System for calculating potential for player to leave game using artificial intelligence

Abstract

Disclosed is a system for calculating the potential for a player who is playing a game to leave the game by using an artificial intelligence algorithm. Training data includes play data which is data related to the play history for the game of the player, player information which is information about the personal characteristics of the player, and game data which is data representing the characteristics of the game. In addition, for supervised learning of an artificial intelligence algorithm, leave result data, which is data of an actual result of the player's leaving the game, is used for the learning. The potential for leaving is determined on the basis of a prescribed elapsed time from the final time point at which the player actually played the game. The advertising effectiveness can be increased by performing a targeted advertising for a player determined to have a high potential of leaving.

Description

A system that uses artificial intelligence to calculate the likelihood of players leaving the game.

[0001] The present invention relates to a system for calculating the probability of a player leaving a game using artificial intelligence, and to a system for predicting the time at which a player leaves a game played online or on an electronic device.

[0002] Advances in IT have led to a proliferation of diverse computer games, and many users are enjoying online games. As numerous game developers competitively release diverse titles, marketing strategies are becoming increasingly necessary for gamers. However, game developers seeking to market their games often struggle to understand the playstyles of a wide range of players, making it difficult to target specific game segments. Consequently, the marketing effort and costs involved often yield ineffective results. Furthermore, fraudulent manipulation, such as artificially increasing ad traffic, is prevalent in the online advertising market. This leads to increased costs for game developers, ultimately driving up the game fees charged to players.

[0003] Targeted marketing fundamentally requires analyzing player tendencies. Simply knowing how frequently and for how long a player plays a game allows you to target specific games to those players, increasing their effectiveness. Therefore, player tendencies analysis has primarily been conducted to identify games suited to each player's characteristics.

[0004] However, this propensity analysis alone still has limitations in the effectiveness of targeted advertising. If future play trends could be predicted based on a user's past play patterns, then these predictions could be used to deliver more relevant and effective targeted advertising.

[0005] Traditional play trend analysis primarily aimed to identify players likely to be interested in a specific game and then target them with ads for that game. Therefore, when predicting future play patterns, the concept of predicting the likelihood of a player churning from a game didn't exist. However, rather than advertising to encourage new games, marketing is more effective when it encourages continued play of existing games. Therefore, if it's possible to predict a player's likelihood of quitting a game, or churn, marketing efforts, such as offering various benefits before they churn, can be strengthened to encourage continued enjoyment of the game. Targeted advertising can also be enhanced by introducing similar new games to players at high risk of churning.

[0006] Taking these points into consideration, the present invention aims to provide a system that analyzes and predicts whether a player is likely to leave a game.

[0007] In order to achieve the above object, the present invention provides a system for calculating a player's probability of leaving a game using an artificial intelligence algorithm, the system comprising: an input unit for receiving learning data including play data, which is data related to the player's play history for the game, player information, which is information on the player's personal characteristics, and game data, which is data expressing the characteristics of the game, and departure result data, which is data of the player's actual departure from the game; a learning unit for learning the artificial intelligence algorithm based on the learning data and the departure result data input through the input unit; and a judgment unit for receiving the play data, the player information, and the game data for the player to be judged, and judging the player's probability of leaving the game using the artificial intelligence algorithm learned by the learning unit.

[0008] The system of the present invention may further include a vector generation unit that generates a plurality of learning vectors using the learning data input through the input unit and provides the vectors to the learning unit.

[0009] The above play data may include data on the proportion of the player's play time for the plurality of games, and log data, which is information on the player's log history for the plurality of games.

[0010] The above learning vector may be configured to include a combined vector created by combining a weight vector that vectorizes the weight data among the play data and a game vector that vectorizes the game data, a preference vector that vectorizes the player's preference for the game based on the log data among the play data, and a player information vector that vectorizes the player information.

[0011] The above player information may include one or more of the player's age, gender, type of device used to play the game, and type of communication network used to play the game.

[0012] The above game data may include category information indicating the classification of the game, and contextual information obtained through a pre-learned language model based on information describing the game.

[0013] The above game vector is obtained by combining a category vector that vectorizes the above category information and a context vector that vectorizes the above contextual information.

[0014] The above-mentioned exit result data may be composed of data calculated based on the elapsed time since the last time the player actually played the game while not playing. In this case, the exit result data may be composed of data determining a player has exited if the elapsed time from the last time is greater than a predetermined value.

[0015] The above-mentioned judgment unit may determine the possibility of withdrawal based on a predetermined amount of time elapsed since the last time the player actually played the game. The above-mentioned judgment unit may determine the possibility of withdrawal by dividing the amount of time elapsed since the last time the player actually played the game.

