JP7587348B2 - Electronic game evaluation device, electronic game evaluation program, and electronic game evaluation method - Google Patents

Description

The present invention relates to an electronic game evaluation device, an electronic game evaluation program, and an electronic game evaluation method.

There are known electronic games in which a player plays a game using a number of different characters with different game parameters. In such electronic games, the game parameters assigned to the characters are a major factor in determining the game's progress. Therefore, the presence of an extremely strong character can upset the balance of the electronic game's progress.

Therefore, when a new character is introduced into an electronic game, it is necessary to evaluate the impact of introducing the character on the electronic game. For example, a method is used in which the new character is actually made available for use in an electronic game, played, a play log is accumulated, and the character is evaluated through supervised learning using the play log. Also, a method is used in which reinforcement learning is used to train an artificial intelligence agent that plays an electronic game using the character (Patent Document 1).

JP 2019-197592 A

However, with conventional technology, it is not possible to evaluate a new character without actually introducing it into a game. Furthermore, when using reinforcement learning, it is difficult to improve the accuracy of learning, and there is a risk that learning will take an enormous amount of time. Therefore, there is a need for technology that can effectively evaluate the characteristics of a new character before introducing it.

One aspect of the present invention is an electronic game evaluation device for evaluating a competitive electronic game, which includes an expression vector search means for clustering feature vectors representing objects to be evaluated, determining a representative vector from the set of clustered feature vectors, selecting one of the representative vectors as an alternative vector based on an expected reward distribution indicating the probability of an expected reward to be obtained in an electronic game for an object corresponding to the representative vector, and setting the object corresponding to the alternative vector as a new object, a match evaluation means for performing a match evaluation between a state in the electronic game including the new object and a state in the electronic game not including the new object using an existing strategy learner, and an expected reward distribution update means for updating the expected reward distribution for the representative vector selected as the alternative vector according to the result of the match evaluation by the match evaluation means, and is characterized in that the electronic game evaluation device repeats the processes in the expression vector search means, the match evaluation means, and the expected reward distribution update means.

Another aspect of the present invention is an electronic game evaluation program for evaluating a competitive electronic game, which causes a computer to function as an expression vector search means for clustering feature vectors indicating an object to be evaluated, determining a representative vector from the set of clustered feature vectors, selecting one of the representative vectors as an alternative vector based on an expected reward distribution indicating the probability of an expected reward to be obtained in an electronic game for the object corresponding to the representative vector, and setting the object corresponding to the alternative vector as a new object, a match evaluation means for performing a match evaluation between a state in the electronic game including the new object and a state in the electronic game not including the new object using an existing strategy learner, and an expected reward distribution update means for updating the expected reward distribution for the representative vector selected as the alternative vector according to the result of the match evaluation by the match evaluation means, and is characterized in that the electronic game evaluation program repeats the processes in the expression vector search means, the match evaluation means, and the expected reward distribution update means.

Another aspect of the present invention is an electronic game evaluation method for evaluating a competitive electronic game, comprising: an expression vector search step of clustering feature vectors representing an object to be evaluated, determining a representative vector from the set of clustered feature vectors, selecting one of the representative vectors as an alternative vector based on an expected reward distribution indicating the probability of an expected reward to be obtained in an electronic game for an object corresponding to the representative vector, and setting the object corresponding to the alternative vector as a new object; a match evaluation step of using an existing strategy learner to perform a match evaluation between a state in the electronic game including the new object and a state in the electronic game not including the new object; and an expected reward distribution update step of updating the expected reward distribution for the representative vector selected as the alternative vector according to the result of the match evaluation in the match evaluation step, and the electronic game evaluation method is characterized in that the process in the expression vector search step, the match evaluation step, and the expected reward distribution update step are repeated.

Here, it is preferable to provide an evaluation means for presenting to the user statistical values based on the results of the match evaluation obtained by the match evaluation means.

Furthermore, it is preferable that the expected reward is an expected win rate in an electronic game, and the expected reward distribution update means updates the expected reward distribution for the representative vector according to the outcome of a match in the match evaluation means.

It is also preferable that the object is a character used in an electronic game, and the battle evaluation means performs a battle evaluation when a deck in the electronic game that combines a new character as the new object with an existing character is pitted against a deck in the electronic game that does not include the new character and combines existing characters.

One of the objectives of the embodiments of the present invention is to realize an electronic game evaluation device, an electronic game evaluation program, and an electronic game evaluation method that can effectively evaluate the characteristics of a new character before the character is introduced. Other objectives of the embodiments of the present invention will become clear by referring to this specification as a whole.

