A game level difficulty assessment method and system

Abstract

The application provides a game level difficulty evaluation method and system, the method comprises the following steps: obtaining a current game level to be tested; obtaining random AI parameters and specific level pass parameters, setting and using a preset random AI test tool to simulate the pass operation of the current game level to be tested to obtain initial pass test data; correcting the initial pass test data by using a preset probability model to obtain current level pass difficulty data; and performing personalized setting on the game level to be tested according to the current level pass difficulty data. The application solves the technical problems of high dependence on human engineering, low consistency of test results, low test configuration efficiency and poor applicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of game testing, and in particular to a game level difficulty evaluation method and system. BACKGROUND

[0002] With the continuous development of electronic game software development technology and the rise of emerging industries such as electronic sports, modeling for electronic game player experience and personalized settings has become a new research and development direction in the field of game development.

[0003] The existing electronic game level difficulty and pass difficulty calculation mainly relies on repeated level testing by test officers to count difficulty information. Based on the game state, experts extract level difficulty and player experience features, combine the win rate data of a level, and fit an approximate curve of the difficulty. In addition, game level operation can also be performed by random AI to count the level pass difficulty data. In the prior art, the historical record information of user level pass data corresponding to the existing level can be used as a sample to train AI for simulating level pass operation, and the AI level pass is used to estimate the level pass difficulty of a specific level in the electronic game.

[0004] In the prior art, the level pass difficulty is counted by repeated level testing by test officers, which requires a large amount of manpower and time cost of game developers and testing departments, and requires manual test case preparation and manual feedback by test officers and other recorders, resulting in slow feedback speed of level pass testing and low testing efficiency. At the same time, since the test officers perform manual level operation, the test results are easily affected by the subjective factors of the operators, resulting in low consistency of the level pass test results.

[0005] In the prior art, different features are extracted for each game, and the accuracy of the feature extraction operation depends on the professionalism of the professional personnel, which is easily affected by the differences in personal technical level and subjective speculation, resulting in that the feature extraction of a specific level cannot fully and truly reflect the difficulty of passing the level, and there is a large error between the consistency of the test results and the actual value.

[0006] The random AI used in the prior art has a large difference from the behavior of human players, and the random AI cannot completely simulate the operation habit features and other key parameters of human players, so there is still a gap between the accuracy of the level pass difficulty calculation and the real test, and the test results obtained by the random AI virtual test of the level pass difficulty are not representative.

[0007] With the continuous development of information technology, the technology of software testing based on neural network and sample training is continuously applied to game operation and user experience testing work. However, the existing technology based on data collection and training AI needs a large amount of time and computing cost, the deployment and setting of algorithm and model are relatively complex, the algorithm complexity is high, and the data collection and training AI need a large amount of time and computing cost. Meanwhile, the simulation test performance of AI trained by data collection and the generalization ability of AI for new levels cannot be guaranteed. For example, the existing patent application document with the publication number CN113660987A “Generating a game bot for a game application” generates a game bot through a system including a processor; receives game telemetry data corresponding to an actual player from a game application through the system; generates game telemetry data corresponding to the game bot from the game application through the system; based on the game telemetry data corresponding to the actual player and the game telemetry data corresponding to the game bot, generates difference data through the system, the difference data representing the difference between a first character generated by the actual player and a second character generated by the game bot over time; and based on the difference data, updates the game bot through the system. The technical solution of the existing application document generates a game bot to run a game to obtain difference data. Since there is a big difference between the behavior mode of the game bot and that of the human player, the simulation running effect is poor, and the test data obtained is not representative. The existing patent application document with the publication number CN111841009A “Method and device for controlling game difficulty, electronic equipment and medium” obtains a current game state of a current level during a user game process. Then, according to user information of the user and the current game state, it is determined whether the game difficulty of the current level needs to be adjusted, wherein the user information is determined based on historical operations of the user on the game. If the game difficulty of the current level needs to be adjusted, the predetermined game parameters of the next game state corresponding to the current game state are adjusted, and a new game state is generated based on the adjusted game parameters. The technical solution disclosed in the existing application document needs to adjust the difficulty of the effective level according to the user information and the current game state. This way of generating a new effective state relies on manual operation and is easily affected by the subjectivity of the operator, resulting in low consistency of the results.

