path three.

[0101] If path three is chosen, then paths one and two will run in parallel.

[0102] Because in actual player operation, behavior is highly random and lacks patterns. Sometimes players focus on strategy, and sometimes they focus on completing tasks. There are also players who actively change their behavior to test the game and try to gain a more favorable difficulty in order to understand the game's adjustment rules. At this time, the simple baseline difficulty model path one and two may not be enough, and a more complex virtual game model path three is needed to deal with it.

[0103] In one possible implementation, the player's first ability deviation is obtained based on the player's ability level model, including: In the current cycle, based on the player's ability level model and a preset baseline difficulty model and / or virtual game model, the player's first ability deviation is obtained, including: Based on the player's ability level model, and based on at least two preset baseline difficulty models, obtain the corresponding initial ability deviation. When the initial ability deviations calculated by different benchmark difficulty models indicate different directions of difficulty adjustment, the game trigger condition is deemed met; or, When the path conflict intensity value calculated based on the initial capability deviation exceeds the preset conflict threshold, it is determined that the game triggering condition is met. In response to the fulfillment of any game trigger condition, a virtual game model is initiated to generate the first capability deviation for the current period; Otherwise, the initial capability deviation generated by the baseline difficulty model will be used as the first capability deviation for the current cycle; Based on the final determined first capability deviation, a difficulty parameter matrix is ​​obtained for configuring the next cycle.

[0104] The baseline difficulty model includes at least paths one and two, and the virtual game model includes at least path three, including: Path 1 is configured as follows: Based on the player's ability level model and a preset target challenge curve, obtain the first ability deviation. Path 2 is configured as follows: Based on the player's ability level model and the socially comfortable difficulty range, obtain the first ability deviation. Path 3 is configured as follows: based on the player's ability level model and an algorithm based on virtual game theory, obtain the first ability deviation.

[0105] When implementing the embodiments of this application, please refer to the following: Figure 4 , Figure 4 The flowchart for selecting three paths provided in this embodiment of the invention checks whether the triggering conditions of the virtual game are met. The triggering conditions include two: The system detects directional conflicts in the initial ability deviation obtained from the baseline difficulty model by comparing the difficulty adjustment directions calculated for path one and path two. If path one suggests increasing the difficulty while path two suggests decreasing the difficulty, meaning the calculated results are opposite, then a directional conflict is considered to have occurred.

[0106] The strength of the conflict is detected by analyzing the initial capability deviation obtained from the baseline difficulty model. The difference between the results of path one and path two is calculated, and the difference between the two initial capability deviations is directly used as the path conflict strength value. If this value exceeds a preset threshold, usually set to 0.5, the conflict is considered severe enough, and a virtual game needs to be initiated to proceed to path three.

[0107] This embodiment also includes monitoring the information entropy of the player's behavior sequence, which remains high for multiple consecutive cycles. This is because it is impossible for a normal player to exhibit highly random or patternless behavior, or to detect patterns that could cheat the game, such as periodic, intentionally inefficient behavior. In such cases, the player may be actively testing the game's difficulty adjustment rules.

[0108] The virtual game algorithm of path three will be activated as long as any one of the above conditions is met.

[0109] In one possible implementation, path three is configured as follows: based on the player ability level model and an algorithm based on virtual game theory, the first ability deviation is obtained, including: Obtain at least two initial ability deviations calculated from the baseline difficulty model; When a conflict in adjustment direction or adjustment intensity is detected between the initial capability deviations, it is determined that the game triggering condition is met. In response to the fulfillment of the game triggering conditions, a virtual game model is initiated to obtain the first capability deviation, corrected for game equilibrium. This first capability deviation includes: Based on the counterfactual reasoning framework, a virtual opponent model is obtained. The virtual opponent model is used to simulate the player's response behavior to the system's difficulty strategy. Its parameters are updated by minimizing the KL divergence between the predicted player behavior distribution and the actual player behavior distribution. Using long-term player engagement metrics as the optimization objective, we solve for the Nash equilibrium point in the system difficulty strategy space to obtain an balanced difficulty strategy. Based on the equilibrium difficulty strategy and the player's current ability level, obtain the first ability deviation after game equilibrium correction; The first capability deviation, corrected by game equilibrium, is taken as the first capability deviation.