1 is a diagram showing a configuration of an electronic game evaluation device in an embodiment of the present invention. 2 is a diagram illustrating an expression learning unit and a strategy learning device (battle AI) in an embodiment of the present invention. FIG. 1 is a flowchart showing an electronic game evaluation method according to an embodiment of the present invention. 5A to 5C are diagrams illustrating examples of expression vectors of each character according to an embodiment of the present invention. FIG. 13 is a diagram showing an example of an expected reward distribution of a representative vector in an embodiment of the present invention. 11A to 11C are diagrams illustrating a selection process of an alternative vector according to an embodiment of the present invention. 11A to 11C are diagrams illustrating a deck search process according to an embodiment of the present invention. 13A to 13C are diagrams illustrating the process of constructing a deck including a new character in an embodiment of the present invention. FIG. 13 is a diagram illustrating the update process of the expected reward distribution in the embodiment of the present invention. FIG. 11 is a diagram showing an example of an evaluation of an electronic game according to an embodiment of the present invention. FIG. 11 is a diagram showing an example of an evaluation of an electronic game according to an embodiment of the present invention. FIG. 11 is a diagram showing an example of an evaluation of an electronic game according to an embodiment of the present invention. FIG. 11 is a diagram showing an example of an evaluation of an electronic game according to an embodiment of the present invention.

[Device configuration]
As shown in FIG. 1, the electronic game evaluation device 100 in the embodiment of the present invention includes a processing unit 10, a storage unit 12, an input unit 14, an output unit 16, and a communication unit 18. The processing unit 10 includes a means for performing arithmetic processing such as a CPU. The processing unit 10 executes an electronic game evaluation program stored in the storage unit 12 to realize the evaluation process in the electronic game in the embodiment. The storage unit 12 includes a storage means such as a semiconductor memory or a memory card. The storage unit 12 is connected to the processing unit 10 so as to be accessible, and stores the electronic game evaluation program and information required for the processing. The input unit 14 includes a means for inputting information. The input unit 14 includes, for example, a keyboard, a touch panel, a button, etc. for receiving input from an administrator. The output unit 16 includes a means for outputting the processing result in the electronic game evaluation device 100, such as a user interface screen (UI) for receiving input information from the administrator. The output unit 16 includes, for example, a display for presenting an image to the administrator. The communication unit 18 includes an interface for communicating information with an external user terminal, etc., via the network 102. The communication by the communication unit 18 may be wired or wireless.

[Electronic Game Evaluation Processing]
In this embodiment, an electronic game evaluation process that enables an appropriate evaluation of the possible impact on a competitive electronic game when a new character is introduced into the electronic game will be described. Specifically, an electronic game evaluation program that executes the following process is executed in the electronic game evaluation device 100, and a process is performed to present the impact of the new character on the electronic game to the game planner, etc., using a multi-armed bandit method.

As an example of an electronic game, we will explain a game in which each player creates a deck consisting of a group of multiple characters, and the characters that make up each player's deck compete against each other to determine the winner. However, the scope of application of the electronic game evaluation method in this embodiment is not limited to this.

The electronic game evaluation device 100 uses an artificial intelligence agent for playing an electronic game using multiple characters. The artificial intelligence agent can be, for example, an information processing device described in JP 2019-197592 A. Specifically, as shown in FIG. 2, expression learning is performed based on a play log (battle log), which is data showing the progress history of a game played in an electronic game in the past. This makes it possible to obtain an expression learning unit that outputs a feature vector (hereinafter referred to as an expression vector) showing the features of each character by inputting the game parameters of each character used in the electronic game. In addition, using the feature vectors of each character learned by the expression learning unit, learning of an electronic game strategy regarding the use of each character (or the use of a deck including each character) is performed. This makes it possible to obtain a strategy learning unit (battle AI) that outputs a preferred strategy, win/loss rate, etc. in an electronic game when each character is used (or a deck including each character is used). In the electronic game evaluation device 100 of this embodiment, such expression learning unit and strategy learning unit are used to generate and output information for evaluating the impact of introducing a new character into an electronic game.

The electronic game evaluation method executed by the electronic game evaluation device 100 will be described below with reference to the flowchart in Figure 3.

In step S10, a search process for an expression vector of a new character is performed. Through the process in this step, the electronic game evaluation device 100 functions as an expression vector search means.