[0016] According to the present invention, the likelihood that a player will abandon a game they are currently playing can be predicted using a pre-trained artificial intelligence algorithm. This can be used to provide various benefits to players at high risk of abandonment, thereby encouraging them to continue playing the existing game in the future. Furthermore, targeted advertising for new games tailored to the player's preferences can be used to enhance advertising effectiveness.

[0017] Figure 1 is a schematic diagram illustrating a system for calculating a player's probability of leaving a game using artificial intelligence according to the present invention.

[0018] Figure 2 is a drawing showing the configuration of a vector generated by the vector generation unit of Figure 1.

[0019] Hereinafter, the present invention will be described in more detail with reference to the drawings.

[0020] In the following description, the components described as ~part, ~module, etc. may each be implemented by a single computer or server, or may be configured to be provided as separate functional modules within a single computer or server. In the description of the present invention, general components that are not directly related to the gist of the present invention or are typically self-evidently provided are omitted from illustration and description.

[0021] Figure 1 is a schematic diagram illustrating a system for calculating a player's probability of leaving a game using artificial intelligence according to the present invention.

[0022] The system of the present invention calculates the likelihood of a player leaving the game using an artificial intelligence algorithm. The system of the present invention comprises an input unit (10), a vector generation unit (20), a learning unit (30), an AI module (40), and a judgment unit (50).

[0023] The input unit (10) receives training data used by the learning unit (30) to train the artificial intelligence algorithm within the AI ​​module (40). The training data includes play data, player information, and game data, which will be described in detail later. In addition, the input unit (10) receives departure result data, which is data on the actual departure result of the player from the game.

[0024] The vector generation unit (20) uses learning data to generate learning vectors in a vector form suitable for learning by the AI ​​module (40). The learning vectors include a combination vector, a preference vector, and a player information vector, which will be described in detail later.

[0025] The learning unit (30) trains the AI ​​module (40) using the learning vectors generated by the vector generation unit (20) and the deviation result data input to the input unit (10).

[0026] The AI ​​module (40) is trained using the training vectors and deviation result data provided by the learning unit (30). The AI ​​module (40) can employ various types of artificial intelligence algorithms, and for example, can employ the CNN (Collective Neural Network) algorithm, which is known to be suitable for artificial intelligence learning of vectorized data.

[0027] The judgment unit (50) determines the possibility of the player being judged leaving the game using the AI ​​module (40) learned by the learning unit (30). The judgment unit (50) receives play data, player information, and game data regarding the player being judged, and provides this to the AI ​​module (40) to determine the possibility of leaving.

[0028] FIG. 2 is a drawing showing the configuration of a vector generated by the vector generation unit of FIG. 1, and is a drawing showing the configuration of data used for learning of the AI ​​module (40) and the process of vectorizing such data for use in learning.

[0029] The training data consists of play data, game data, player information, and dropout result data. The training data is used to generate training vectors.

[0030] Play data refers to a player's play history for each game. Play data is divided into weighted data and log data. The two types of data may be identical or similar in their underlying raw data, but they are extracted from different aspects of this raw data.

[0031] Prevalence data represents the proportion of a player's play time across multiple games. Specifically, prevalence data is data calculated for each player, estimating the proportion of time spent playing each game when the player plays one or more games. For example, if player "A" played Game (A) and Game (B) for 2 hours each on a specific day, player "A's" prevalence data would be Game (A) = 0.5, Game (B) = 0.5. If player "A" played Game (A) and Game (B) for 4 hours and 1 hour respectively the next day, player "A's" prevalence data would be Game (A) = 0.8, Game (B) = 0.2.

[0032] Log data is information about a player's log history for multiple games. Specifically, it can be used to determine changes in the player's play time for each game by looking at the time between the player's log-in and log-out. Log data can be fundamentally similar to weight data. While weight data describes the relative importance of each game, log data describes changes in the player's play time for each game. This log data captures changes in the player's game preferences. For example, if the play time for game (A) sequentially decreased from 5 hours to 1 hour on each day over a one-week period, while it sequentially increased from 2 hours to 6 hours for game (B), this indicates a shift in the player's preference from game (A) to game (B).

[0033] Player information is information about a player's personal characteristics. It primarily includes personal information about the player, such as gender, age, and occupation. It also includes information about the type of device used to play the game and the type of network used to play the game.

[0034] Game data is data that expresses the characteristics of a game, and includes game category information and contextual information about the game's description.

[0035] Category information is category information that indicates the classification of the game, and includes the game's genre (action, strategy, etc.), game's subgenre (io, tower defense, etc.), gameplay elements (competitive mode, daily and login rewards, leaderboards, achievements, lucky spins, and card scratching, etc.), monetization methods (in-app advertising, gacha / loot boxes, VIP, consumables, starter packs, etc.), drawing style (cartoon, landscape mode, 3D, 2D, etc.), camera view (quarter view, 2D, etc.), and fun elements (social bonding, visual fun, fun of discovery, etc.).