[0008] In summary, the existing technology has the technical problems of high dependence on human engineering, low consistency of test results, low efficiency of test configuration, and poor applicability. SUMMARY

[0009] The technical problem to be solved by the present application is how to solve the technical problems of high dependence on human engineering, low consistency of test results, low efficiency of test configuration, and poor applicability in the existing technology.

[0010] The application solves the above technical problems by adopting the following technical scheme: a game level difficulty evaluation method comprises the following steps:

[0011] S1, acquiring a current game level to be tested;

[0012] S2, acquiring random AI parameters and specific level pass parameters, and setting and using a preset random AI test tool to simulate pass operation on the current level to be tested to obtain initial pass test data;

[0013] S3, correcting the initial pass test data by using a preset probability model to obtain current level pass difficulty data, wherein step S3 comprises:

[0014] S31, acquiring pass step data and pass action data of the current game level to be tested, and using a preset winning probability logic process to acquire actual close-to-situation data;

[0015] S32, using a preset condition independent logic process to process the actual close-to-situation data, so that the conditions of the situation corresponding to the actual close-to-situation data are independent, and independent condition situation data is obtained;

[0016] S33, using a preset pass probability logic process to process the independent condition situation data to obtain a pass step corresponding probability;

[0017] S34, according to the pass step corresponding probability, performing a preset number of hypothetical experiments by using a preset experiment logic to obtain applicable distribution hypothetical experiment data, processing the applicable distribution hypothetical experiment data by using a preset condition probability logic to obtain level pass condition probability, and generating current level pass difficulty data;

[0018] S4, performing personalized setting on the game level to be tested according to the current level pass difficulty data.

[0019] The level difficulty in the application can be represented by the win rate when the number of steps is limited (the type of the game needs to be limited). For a new level, first correct it to an infinite step mode, and by applying random AI to pass the level many times, the win probability under different limited steps can be obtained. The application applies random AI, reduces the threshold of AI training, and eliminates the dependence on user data collection. Due to the behavior difference between real users and random AI, we need to correct the results. After correction, the level win rate and the best step number can be calculated to provide a reference for new level design. The application combines AI technology and probability modeling to provide more accurate and fast game level difficulty evaluation, to accelerate game design and improve user experience.

[0020] In a more specific technical scheme, in step S1, the game level to be tested is modified and set to an infinite step mode.

[0021] In a more specific technical solution, the specific checkpoint passing parameter in step S2 includes: the number of steps required for victory.

[0022] In a more specific technical solution, step S31 includes:

[0023] S311, set the random passing n times, and record the step number of each passing as Si, and the number of executable actions at the last step before passing as Ci;

[0024] S312, record the i-th game as Ai, and set the situation of the i-th game Ai before the last step as Bi;

[0025] S313, utilize the preset winning probability logic to process the passing step number data and the passing action data to obtain actual situation data close to the situation.

[0026] The present application does not need to test the repeated passing of the test officer to count the difficulty, which saves a lot of manpower and time cost, and utilizes the preset winning probability logic to process the passing step number data and the passing action data to obtain actual situation data close to the situation, which overcomes the difference between the random AI and the behavior of the human player in the prior art, and improves the representativeness of the difficulty calculation. The present application improves the system feedback speed, and improves the consistency of the detection result.

[0027] In a more specific technical solution, in step S313, the following logic is used to process the passing step number data and the passing action data to obtain actual situation data close to the situation:

[0028]

[0029]

[0030]

[0031] Wherein, the i subscript represents a certain time of simulation operation, c is the number of selectable actions at the last step, A represents the last state, and B is the state before the last state.

[0032] In a more specific technical solution, step S32 includes:

[0033] S321, record the optimal passing step number as Smin;

[0034] S322, when the random passing number n reaches the preset number threshold, the following logic is used to represent the optimal passing step number Smin:

[0035]

[0036] S323, the following logic is used to process the optimal passing number Smin to obtain the corresponding conditional probability:

[0037]

[0038]

[0039] S324, subtract Sj<Smin+k from the random pass number n with the following logic, and eliminate the i-th game Ai to obtain actual close-to-situation data:

[0040]

[0041] S325, process the actual close-to-situation data with a preset condition independent logic to obtain independent condition situation data.