The processing unit 10 clusters expression vectors for existing characters obtained by expression learning for past play logs (battle logs). For example, k-means processing can be applied to the expression vector clustering process. However, this is not limited to this, and other clustering processing methods may also be applied.

Figure 4 shows an example of multiple existing characters in an electronic game being represented by their respective expression vectors using an existing expression learning unit. Figure 4 shows an example of the expression vectors of each character being represented by two-dimensional expression vectors (dots in the figure). However, the dimensions of the vectors representing the expression vectors are not limited to this.

After clustering the expression vectors, a representative vector is determined from each cluster. The representative vector is preferably the centroid (center of gravity vector) of each cluster. Note that the representative vector can also be selected by randomly selecting a character from each cluster, or by using a bandit algorithm or the like from the characters within a cluster.

Figure 4 also shows the results of clustering the expression vectors of multiple existing characters in an electronic game. In Figure 4, each cluster is shown surrounded by an ellipse. The centroid (center of gravity vector) of each cluster is also shown with a star-shaped mark. Therefore, the representative vector of each cluster is represented as the expression vector shown with a star-shaped mark.

Next, an expected reward distribution is generated for each representative vector. The expected reward distribution is a distribution that indicates the probability that an expected reward for an electronic game for each representative vector of each cluster will occur in the electronic game. For example, if the expected reward is the expected win rate, the distribution indicates the probability that the expected win rate will be obtained when a character having the representative vector of each cluster is used in an electronic game. The expected reward is not particularly limited as long as it is a value related to an electronic game, but in addition to the expected win rate, it can be the expected maximum offensive power, the expected maximum defensive power, the expected occurrence rate of some in-game skill, etc.

As shown in Figure 5, in this embodiment, an example will be described in which the expected element value is the expected win rate. In Figure 5, the probability of obtaining the expected win rate for the representative vector in each cluster is shown as a mountain-shaped probability distribution.

The probability distribution may be set appropriately, but it is preferable to apply the beta distribution (β distribution) shown in Equation (1). In the beta distribution, the initial values of the parameters α and β are 1. The expected reward distribution is changed by an update process described later. However, the specific expected reward distribution and update process are not limited to this, and a frequency count or frequency distribution based on a binomial distribution or trial results may be applied.

Next, for each cluster, the expected reward for each representative vector is calculated from the expected reward distribution, and the representative vector with the highest expected reward is selected as the replacement vector for the newly introduced character. The selection may be made probabilistically based on the Boltzmann distribution or beta distribution with reference to the expected reward. In other cases, the selection may be made deterministically using an evaluation function such as UCB1 or its derivatives. Figure 6 shows an example of calculating the expected reward for the representative vectors in each cluster of the expression vectors shown in Figure 4. In this case, representative vector 1 with the highest expected reward is selected as the replacement vector for the new character.

The expected reward and selection conditions may be determined depending on what kind of expected reward a new character is evaluated for in the expected reward that is the subject of evaluation of the electronic game. In the case of this embodiment, the win rate is a representative evaluation subject, and the purpose is to evaluate changes in the electronic game when a new character that increases the win rate is introduced, so the expected reward is set as the win rate, and the representative vector with the maximum expected reward is set as the condition for selecting as an alternative vector for the new character. For example, if the damage inflicted on an enemy character is a representative evaluation subject, and the purpose is to evaluate changes in the electronic game when a new character that increases the damage is introduced, the representative vector with the maximum damage may be set as the condition for selecting as an alternative vector based on the expected reward distribution that indicates the probability of inflicting the damage as the expected reward.

In step S12, a search process is performed for a deck that includes a new character. Through the process in this step, the electronic game evaluation device 100 functions as a deck search means.

In this step, first, a process is performed to prepare a deck into which the new character represented by the alternative vector selected in step S10 will be incorporated. The deck to be prepared is a deck that is constructed by combining multiple existing characters. In this embodiment, as shown in FIG. 7, multiple decks that are candidates for incorporating the new character are prepared.

For example, multiple decks that combine existing characters are prepared according to the intentions of the electronic game planner. Also, multiple decks are prepared that include existing characters that are thought to be similar to the new characters according to the planner's intentions for using new characters in the electronic game (such as a strategy using a new character in an electronic game). Also, multiple decks may be selected from decks that have been used in the past from the play log of the electronic game. In this case, for example, decks with a high win rate or decks that are likely to be used for a specific strategy may be extracted from the characteristics in the play log.