[0036] Contextual data is information derived from game descriptions, extracted using a pre-trained language model from descriptions that describe the game's features. Game descriptions are typically written by the game developer. For example, a game title and its description might look like this:

[0037] [Game Name]

[0038] Mysterious Wizard's Tower

[0039] [Game Description]

[0040] Welcome to the mystical Wizard's Tower, a world filled with magic and adventure! This game combines exploration, magic, and mystery to create a new level of adventure. Become a wizard and build your own tower, expand your magical knowledge, and master powerful spells. Mystical Wizard's Tower features a unique magic system and a deep crafting system that allows you to collect various magical ingredients and brew magical potions. You will also experience a vast world full of hidden knowledge and treasures to discover through exploration. Each wizard can customize their tower, raise a variety of magical creatures, and explore mysterious magical worlds. Cooperate or compete with other wizards in a large online community, and compete in wizarding tournaments. Win wizarding tournaments or complete special quests to earn rare items and spellbooks. This game is the ideal choice for anyone who dreams of living the life of a wizard. Write your own story in this mystical world filled with magic and adventure. Endless possibilities await you in Mystical Wizard's Tower!

[0041] By applying a natural language analysis AI algorithm to analyze these descriptions, information indicating the contextual characteristics of the game can be obtained, and the information obtained in this way is contextual information.

[0042] The attrition outcome data represents the actual attrition outcomes of each player from each game, corresponding to each training data set. By using this attrition outcome data for training, the actual attrition outcomes of each player with these values ​​are reflected along with the values ​​represented by the aforementioned training vectors, enabling learning.

[0043] The churn result data may be composed of data calculated based on the elapsed time since the last time a player actually played the game. Specifically, the churn result data is composed of data that determines a player has churned if the elapsed time from the last time is greater than a predetermined time. That is, in the present invention, the concept of "churn" refers to a state in which the player has not played the game for a predetermined period of time. For example, a player who has not played the game for three days is considered to have churned. This predetermined period may be set differently, such as one week or one month, as needed. The churn data may be composed in stages based on the elapsed time. For example, the elapsed time during which the player has not played the game may be set in stages, such as 3 days, 7 days, 14 days, and 30 days. If the player resumes playing after 15 days, the player may be treated as churned for 3, 7, and 14 days, but not churned for 30 days.

[0044] The vector generation unit (20) uses each of these data to generate multiple learning vectors to be used for learning. The learning vectors are composed of a combination vector, a preference vector, and a player information vector.

[0045] The combination vector is obtained by combining the weight vector and the game vector.

[0046] The weight vector is the vectorized weight data described above. For example, in the example described above, the weight data are obtained as [0.5, 0.5] and [0.8, 0.2] for each day for Game (A) and Game (B).

[0047] A game vector is obtained by concatenating the categorical vector and the contextual vector, which are vectorized categorical data and contextual data, respectively. For example, if the categorical vector is [0, 1, 1, 1, 0, 0, 1, .....] and the contextual vector is [0, 0, 1, 1, 0, 0, 1, .....], the two can be concatenated to create a game vector of [0, 1, 1, 1, 0, 0, 1, ....., 0, 0, 1, 1, 0, 0, 1, .....]. Such a game vector is created one by one for each game.

[0048] Combining the weight data and game vectors to obtain a combined vector can be accomplished in various ways. A simple example is the matrix multiplication of these two vectors. Specifically, by multiplying the weight vector [0.9, 0.1] obtained for a given player's game (A) and game (B) by the game vector for each game, a combined vector for each game of that player is obtained.

[0049] The combined vector obtained in this way has a value that represents the proportion of games with specific characteristics that the player actually played.

[0050] A taste vector is a vectorized version of log data. For example, it can be obtained as [0.9, 0.1] depending on the degree to which tastes for Game (A) and Game (B) gradually increase.

[0051] A player information vector is a vectorized representation of player information. A player information vector is obtained by assigning values ​​to each factor that constitutes the player information.

[0052] The learning unit (30) uses each of these learning vectors to train the AI ​​module (40). As the combined vector, preference vector, and play information vector are used for learning by the learning unit (30), learning is performed based on data that combines actual play proportions according to the characteristics of the game, changes in player preferences, and player characteristics. In addition, the learning unit (30) uses the attrition result data along with the learning vector to perform supervised learning. Therefore, learning is performed that reflects the attrition results of actual players who obtained the learning data.