[0042] In a more specific technical solution, in step S325, the actual close-to-situation data is processed with the following logic:

[0043]

[0044] S i =S min +k,

[0045]

[0046] wherein, represents the probability of the occurrence of a specific state B in the simulation with more than s_min+k steps when it is known that there are not less than 2 endings, and c corresponds to the aforementioned specific state, represents the number of times of the event more than s_min+k.

[0047] In a more specific technical solution, in step S33, the independent condition situation data is processed with the following logic to obtain the pass step number corresponding probability:

[0048]

[0049] In a more specific technical solution, step S34 includes:

[0050] S341, according to the pass step number corresponding probability, perform mn(m+∞) times of hypothetical experiments to make the result expectation distribution the same as the distribution of n times of real experiments to obtain applicable distribution hypothetical experiment data;

[0051] S342, process the applicable distribution hypothetical experiment data with the following logic to obtain the level pass condition probability:

[0052]

[0053] S343, according to the level pass condition probability, generate the current level pass difficulty data with the following logic:

[0054]

[0055] wherein,

[0056] wherein, represents the number of events in the experiment in which the step number is greater than s_min+k+1, s>s min +k represents the number of events in which the corrected step number is greater than s_min+k, is the number of events in the experiment in which the step number is greater than s_min+k;

[0057]

[0058]

[0059] The present application uses probability modeling for correction, improves the accuracy of difficulty calculation, speeds up the difficulty calculation speed, reduces the difficulty calculation cost, eliminates the dependence on expert analysis, can be applied to multiple similar scenarios, and further improves the consistency of the results.

[0060] In a more specific technical solution, a game level difficulty evaluation system includes:

[0061] A game level module is configured to obtain a current game level to be tested.

[0062] An initial test data module is configured to obtain random AI parameters and specific level pass parameters, based on which a preset random AI test tool is set and used to simulate a pass operation on the current level to be tested to obtain initial pass test data. The initial test data module is connected to the game level module.

[0063] A pass difficulty module is configured to correct the initial pass test data using a preset probability model to obtain current level pass difficulty data. The pass difficulty module is connected to the initial test data module and the game level module. The pass difficulty module includes:

[0064] A realistic situation module is configured to obtain pass step data and pass action data of the current game level to be tested, based on which a preset winning probability logic process is used to obtain actual realistic situation data.

[0065] An independent condition situation module is configured to process the actual realistic situation data using a preset condition independent logic process, so that the conditions of the situation corresponding to the actual realistic situation data are independent, based on which independent condition situation data is obtained. The independent condition situation module is connected to the realistic situation module.

[0066] A pass probability module is configured to process the independent condition situation data using a preset pass probability logic process to obtain a pass step number corresponding probability. The pass probability module is connected to the independent condition situation module.

[0067] The difficulty data processing module is used for performing preset number of times of hypothetical experiments on the preset experiment logic according to the pass step number corresponding probability, obtaining applicable distribution hypothetical experiment data, processing the applicable distribution hypothetical experiment data according to the preset conditional probability logic, and obtaining the level pass condition probability, so as to generate the current level pass difficulty data.

[0068] The individualized configuration module is used for performing individualized setting on the to-be-tested game level according to the current level pass difficulty data, and the individualized configuration module is connected with the pass difficulty module.

[0069] Compared with the prior art, the level difficulty in the application can be represented by the win rate when the number of steps is limited (TODO, the type of the game needs to be limited). For a new level, first, the infinite step mode is corrected, and the win probability under different limited steps can be obtained by applying random AI multiple times. The application applies random AI, reduces the threshold of AI training, and eliminates the dependence on user data collection. Since there is a behavior difference between real users and random AI, the result needs to be corrected. After correction, the level win rate and the optimal step number can be calculated to provide a reference for new level design. The application combines AI technology and probability modeling to provide more accurate and fast game level difficulty evaluation, so as to accelerate game design and improve user experience.

[0070] The application does not need a test officer to repeatedly play and count the difficulty, which saves a large amount of manpower and time cost. At the same time, the preset win probability logic is used to process the pass step number data and the pass action data to obtain the actual close-to-situation data, which overcomes the difference between the behaviors of random AI and human players in the prior art, and improves the representativeness of difficulty calculation. The application improves the system feedback speed and the consistency of the detection result.