[0110] When implementing the embodiments of this application, if the first capability deviation is calculated by simultaneously starting three paths, path three does not run in every cycle. Its start depends on the satisfaction of any triggering condition.

[0111] This invention introduces path three as a virtual game model, which works in conjunction with the baseline difficulty model formed by paths one and two. The initial capability deviation obtained from paths one and two is calculated. and Then, check for any conflicts in adjustment direction or intensity.

[0112] Then, direction conflict is defined as... The symbol indicates deviation from the direction. This means that the player's ability is advanced and the difficulty needs to be increased. Negative representation indicates a lag in capability, requiring a reduction in difficulty. If the directions of these two paths conflict, it means that the two baseline models are giving opposing adjustment suggestions. Define intensity conflict as Even if the two paths adjust in the same direction, the difference in adjustment magnitude is too large, indicating a significant disagreement in the assessment of the player's state, which will trigger path three for further game analysis.

[0113] Furthermore, if the information entropy of the player's first and second behavior sequences is detected to be higher than a threshold, for example, 0.8, for several consecutive cycles (e.g., three cycles), it indicates that the player's behavior is highly random and lacks a stable pattern.

[0114] Alternatively, pattern recognition algorithms can be used to identify periodic inefficient operations, primarily by detecting specific behavioral patterns, such as repeatedly synthesizing low-value items or frequently canceling layouts. These behaviors indicate that players may be testing and adjusting the rules.

[0115] In actual operation, all of the above are considered to meet the game triggering conditions.

[0116] Once any game trigger condition is met, the virtual game algorithm of Path 3 is activated to generate the first capability deviation.

[0117] Then, the virtual game algorithm of path three is built on the counterfactual reasoning framework.

[0118] Then, a virtual opponent model M is built using the structure of a recurrent neural network (RNN), which models player behavior as a response to difficulty strategies. response function ,in In game state, The virtual opponent model uses a recurrent neural network (RNN) structure. The input is a sequence of historical actions, and the output is the predicted distribution of actions in the next cycle. The historical action sequence here is a concatenation of the first to third historical action sequences. Then, by minimizing the distribution of predicted behavior... Distribution of actual behavior The KL divergence between the two models updates the parameters of the virtual opponent model: ; Then solve for the Nash equilibrium point. This makes the difficulty strategy at that point... Optimal response strategy with players To establish an equilibrium pair, in practice, the player's current ability level vector C is input into the virtual game model. The gradient ascent method is then used to search the difficulty parameter space for the long-term participation index, which is directly defined as maximizing the weighted activity over the next five periods. And calculate the Corresponding ability deviation As the first capability deviation after game equilibrium correction .

[0119] In this embodiment, path three is constructed based on counterfactual reasoning, maintaining a virtual opponent model M to simulate the player's optimal response to strategies of varying difficulty, thus modeling player behavior as a response to system difficulty parameters. The response function is expressed as Where: s is the game state, including historical behavior sequences, resource status, etc.; a is the player action, such as arranging furniture, crafting resources, etc. It is a pre-set vector of difficulty parameters, including task objectives, stamina recovery rate, etc.

[0120] The goal is to predict a player's future performance in a given context by learning from their historical behavior. The difficulty lies in understanding the distribution of behaviors and finding the equilibrium that maximizes long-term engagement. .

[0121] A virtual opponent model is constructed using a recurrent neural network, where the input is a historical behavior sequence s, and the output is the predicted behavior distribution for the next cycle. .

[0122] Then, the model is updated using the loss function, which is calculated by minimizing the prediction distribution. Distribution of behaviors actually observed Use the KL divergence between them to update the model parameters: ; Actual distribution It is obtained from the statistical analysis of the behavior sequence collected in the current period, such as the distribution of action frequency.

[0123] Then, training is performed. After collecting player behavior data in each cycle, the RNN parameters are updated using the loss function, so that the virtual opponent model can more accurately simulate the player's response to difficulty.

[0124] The problem of difficulty adjustment is modeled as a virtual two-player game, and the Nash equilibrium is solved. Player strategy is a response function The goal is to maximize one's own gaming experience, such as efficiency in completing levels and resource accumulation; Difficulty parameters of system difficulty strategy The goal is to maximize long-term player engagement, such as weighted activity over the next five cycles.