Next, as shown in FIG. 7, an expected reward distribution is generated for each existing character that composes each prepared deck. Then, one of the prepared decks is randomly selected. Then, as shown in FIG. 8, the expected reward for each character is calculated from the expected reward distribution generated for each existing character that composes the selected deck, and the character with the highest expected reward is replaced with the new character selected in step S10 to generate a deck including a new character. In the example of FIG. 8, the expected reward of character 1 is 0.9, which is the highest value among characters 1 to k included in the randomly selected deck 1, so character 1 is replaced with the new character in deck 1 to construct a new deck.

Note that this step may be omitted if the electronic game does not use a deck that combines multiple characters, but instead involves individual characters battling each other.

In step S14, a match evaluation process is performed between the deck containing the new character and another deck. Through the process in this step, the electronic game evaluation device 100 functions as a match evaluation means.

The processing unit 10 selects one deck that does not include a new character from the decks prepared in step S12. The deck in question may be selected randomly from the decks prepared in step S12. Then, using a strategy learning unit (battle AI) that has been machine-learned using past play logs, the deck selected here is pitted against the deck that includes the new character generated in step S12, and the results of the battle are obtained.

Match results include any information obtained when an electronic game match is actually played, such as the outcome of the match between the decks, the outcome of the match between the characters, the parameter values of each character at the end of the match (e.g., hit points, attack power, defense power, etc.), the damage each character inflicted on the opponent's character, and the damage each character was dealt, etc.

In the case of an electronic game in which a deck is not constructed and characters battle each other, the new character selected in step S10 is pitted against an existing character, and the battle results are obtained.

In step S16, an expected reward distribution update process is performed. Through the process in this step, the electronic game evaluation device 100 functions as an expected reward distribution update means.

The processing unit 10 uses the battle results in step S14 to update the expected reward distribution for the characters that make up the deck, including the new character. Specifically, the processing unit 10 updates the expected reward distribution for the representative vector for the cluster corresponding to the alternative vector selected as the new character in step S10. For example, if representative vector 1, which has the highest expected reward, is selected as the alternative vector for the new character in step S10, the processing unit 10 performs processing to update the expected reward distribution for representative vector 1.

In addition, for other characters that make up the deck including the new character, the expected reward distribution for the expression vector for that character may be updated, taking into consideration the possibility of a synergistic effect with the new character. In addition, using the battle results in step S14, the expected reward distribution for the characters in the original deck that were replaced with the deck including the new character prepared in step S12 may be updated.

At this time, if the match result in step S14 is a win, the expected reward distribution is updated to a probability distribution in which the expected reward increases. Also, if the match result in step S14 is a loss, the expected reward distribution is updated to a probability distribution in which the expected reward decreases.

When the expected reward distribution is expressed by a beta distribution, the expected reward distribution is updated by changing the parameters α and β of the beta distribution according to the match result in step S14. If the match result in step S14 is a win, a process is performed to increase the parameter α by a predetermined adjustment value. The predetermined adjustment value can be, for example, 1. Furthermore, if the match result in step S14 is a loss, a process is performed to decrease the parameter β by a predetermined adjustment value. The predetermined adjustment value can be, for example, +1. This makes it possible to update the expected reward distribution so that the expected reward in the expected reward distribution increases when the match result is a win, and the expected reward in the expected reward distribution decreases when the match result is a loss.

In the case of electronic games in which characters battle each other without constructing a deck, it is only necessary to update the expected reward distribution of the representative vector that corresponds to the replacement vector of the new character.

In step S18, an end determination process is performed. Through the process in this step, the electronic game evaluation device 100 functions as an end determination means.

The processing unit 10 determines whether or not the condition for terminating the repetition of the processing of steps S10 to S16 is satisfied, and if the termination condition is satisfied, the processing proceeds to step S20, whereas if the termination condition is not satisfied, the processing returns to step S10. When the processing returns to step S10, the expected reward for each representative vector is calculated from the expected reward distribution for each cluster based on the expected reward distribution updated in step S16, and the representative vector with the highest expected reward is newly selected as the replacement vector for the new character. Then, the processing of steps S12 to S16 is performed again for the newly selected new character.