[0053] The judgment unit (50) uses the AI ​​module (40) to determine the likelihood of a player being judged to be a potential candidate for leaving the game. To this end, the judgment unit (50) receives judgment information about the player being judged. The judgment information consists of the same data as the learning data described above.

[0054] As an example, the judgment unit (50) determines the likelihood of a player leaving the game based on a predetermined amount of time elapsed since the last time the player actually played the game. For example, if the elapsed time is set to 3 days, the likelihood that the player will not play for more than 3 days is determined to be, for example, 70%. This method is suitable for cases where the elapsed time determined as leaving is set to 3 days in the departure result data input during the aforementioned learning process.

[0055] As another example, the judgment unit (50) may be configured to determine the likelihood of a player leaving the game based on the elapsed time since the last time the player actually played the game. For example, the likelihood of the player leaving the game may be determined as 30% after 3 days, 50% after 7 days, 60% after 14 days, 70% after 30 days, etc. This method is suitable for cases where the elapsed time determined as leaving is set in stages, such as 3 days, 7 days, 14 days, and 30 days, in the departure result data input during the aforementioned learning process.

[0056] According to the present invention, the likelihood that a player will abandon a game they are currently playing can be predicted using a pre-trained artificial intelligence algorithm. This can be used to provide various benefits to players at high risk of abandonment, thereby encouraging them to continue playing the existing game. Furthermore, targeted advertising for new games tailored to the player's preferences can be used to enhance advertising effectiveness.

[0057] While the above description has been made with reference to one embodiment illustrated in the drawings, this is merely exemplary, and those skilled in the art will understand that various modifications and equivalent embodiments are possible. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the appended claims.

Claims

1. A system for calculating the probability of a player leaving a game using artificial intelligence, for calculating the probability of a player leaving a game using an artificial intelligence algorithm. An input unit that receives learning data including play data, which is data related to the player's play history for the game, player information, which is information on the player's personal characteristics, and game data, which is data expressing the characteristics of the game, and departure result data, which is data on the actual departure result of the player from the game; A learning unit that trains the artificial intelligence algorithm based on the learning data and the departure result data input through the input unit; and A judgment unit that receives the play data, player information, and game data for the player to be judged, and judges the possibility of the player leaving the game using the artificial intelligence algorithm learned by the learning unit; A system for calculating the probability of a player leaving a game using artificial intelligence, characterized by including:

2. In paragraph 1, A vector generation unit that generates a plurality of learning vectors using the learning data input through the input unit and provides the vectors to the learning unit; A system for calculating the probability of a player leaving a game using artificial intelligence, characterized by further including:

3. In paragraph 1, A system for calculating the probability of a player leaving a game using artificial intelligence, characterized in that the play data includes data on the proportion of the player's play time for a plurality of the games and log data, which is information on the player's log history for a plurality of the games.

4. In paragraph 3, The above learning vector is, A combined vector created by combining the weight vector obtained by vectorizing the weight data among the above play data and the game vector obtained by vectorizing the game data, A taste vector that vectorizes the player's taste for the game based on the log data among the play data, and A player information vector that vectorizes the above player information, A system for calculating the probability of a player leaving a game using artificial intelligence, characterized by including:

5. In paragraph 4, A system for calculating the probability of a player leaving a game using artificial intelligence, characterized in that the player information includes at least one of the player's age, gender, type of device used to play the game, and type of communication network used to play the game.

6. In paragraph 5, A system for calculating the probability of a player leaving a game using artificial intelligence, characterized in that the game data includes category information indicating the classification of the game and contextual information obtained through a pre-learned language model based on information describing the game.

7. In paragraph 6, A system for calculating a player's likelihood of leaving a game using artificial intelligence, characterized in that the above game vector is obtained by combining a category vector that vectorizes the above category information and a context vector that vectorizes the above contextual information.

8. In any one of paragraphs 1 to 7, A system for calculating the probability of a player leaving a game using artificial intelligence, characterized in that the above-mentioned exit result data is data calculated based on the elapsed time from the last time the player actually played the game until the player did not play.

9. In paragraph 8, A system for calculating the probability of a player leaving a game using artificial intelligence, characterized in that the above-mentioned exit result data is data of a result determined to be an exit when the elapsed time from the final point in time is greater than a predetermined value.

10. In any one of paragraphs 1 to 7, A system for calculating a player's probability of leaving the game using artificial intelligence, characterized in that the judgment unit determines the probability of leaving based on a predetermined amount of time elapsed from the last time the player actually played the game.

11. In any one of paragraphs 1 to 7, A system for calculating the probability of a player leaving a game using artificial intelligence, characterized in that the judgment unit determines the probability of leaving by dividing it by the elapsed time from the last time the player actually played the game.

Images