[0125] Nash equilibrium point Satisfy: At this point, the difficulty parameter selected by the system... And the optimal response strategy for players They form an equilibrium pair, where neither side gains any additional benefit from unilaterally changing its strategy.

[0126] Specifically, the current player's ability level vector C is input into the virtual opponent model.

[0127] Define long-term engagement metrics For example, the weighted activity over the next five cycles:

[0128] in This is the predicted activity level for cycle t, which can be obtained by regression analysis based on the relationship between historical activity and difficulty. The discount factor can be set higher, for example, 0.9.

[0129] Using gradient ascent to search the difficulty parameter space... Maximize , can be represented as: ; The specific iterative formula is as follows: ; in For learning rate, gradient Behavioral distribution pairs that can be predicted by virtual opponent models The differential approximation is calculated using the policy gradient method.

[0130] Finally, the deviation of the first ability is calculated to obtain the balanced difficulty parameter. Then, it is converted into the first capability deviation. ,Will Mapping to target capability vector The adjustment amount, and then setting the target capability vector corresponding to the baseline difficulty as The adjusted target vector is then expressed as: ; in The mapping coefficient matrix, This is the baseline difficulty parameter.

[0131] Calculate the Euclidean distance between the player's current ability vector C and the adjusted target vector as... , can be represented as: ; Then, the sign of its calculation is defined, if C precedes... If the result is positive, then the difficulty needs to be increased; conversely, if the result is negative, then the difficulty needs to be decreased.

[0132] Finally, the first capability deviation, corrected for game equilibrium, is used. As the first capability deviation This is used to generate the difficulty parameter matrix for the next cycle.

[0133] Path three uses a virtual game model to simulate player behavior as a response to increasingly difficult strategies and searches for a Nash equilibrium that maximizes long-term engagement. This method effectively addresses players' strategic probing behavior and provides more robust difficulty adjustments when the baseline model conflicts, thereby improving game adaptability and player experience.

[0134] Step S4: When entering the next cycle, perform difficulty adjustment according to the difficulty parameter matrix and obtain the player's second interactive behavior.

[0135] When implementing the embodiments of this application, adjusting the game difficulty involves adjusting at least one adjustable game parameter, including: the target value of daily tasks, the natural recovery rate of stamina, the difficulty of crafting recipes, and spatial arrangement restrictions.

[0136] In the implementation of this application embodiment, when entering the next cycle, difficulty adjustment is performed according to the difficulty parameter matrix D.

[0137] This matrix contains four categories of adjustable game parameters: the number of daily mission objectives, stamina recovery rate, crafting recipe difficulty, and initial space layout constraints. Each parameter has an independent influence across three dimensions: strategic planning, execution efficiency, and economic investment.

[0138] For example, the adjustment of the number of daily task objectives is mainly influenced by the strategic planning ability score. The impact, and the adjustment range is ,in The preset sensitivity coefficient is used; the physical recovery rate is mainly affected by the execution efficiency score. Modulation, adjustment formula is ,in The base value for recovery rate; the difficulty coefficient of the synthesis formula and the score for economic input capability. A negative correlation is reflected in the fact that players with a high willingness to pay can obtain more lenient crafting conditions; initial spatial layout restrictions, such as the initial number of available plots, are then considered, and a weighted fusion formula is applied to all three factors. ,in .

[0139] The daily task objective refers to the target value of the daily task, which means adjusting the target value of the daily task. For example, based on the difficulty parameter matrix calculated in the previous step, the daily task objective increment in the adjustable game parameters is -2 times, which means that the number of successful synthesis attempts is required to be reduced from 5 to 3. Modifying the natural stamina recovery rate can increase it from 1 point every 5 minutes to 1 point every 4 minutes. Change the difficulty of the raw material synthesis formula, such as increasing the success rate of a certain high-level item from 40% to 60%; Reset spatial layout constraints, for example, expand the initial unlock area from 3×3 to 4×4; All adjustments are injected through the parameter interface of the game logic layer to ensure that they take effect in real time.