In this way, by repeating the processes of steps S10 to S16 multiple times, the expected reward distribution of the representative vector corresponding to the alternative vector of the new character is updated based on the results of the battle evaluation in step S14 for the deck including the new character that is reselected each time. That is, as shown in FIG. 9, the parameters of the expected reward distribution change and the expected reward distribution is updated. As the battle evaluation is repeated, the variance of the expected reward distribution becomes smaller, and when the expected win rate increases, the position of the peak of the expected reward distribution moves toward the higher expected reward, and when the expected win rate decreases, the position of the peak of the expected reward distribution moves toward the lower expected reward. And, the more the position of the peak of the expected reward distribution moves toward the higher expected reward, the more likely it is to be selected as an alternative vector in step S10, and the more times it is evaluated for battle. Note that when the outcome is a 50-50 split, the expected reward distribution peaks in the middle (around 0.5 in the example of FIG. 9), and shows a distribution that is greatly spread out with respect to the value of the expected reward. Therefore, when an expected reward distribution with such an expected reward distribution begins to exist with a high expected reward peak, it becomes difficult for the representative vector with such an expected reward distribution to be selected as an alternative vector in step S10.

The termination condition for this step may be an upper limit set for the time or number of times steps S10 to S16 are repeated, and the process may terminate when the upper limit is exceeded. However, the termination condition is not limited to this.

In step S20, processing related to the evaluation of the new character is carried out. Through the processing in this step, the electronic game evaluation device 100 functions as an evaluation means. Here, the processing related to the evaluation is carried out using the statistical index obtained by applying the multi-bowl bandit realized in the above steps S10 to S18. For example, the processing is carried out to present the obtained statistical index (statistical value) to a user such as a planner, or to display a warning if the obtained statistical index exceeds a predetermined reference value.

For example, as shown in FIG. 10, when a deck is constructed by replacing each character constituting each pre-prepared deck with a new character, the evaluation value of the deck with the highest win rate in the electronic game can be presented to the user.

In addition, for example, as shown in FIG. 11, when a deck is constructed in which each character constituting each pre-prepared deck is replaced with a new character, the deck with the highest win rate in an electronic game is plotted on the vertical axis, and the deck that does not include new characters and is composed only of existing characters is plotted on the horizontal axis, and information showing the win rate for each combination can be presented to the user. In FIG. 11, combinations with a high win rate for decks including new characters compared to decks containing only existing characters are shown with a low brightness, and combinations with a low win rate are shown with a high brightness.

This information makes it possible to extract decks that include new characters that have a uniformly high win rate compared to decks that do not include the new character. Since such decks may upset the balance of the electronic game, it is possible to make decisions such as not introducing new characters that would allow the construction of such decks.

In addition, as shown in FIG. 12, it is also possible to extract a new method of using characters that increases the winning rate. For example, it is possible to know information such as how the winning rate changes depending on the order in which characters are used in a deck that is replaced with new characters, compared to the winning rate when characters are used in a specified order in a deck that is composed only of existing characters.

It is also possible to identify in advance which decks will have a higher average win rate if they include new characters, as shown in Figure 13.

In the above evaluation, the winning rate for the combination of decks is shown, but there is no particular limitation as long as it is information that can be used to evaluate the electronic game and that can be obtained in the battle evaluation in step S14, such as the remaining hit point difference after the battle, the difference in maximum attack power, etc.

In the case of electronic games in which characters battle each other without constructing a deck, the difference in winning rate between the new character and other existing characters in the electronic game can be evaluated.

In this embodiment, the character used in the electronic game is the subject of evaluation, but other objects used in the electronic game can also be the subject of evaluation. For example, the skills of a character in an electronic game (parameters that give attack power, defense power, and other game effects), items in an electronic game, and other objects used in the electronic game can be the subject of evaluation. In this case, the object to be evaluated can be applied instead of the character in the above explanation and processed.

In addition, in the electronic game evaluation device 100 of this embodiment, the expression vector search means, deck search means, battle evaluation means, expected reward distribution update means, end determination means, and evaluation means are realized in a single computer system. However, these means may be realized in a distributed manner in different devices or different execution entities. For example, some of these means may be realized by sharing the functions of multiple computers that can communicate with each other.

REFERENCE SIGNS LIST 10 Processing unit, 12 Memory unit, 14 Input unit, 16 Output unit, 18 Communication unit, 100 Electronic game evaluation device, 102 Network.