[0071] The application uses probability modeling for correction, improves the accuracy of difficulty calculation, speeds up the difficulty calculation speed, reduces the difficulty calculation cost, eliminates the dependence on expert analysis, can be applied to multiple similar scenes, and further improves the consistency of the result. The application solves the technical problems of high dependence on human engineering, low consistency of test results, low test configuration efficiency and poor applicability in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A game level difficulty evaluation method step schematic diagram of embodiment 1 of the application

[0073] Figure 2 A game flow representation schematic diagram of embodiment 1 of the application

[0074] Figure 3A game level difficulty evaluation system module schematic diagram for the embodiment 1 of the present application;

[0075] Figure 4 A game level test difficulty running result data schematic diagram for the embodiment 2 of the present application.

[0076] Figure 5 A game level test difficulty running result data schematic diagram for the embodiment 2 of the present application. DETAILED DESCRIPTION

[0077] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in a clear and complete manner in combination with the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0078] Embodiment 1

[0079] As shown in the following table, the level difficulty can be represented by the win rate when the number of steps is limited (TODO, the type of game needs to be limited). Figure 1

[0080] S1, for a new level, first correct to the infinite step mode;

[0081] S2, by applying random AI to play the game for many times, the win probability under different limited steps can be obtained;

[0082] S3, according to the behavior difference between the real user and the random AI, the result is corrected;

[0083] S4, according to the corrected data, the level difficulty is obtained and output; in this embodiment, the level win rate and the best step number can be calculated after correction, which provides a reference for the design of new level. The principle and derivation process of the correction algorithm are introduced as follows.

[0084] As shown in the following table, the level difficulty can be represented by the win rate when the number of steps is limited (TODO, the type of game needs to be limited). Figure 2 In this embodiment, the game process can be represented as a Markov process, at a certain step s, the player faces a situation B, the player selects an action from the action set C to reach a new situation A. Under the premise of determining the action, the state transition is determined. The strategy of the random AI is to randomly select an action at each step, that is, if there are n available actions under the current situation, the probability of selecting each action is 1 / n.

[0085] ​In this embodiment, the difficulty of the level can be represented by the win rate when the number of steps is limited. For a new level, first correct it to the infinite step mode, and by applying random AI to play the level multiple times, the winning probability under different limited steps can be obtained. The following table is an example of data results (the data is illustrative data), a total of 10 times, the minimum pass step is 3 steps, which occurs 2 times, and the maximum pass step is 5 steps, which occurs 6 times.

[0086] Passing steps Occurrence times Number of optional actions before the last step 1 0 2 0 3 2 [3,5] 4 2 [2,4] 5 6 [2,4,1,3，5,3] Total 10

[0087] From the above table, we can get the pass probability of random AI under a certain number of steps, however, due to the behavior difference between real users and random AI, we need to correct the results. After correction, the level win rate and the best step number can be calculated, which provides a reference for new level design. The following introduces the principle and derivation process of the correction algorithm.

[0088] Assumption one: when the game is only one step away from passing, the probability of human players finding the correct winning method is 1

[0089] Randomly pass n times, record each pass step number as Si, and the number of actions that can be performed at the last step before passing is Ci.

[0090] Let the ith game be Ai, and the situation before the last step of Ai be Bi, and

[0091]

[0092]

[0093]

[0094] Where, the subscript i represents a certain simulation run, c is the number of actions available at the last step, A represents the last state, and B is the state before the last step. In this embodiment, there are c actions available from B to A, and only one action can reach A.

[0095] According to assumption one, Bi is closer to the actual difficulty of the game than Ai. Since In this embodiment, referring to the schematic diagram of the game flow, each pre-state corresponds to multiple post-states, so each Bi is not independent of each other, and appropriate conditions are needed to make them conditionally independent.

[0096] Let the optimal pass step number be Smin, and when n is large enough, there is

[0097]

[0098] The above formula represents the minimum pass step number.

[0099] Based on this, remember:

[0100]

[0101] As can be seen from the above logic, when randomly playing levels, S was used exactly. min +k steps to the number of events. Consider S i =S min When calculating the conditional probability, +k must first be subtracted from the denominator n.

[0102] The case where Sj < Smin + k, i.e. Next, remove Ai, that is, reduce the numerator and denominator by 1. Except for Ai, other cases where Sj < Smin + k do not need to be removed, because they are independent of Bi.