[0140] Step S5: Obtain the second capability deviation based on the second interaction behavior; When the deviation of the second ability is less than that of the first ability, the difficulty parameter matrix is ​​updated based on the adjustment strategy function.

[0141] In one possible implementation, a second capability deviation is obtained based on the second interaction behavior. In response to the second capability deviation being greater than the first capability deviation, the difficulty parameter matrix is ​​updated, including: In response to the second ability deviation being greater than the first ability deviation, the difficulty parameter matrix is ​​updated; When the deviation of the second ability is greater than that of the first ability, it indicates that the difficulty parameter matrix obtained in the previous period has no effect on the current player's difficulty adjustment. In order to adjust the difficulty parameter matrix according to the player's ability level, the difficulty parameter matrix is ​​obtained directly based on the preset mapping library of in-game events and parameter adjustments.

[0142] In the next cycle, collect the player's second interaction behavior synchronously, and repeat steps S1 to S2 to obtain the updated player ability level model. And calculate the second capability deviation. Then compare and .

[0143] when This indicates that the previous adjustment failed to effectively converge player abilities, forcibly switching to the event and parameter mapping library path to regenerate the difficulty parameter matrix, even if the current... This mechanism prevents the adjustment policy function from getting trapped in local optima.

[0144] This embodiment also includes step S6, which, in response to the change in the third row sequence, obtains a difficulty parameter matrix based on a preset adjustment strategy function. This step only considers economic factors and uses the change in the third row sequence as a basis to adjust the difficulty and attract players to spend money.

[0145] In one possible implementation, after obtaining the second ability deviation based on the second interaction behavior, and updating the difficulty parameter matrix in response to the second ability deviation being greater than the first ability deviation, the method further includes: In the current cycle, based on the player's first interaction, obtain the third action sequence for the current cycle; When entering the next cycle, the third action sequence for the next cycle is obtained based on the player's second interaction action; Based on the third row sequence of the current period and the third row sequence of the next period, obtain the change in the third row sequence. When the change in the third row sequence exceeds a preset threshold, the difficulty parameter matrix is ​​updated based on the adjustment strategy function.

[0146] Detect changes in the third row of the sequence; When the change in the third row sequence exceeds a preset threshold, the difficulty parameter matrix is ​​updated based on the adjustment strategy function, without considering the relationship between the second ability deviation and the first ability deviation.

[0147] Detecting changes in the third row of the sequence, including: Calculate the changes in the dimension of economic input willingness between the old and new cycles; When the absolute value of the change is greater than the preset payment sensitivity threshold, it is determined that the third behavior sequence has changed significantly.

[0148] In response to the sequence change in the third row, a difficulty parameter matrix is ​​obtained based on a preset adjustment strategy function, including: While updating the difficulty parameter matrix based on the second interaction behavior, it detects whether the third behavior sequence has changed significantly.

[0149] Calculate new and old The absolute value of the difference between the values, if The payment sensitivity threshold can be set to... Then regardless The size is fine-tuned using the adjustment strategy function.

[0150] For example, if players suddenly increase their spending significantly, the mission objectives will be appropriately raised to match their increased willingness to invest money, in order to avoid content being consumed too quickly, resulting in a lack of content or empty missions in the later stages.

[0151] Example 3 This is the third embodiment of the present invention. Based on embodiments 1 and 2, please refer to the following references. Figure 5 , Figure 5 This is a schematic diagram of a difficulty adjustment system based on player behavior data provided in an embodiment of the present invention. Further, the difficulty adjustment system based on player behavior data may specifically include a behavior acquisition unit, a player level calculation unit, and a difficulty adjustment unit connected in sequence. The behavior collection unit is configured to: obtain the first behavior sequence and the second behavior sequence of the current cycle; the first behavior sequence is configured as the player's functional item placement behavior data, and the second behavior sequence is configured as the player's resource synthesis operation and stamina status change data; Obtain the first and second row sequences of the current period; The player level calculation unit is configured to obtain a player ability level model based on the first and second line sequences. The difficulty adjustment unit is configured to: obtain the player's first ability deviation based on the player's ability level model, in order to obtain the difficulty parameter matrix; The behavior collection unit is also configured to: when entering the next cycle, perform difficulty adjustment according to the difficulty parameter matrix, and repeat the above steps to obtain the player's second interaction behavior; The difficulty adjustment unit is also configured to: obtain a second ability deviation based on the second interaction behavior, and update the difficulty parameter matrix in response to the second ability deviation being greater than the first ability deviation.