Claims (6)

An electronic game evaluation device for evaluating a competitive electronic game using an object, comprising:
an expression vector search means for determining a representative vector from a set of clustered feature vectors representing a plurality of existing objects, which is different from the existing objects , selecting the representative vector having the highest expected reward as an alternative vector based on an expected reward distribution indicating the probability of an expected reward of a predetermined type to be obtained in the electronic game when the object corresponding to the representative vector is used in the electronic game, and designating the object corresponding to the alternative vector as a new object;
a match evaluation means for evaluating an electronic game between a deck including the new object and a deck not including the new object, using an existing strategy learner;
an expected reward distribution update means for updating an expected reward distribution for the predetermined type of reward for the representative vector selected as the alternative vector in accordance with the predetermined type of reward obtained as a result of the battle;
Equipped with
Setting an initial value of the expected reward distribution for the predetermined type of reward for the representative vector, and selecting one of the representative vectors as the alternative vector;
In the match evaluation means, an object corresponding to the selected alternative vector is set as the new object, and an electronic game match is performed between a deck including the new object and a deck not including the new object;
the expected reward distribution update means updates an expected reward distribution for the representative vector selected as the alternative vector according to the predetermined type of reward obtained as a result of the battle;
the representative vector search means selects, as a new alternative vector, the representative vector having the highest expected reward based on an expected reward distribution for the predetermined type of reward for the updated representative vector;
The electronic game evaluation device is characterized in that the above process is repeated to evaluate the change in balance of the electronic game when the new object is added. 2. The electronic game evaluation device according to claim 1,
An electronic game evaluation device comprising evaluation means for presenting to a user a statistical value based on the results of the match evaluation obtained by said match evaluation means. 3. The electronic game evaluation device according to claim 1,
the predetermined type of reward is a winning rate in an electronic game, and the expected reward is an expected winning rate in the electronic game;
The electronic game evaluation device, wherein the expected reward distribution update means updates the expected reward distribution for the representative vector in accordance with the outcome of the match in the match evaluation means. The electronic game evaluation device according to any one of claims 1 to 3,
the object is a character used in an electronic game,
The electronic game evaluation device is characterized in that the battle evaluation means performs a battle evaluation when a deck in an electronic game that combines a new character and an existing character as the new object is pitted against a deck in an electronic game that does not include the new character and combines existing characters. An electronic game evaluation program for evaluating a competitive electronic game using objects, comprising:
Computer,
an expression vector search means for determining a representative vector from a set of clustered feature vectors representing a plurality of existing objects, which is different from the existing objects , selecting the representative vector having the highest expected reward as an alternative vector based on an expected reward distribution indicating the probability of an expected reward of a predetermined type to be obtained in the electronic game when the object corresponding to the representative vector is used in the electronic game, and designating the object corresponding to the alternative vector as a new object;
a match evaluation means for evaluating an electronic game between a deck including the new object and a deck not including the new object, using an existing strategy learner;
an expected reward distribution update means for updating an expected reward distribution for the predetermined type of reward for the representative vector selected as the alternative vector in accordance with the predetermined type of reward obtained as a result of the battle;
Function as a
Setting an initial value of the expected reward distribution for the predetermined type of reward for the representative vector, and selecting one of the representative vectors as the alternative vector;
In the match evaluation means, an object corresponding to the selected alternative vector is set as the new object, and an electronic game match is performed between a deck including the new object and a deck not including the new object;
the expected reward distribution update means updates an expected reward distribution for the representative vector selected as the alternative vector according to the predetermined type of reward obtained as a result of the battle;
the representative vector search means selects, as a new alternative vector, the representative vector having the highest expected reward based on an expected reward distribution for the predetermined type of reward for the updated representative vector;
The electronic game evaluation program is characterized in that the above process is repeated to evaluate a change in balance of the electronic game when the new object is added. 1. A method for evaluating a competitive electronic game using an object, comprising:
determining a representative vector from a set of clustered feature vectors representing a plurality of existing objects, the representative vector being a vector different from the existing objects; setting an initial value of the expected reward distribution for the predetermined type of reward for the representative vector; selecting one of the representative vectors as the alternative vector; and setting the object corresponding to the alternative vector as a new object;
a match evaluation step of playing an electronic game between a deck including the new object and a deck not including the new object by using an existing strategy learner;
an expected reward distribution update step of updating an expected reward distribution regarding the predetermined type of reward for the representative vector selected as the alternative vector according to the predetermined type of reward obtained as a result of the battle;
a representation vector search step of selecting the representative vector having the highest expected reward based on the expected reward distribution for the predetermined type of reward for the updated representative vector as the new alternative vector;
and evaluating a change in balance of the electronic game when the new object is added by repeating the above steps.

Images