[0103]

[0104] S i =S min +k,

[0105]

[0106] In this embodiment, This represents the probability of a specific state B occurring in a simulation where the number of steps exceeds s_min+k at the end of the simulation. The numerator c on the right-hand side of the aforementioned equation corresponds to this specific state, and the denominator... This is the number of events that exceed s_min+k.

[0107] The probability of passing the level within any number of steps S can be calculated below. Let's start with S = Smin:

[0108]

[0109] In the above formula, P(s=s) min Let ) represent the probability of all prerequisite events that allow you to pass the level in s steps. We take the limit to ensure that n >> Ci.

[0110] In numerical calculations, this can be handled as follows:

[0111] When conducting mn(m+∞) hypothetical experiments, the expected distribution of the results should be the same as the distribution of n real experiments. In this way, both n and nSmin become (m+1) times their original values, yielding...

[0112]

[0113] The second step in the above logic transforms the multiplication into addition by taking the logarithm, and L'Hôpital's rule is used in the third step.

[0114] Similarly, P(S≤Smin+k) can be calculated.

[0115] For convenience, we will not calculate the probability directly here, but instead calculate the conditional probability as shown below. The process is basically the same as above.

[0116]

[0117] in,

[0118] In this embodiment, based on the concepts of subordination and conditional independence, and using other symbols to represent the following, it is equivalent to:

[0119] X and Y are subsets of Z, and X and Y are conditionally independent with respect to Z, i.e.

[0120] P(X|Y)=P(X|Z), where X represents Let Y represent s>s min +k, represented by Z

[0121] in, This represents the number of events in the experiment where the number of steps is greater than s_min+k+1, where s>s min +k indicates an event where the corrected step count is greater than s_min+k. This represents the number of events in the experiment where the number of steps is greater than s_min+k.

[0122] because With s>s min +k are all A subset of, and both are given Time-conditional independence, from conditional probability we can obtain P(s≤s) min The recurrence relation for (+k) is as follows:

[0123]

[0124]

[0125] like Figure 3 As shown, the game level difficulty assessment system provided by the present invention includes:

[0126] Game Level Module 1 is used to obtain the current game level to be tested;

[0127] Initial test data module 2 is used to obtain random AI parameters and specific level completion parameters, and to set and use a preset random AI test tool to simulate the level to be tested to obtain initial level completion test data. Initial test data module 2 is connected to game level module 1.

[0128] A difficulty module 3 is configured to correct the initial test data according to a preset probability model to obtain the current level difficulty data, and the difficulty module 3 is connected with the initial test data module 2 and the game level module 1, wherein the difficulty module 3 comprises:

[0129] An actual situation module 31 is configured to obtain the number of steps and the action data of the current game level to be tested, and the actual situation data is obtained by using a preset winning probability logic processing;

[0130] An independent condition module 32 is configured to process the actual situation data according to a preset independent condition logic to make the conditions of the situation corresponding to the actual situation data independent, and the independent condition data is obtained, and the independent condition module 32 is connected with the actual situation module 31;

[0131] A pass probability module 33 is configured to process the independent condition data according to a preset pass probability logic to obtain the probability corresponding to the number of steps, and the pass probability module 33 is connected with the independent condition module 32;

[0132] A difficulty data processing module 34 is configured to perform a preset number of hypothetical experiments according to the probability corresponding to the number of steps, and the applicable distribution hypothetical experiment data is obtained by using a preset experiment logic, and the level pass condition probability is obtained by using a preset condition probability logic to process the applicable distribution hypothetical experiment data, and the current level difficulty data is generated, and the difficulty data processing module 34 is connected with the pass probability module 33;

[0133] An individual configuration module 4 is configured to perform individual setting on the game level to be tested according to the current level difficulty data, and the individual configuration module 4 is connected with the difficulty module 3.

[0134] Embodiment 2

[0135] We apply the above technique to the difficulty evaluation of one of our match-3 games. Match-3 is a casual game where the player can swap two adjacent elements in a horizontal or vertical direction, and if three or more elements of the same color are aligned, they will be removed. Each level has a certain target for the player to remove, and a certain number of steps limit. The difficulty of a level in a match-3 game is originally a random game. After the random seed of the initial stage is determined (the random seed is hidden from the player, but can be obtained in the background), the randomness is eliminated, that is, the same action will inevitably lead to the same result in the same state. In match-3, the level configuration (the shape of the board and the types of elements on it) and the random seed (which controls the falling elements) together determine the difficulty of the level. When we design new levels (add new elements, change the board, etc.) or adjust existing levels (change the random seed and the maximum number of steps to pass), we need to be able to quickly and effectively estimate the difficulty of the level.