[0152] Example 4 The fourth embodiment of the present invention differs from the previous embodiments in that: Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, devices, or units, or they may be electrical, mechanical, or other forms of connection.

[0154] The units described as separate components may or may not be physically separate. As will be apparent to those skilled in the art, the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0155] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0156] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a grid device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0157] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A difficulty adjustment method based on player behavior data, characterized in that, include: In the current cycle, based on the player's first interaction, obtain the first action sequence and the second action sequence for the current cycle; The first behavior sequence is configured as the player's functional item placement behavior data, and the second behavior sequence is configured as the player's resource synthesis operation and stamina status change data; Based on the first and second behavior sequences, a player ability level model is obtained. Based on the player ability level model, the player's first ability deviation is obtained to obtain the difficulty parameter matrix; When entering the next cycle, according to the difficulty parameter matrix, the difficulty is adjusted, and the above steps are repeated to obtain the player's second interactive behavior; Based on the second interaction behavior, a second ability deviation is obtained, and in response to the second ability deviation being greater than the first ability deviation, the difficulty parameter matrix is ​​updated.

2. The difficulty adjustment method based on player behavior data according to claim 1, characterized in that, After the step of obtaining the second ability deviation based on the second interaction behavior, and updating the difficulty parameter matrix in response to the second ability deviation being greater than the first ability deviation, the method further includes: In the current cycle, based on the player's first interaction, obtain the third action sequence for the current cycle; When entering the next cycle, the third action sequence for the next cycle is obtained based on the player's second interaction action; Based on the third row sequence of the current period and the third row sequence of the next period, obtain the change in the third row sequence. When the change in the third row sequence exceeds a preset threshold, the difficulty parameter matrix is ​​updated based on the adjustment strategy function.

3. The difficulty adjustment method based on player behavior data according to claim 1, characterized in that, The step of obtaining the player's first ability deviation based on the player ability level model includes: Obtain the target capability vector in the current period from the preset target challenge curve; Based on the player ability level model and the target ability vector, calculate the Euclidean distance to obtain the first ability deviation. Based on the first ability deviation, a difficulty parameter matrix is ​​obtained.

4. The difficulty adjustment method based on player behavior data according to claim 1, characterized in that, The step of obtaining the player's first ability deviation based on the player ability level model includes: Based on the second line sequence, obtain the player cluster of active players; Based on the player ability level model of the active player cluster, a socially comfortable difficulty range is obtained; Based on the social comfort difficulty range and the player ability level model, obtain the player's first ability deviation. Based on the first ability deviation, obtain the difficulty parameter matrix; The step of obtaining the socially comfortable difficulty range based on the player ability level model of the active player cluster includes: Obtain player ability level model data for active players; Active players are divided into at least one player cluster through cluster analysis; For each player cluster, calculate the statistical distribution of the player ability level vector in the current period, and take the range of mean plus or minus standard deviation as the social comfort difficulty range of the player cluster.

5. The difficulty adjustment method based on player behavior data according to claim 1, characterized in that, The step of obtaining the player's first ability deviation based on the player ability level model to obtain the difficulty parameter matrix also includes: Based on the player ability level model, at least one path is selected to obtain the player's first ability deviation. Based on the first ability deviation, obtain the difficulty parameter matrix; When the deviation of the first ability is less than or equal to the preset threshold, the difficulty parameter matrix is ​​obtained based on the preset adjustment strategy function; When the deviation of the first ability exceeds a preset threshold, a difficulty parameter matrix is ​​obtained based on a preset mapping library of in-game events and parameter adjustments.

6. The difficulty adjustment method based on player behavior data according to claim 1, characterized in that, The step of obtaining the player's first ability deviation based on the player ability level model includes: In the current cycle, based on the player ability level model and a preset baseline difficulty model and / or virtual game model, the player's first ability deviation is obtained, including: Based on the player ability level model, and based on at least two preset baseline difficulty models, the corresponding initial ability deviation is obtained; When the initial ability deviations calculated by different benchmark difficulty models indicate different directions of difficulty adjustment, the game triggering condition is determined to be met; or, When the path conflict intensity value calculated based on the initial capability deviation exceeds the preset conflict threshold, it is determined that the game triggering condition is met. In response to the satisfaction of any of the game triggering conditions, the virtual game model is activated to generate the first capability deviation for the current period; Otherwise, the initial capability deviation generated by the benchmark difficulty model will be used as the first capability deviation for the current cycle; The baseline difficulty model includes at least paths one and two, and the virtual game model includes at least path three, including: The first path is configured as follows: based on the player's ability level model and a preset target challenge curve, obtain the first ability deviation. The second path is configured as follows: based on the player's ability level model and the socially comfortable difficulty range, obtain the first ability deviation. The third path is configured to: obtain the first ability deviation based on the player's ability level model and a virtual game algorithm.

7. The difficulty adjustment method based on player behavior data according to claim 6, characterized in that, The third path is configured as follows: based on the player's ability level model and an algorithm based on virtual game theory, the first ability deviation is obtained, including: Obtain at least two initial ability deviations calculated from the baseline difficulty model; When a conflict in adjustment direction or adjustment intensity is detected between the initial capability deviations, it is determined that the game triggering condition is met. In response to the fulfillment of the game triggering condition, a virtual game model is initiated to obtain a first capability deviation corrected for game equilibrium. This first capability deviation includes: Based on the counterfactual reasoning framework, a virtual opponent model is obtained. The virtual opponent model is used to simulate the player's response behavior to the system's difficulty strategy. Its parameters are updated by minimizing the KL divergence between the predicted player behavior distribution and the actual player behavior distribution. Using long-term player engagement metrics as the optimization objective, we solve for the Nash equilibrium point in the system difficulty strategy space to obtain an balanced difficulty strategy. Based on the equilibrium difficulty strategy and the player's current ability level, the first ability deviation corrected by game equilibrium is obtained; The first capability deviation, corrected by game equilibrium, is taken as the first capability deviation.

8. The difficulty adjustment method based on player behavior data according to claim 1, characterized in that, Based on the first and second behavior sequences, a player ability level model is obtained, including: Based on the first behavioral feature vector, a first ability score is obtained; The first behavioral feature vector is linearly combined with a preset first weight vector to obtain the first ability score, wherein the first weight vector is obtained by dimensionality reduction calculation based on historical player behavior data. Based on the second behavioral feature vector, the second ability score is obtained; The components of the second behavioral feature vector are normalized and then weighted and fused based on preset weights to obtain the second capability score. Based on the player's first interaction behavior, the third behavior sequence of the current period is also obtained to obtain the third behavior feature vector; the third behavior sequence is configured as the player's paid transaction data; Based on the third behavior feature vector, a third ability score representing the player's willingness to invest economically is obtained; The third row of the feature vector is subjected to a logarithmic transformation at least one component, and then weighted and fused with the remaining components to obtain the third capability score. A player ability level model is obtained based on the first ability score, the second ability score, and the third ability score.

9. A difficulty adjustment system based on player behavior data, characterized in that, This includes a behavior acquisition unit with sequential electrical connections, a player level calculation unit, and a difficulty adjustment unit; The behavior acquisition unit is configured to: in the current cycle, based on the player's first interaction behavior, obtain the first behavior sequence and the second behavior sequence of the current cycle; the first behavior sequence is configured as the player's functional item placement behavior data, and the second behavior sequence is configured as the player's resource synthesis operation and stamina status change data. The player level calculation unit is configured as: a first line sequence and a second line sequence to obtain a player ability level model; The difficulty adjustment unit is configured to: obtain the player's first ability deviation degree according to the player ability level model, so as to obtain a difficulty parameter matrix; The behavior acquisition unit is also configured to: when entering the next cycle, perform difficulty adjustment according to the difficulty parameter matrix, and repeat the above steps to acquire the player's second interactive behavior; The difficulty adjustment unit is further configured to: obtain a second ability deviation degree based on the second interaction behavior, and update the difficulty parameter matrix in response to the second ability deviation degree being greater than the first ability deviation degree.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

Images