[0136] As shown in Figure 4 , two of the action sets that can be performed in this state are marked in the figure. Under the condition of a determined random seed, choosing these two actions will lead the game to two subsequent states respectively.

[0137] Running speed: The performance bottleneck of the difficulty running is the AI random pass. By running game instances in parallel, we can improve the rate of passing. Currently, relying only on a single server (CPU: AMD Ryzen 5 5600X, memory: 32G) and some parallel operations, we can complete the difficulty estimation for 150 levels / seeds within 6 hours.

[0138] As shown in Figure 5 , the running results: we apply the above method to the match-3 game. After obtaining the estimated difficulty, we also test these levels (and seeds) online, and collect real user data to obtain the actual difficulty of the level.

[0139] Compared with the win rate of the random AI pass, the win rate after the correction has a higher correlation with the actual user win rate, although there is a gap in the absolute value. In this example, the correlation increases from 0.24 to 0.63. The estimated results can be used to guide our level design and game operation.

[0140] In summary, the level difficulty in this invention can be represented by the win rate within a limited number of moves (TODO: the game type needs to be specified). For a new level, it is first modified to an infinite move mode. By applying random AI to multiple levels, the win probability under different limited move counts can be obtained. This invention uses random AI, which lowers the threshold for AI training and eliminates the dependence on user data collection. Due to the behavioral differences between real users and random AI, the results need to be corrected. After correction, the level win rate and optimal move count can be calculated, providing a reference for new level design. This invention combines AI technology with probabilistic modeling to provide a more accurate and faster assessment of game level difficulty, thereby accelerating game design and improving user experience.

[0141] This invention eliminates the need for testing specialists to repeatedly play through levels and calculate difficulty, saving significant manpower and time. Simultaneously, it utilizes pre-set winning probability logic to process the number of steps taken and the actions taken to obtain data closely resembling the actual game situation. This overcomes the discrepancy between the behavior of random AI and human players in existing technologies, improving the representativeness of difficulty calculations. This invention also improves system feedback speed and the consistency of detection results.

[0142] This invention employs probabilistic modeling for correction, improving the accuracy and speed of difficulty calculation, reducing its cost, eliminating reliance on expert analysis, and making it applicable to multiple similar scenarios, further enhancing result consistency. This invention solves the technical problems of existing technologies, such as high dependence on manual intervention, low consistency of test results, low test configuration efficiency, and poor applicability.

[0143] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the difficulty of game levels, characterized in that, The method includes: S1. Obtain the current game level to be tested; S2. Obtain random AI parameters and specific level completion parameters, set them accordingly, and use a preset random AI testing tool to simulate the level completion operation of the current game level to be tested, so as to obtain initial level completion test data. S3. Correct the initial level-clearing test data using a preset probability model to obtain the current level's level-clearing difficulty data, wherein step S3 includes: S31. Obtain the number of steps and actions to pass the current game level to be tested, and use the preset winning probability logic to process and obtain the actual situation data. S32. Process the actual close situation data with preset conditional independent logic so that the conditions of the situation corresponding to the actual close situation data are independent, thereby obtaining independent condition situation data. S33. Process the independent conditional situation data with the preset pass probability logic to obtain the probability corresponding to the number of pass steps; In step S33, the independent conditional situation data is processed using the following logic to obtain the probability corresponding to the number of steps to pass the level: In the formula, A i Let B represent the i-th game. i c represents the situation before the final step. i This represents the number of actions that can be performed in the final step before passing the level. S34. Based on the probability corresponding to the number of steps to pass the level, perform a preset number of hypothetical experiments using preset experimental logic to obtain applicable distribution hypothetical experimental data. Process the applicable distribution hypothetical experimental data using preset conditional probability logic to obtain the level-clearing conditional probability, and generate the current level-clearing difficulty data accordingly. S34 includes: S341. Based on the probability corresponding to the number of steps to pass the level, proceed... This hypothetical experiment aims to make the expected distribution of the results the same as the distribution of n real experiments, so as to obtain hypothetical experiment data with an applicable distribution. S342. Process the hypothetical experimental data with the applicable distribution using the following logic to obtain the conditional probability of clearing the level: S343. Based on the probability of clearing the level, generate the current level's difficulty data using the following logic: in, This indicates that the number of steps in the experiment is greater than s. min The number of events plus k+1, This indicates that the number of steps after correction is greater than s. min +k events, For the number of steps in the experiment greater than s min +k events ； S4. Personalize the game level to be tested based on the current level's difficulty data.

2. The method for evaluating the difficulty of game levels according to claim 1, characterized in that, In step S1, the game level to be tested is modified to an unlimited number of steps mode.

3. The method for evaluating the difficulty of game levels according to claim 1, characterized in that, The specific level completion parameters in step S2 include: the number of steps required to win.

4. The method for evaluating the difficulty of game levels according to claim 1, characterized in that, Step S31 includes: S311. Suppose there are n random attempts to pass a level. Let Si be the number of steps to pass each level and Ci be the number of actions that can be performed in the last step before passing a level. S312. Let Ai be the i-th game, and let Bi be the state of Ai before the last move in the i-th game. S313. Process the number of steps to pass the level and the number of actions to pass the level using the preset winning probability logic to obtain the actual close situation data.

5. The method for evaluating the difficulty of game levels according to claim 4, characterized in that, In step S313, the number of steps to clear the level and the action data to clear the level are processed using the following logic to obtain the actual close-to-the-situation data: Where i represents a specific simulation run, c is the number of selectable actions in the last step, A represents the final state, and B is the state one step before the final state.

6. The method for evaluating the difficulty of game levels according to claim 1, characterized in that, Step S32 includes: S321. Let Smin be the optimal number of steps to pass the level. S322. When the number of random attempts n reaches a preset threshold, the optimal number of steps Smin is represented by the following logic: ； S323. Process the optimal number of steps to pass the level, Smin, using the following logic to obtain the corresponding conditional probability: S324. Subtract the cases where Sj < Smin + k from the random number of passes n using the following logic, and remove Ai from the i-th game to obtain the actual close-to-the-situation data: ； S325. The actual close-to-the-situation data is processed by preset conditional independent logic to obtain the independent conditional situation data.

7. The method for evaluating the difficulty of game levels according to claim 6, characterized in that, In step S325, the actual close-to-the-situation data is processed using the following logic: in, This means that if there are at least two known endpoints, the number of steps is greater than s. min The probability of a specific state B occurring in the simulation of +k, where c corresponds to the aforementioned specific state. More than s min The number of events with a value of +k.

8. A game level difficulty assessment system, used to implement the game level difficulty assessment method according to any one of claims 1 to 7, characterized in that, The system includes: The game level module is used to obtain the current game level to be tested; The initial test data module is used to obtain random AI parameters and specific level completion parameters, and to set and use a preset random AI test tool to simulate the level completion operation of the current game level to be tested in order to obtain initial level completion test data. The initial test data module is connected to the game level module. A level completion difficulty module is used to correct the initial level-clearing test data using a preset probability model to obtain the current level completion difficulty data. The level completion difficulty module is connected to the initial test data module and the game level module. The level completion difficulty module includes: The "Real Situation Module" is used to obtain the number of steps and actions to pass the current game level to be tested, and then use the preset winning probability logic to process and obtain real situation data. The independent condition situation module is used to process the actual close situation data with preset condition independent logic, so that the conditions of the situation corresponding to the actual close situation data are independent, thereby obtaining independent condition situation data. The independent condition situation module is connected to the close actual situation module. The success probability module is used to process the independent condition situation data with preset success probability logic to obtain the probability corresponding to the number of success steps. The success probability module is connected to the independent condition situation module. The difficulty data processing module is used to perform a preset number of hypothetical experiments based on the probability corresponding to the number of steps to pass the level, in order to obtain applicable distribution hypothetical experimental data, and to process the applicable distribution hypothetical experimental data with preset conditional probability logic to obtain the level clearance conditional probability, thereby generating the current level clearance difficulty data. The difficulty data processing module is connected to the level clearance probability module. A personalized configuration module is used to personalize the game level to be tested based on the current level's difficulty data. The personalized configuration module is connected to the level difficulty module.

Images