Gaming device and gaming system

Abstract

To improve the accuracy of an operation prediction and an operation plan.SOLUTION: A game device such as a server 6 capable of outputting information on a game machine 1 installed in a game hall, includes: a storage unit 63 (storage means) capable of storing feature information indicating features of the game machine 1 and past game information indicating the operation record of the game machine 1; and model generative means capable of generating a model M that outputs predicative information on the operation of the game machine 1 by performing learning on the relation between the feature information stored in the storage unit 63 and the past game information. The feature information is configured to include image information on a housing of the game machine 1.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a gaming device and a gaming system that can output information about gaming machines installed in an amusement facility. [Background technology]

[0002] 2. Description of the Related Art Gaming devices and gaming systems capable of outputting information about gaming machines installed in gaming parlors are known. For example, Patent Document 1 discloses a system that performs a simulation based on actual performance values ​​over a predetermined period of time. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2003-152397 Summary of the Invention [Problem to be solved by the invention]

[0004] However, there is room for improvement in conventional systems. [Means for solving the problem]

[0005] In order to achieve the above object, the gaming device of the present invention is a gaming device capable of outputting information about gaming machines installed in an amusement facility, and includes a storage means capable of storing characteristic information indicating the characteristics of the gaming machines and past gaming information indicating the operating performance of the gaming machines, and a storage means for storing the characteristic information and past gaming information by learning the relationship between the characteristic information and the past gaming information. In response to the input of the characteristic information, Regarding the operation of the gaming machine Gaming information Forecast Information as and a model generating means for generating a model to be output, wherein the feature information includes to Regarding Includes the number of characters in the SNS information It is structured as follows. [Brief explanation of the drawings]

[0006] [Figure 1] 1 is a schematic diagram of a gaming system according to one embodiment of the present invention. [Figure 2] 1 is a block diagram showing the configuration of a server which is an example of a gaming device of the present invention; [Figure 3] 10 is a chart showing an example of past gaming information (operational performance) at each store. [Figure 4] 10 is an example of characteristic information. [Figure 5] FIG. 1 is a schematic diagram of a neural network. [Figure 6] FIG. 1A is an explanatory diagram of a method for generating an out-prediction model, and FIG. 1B is a diagram showing an out-prediction model generated by the method of FIG. [Figure 7] 10 is an example of a housing image. [Figure 8] This is an example of SNS information. [Figure 9] 10 is a flowchart showing a procedure for generating a model. [Figure 10] 10 is a flowchart showing a procedure for creating an operation plan. [Figure 11] 10A is a diagram showing an example of an S plan screen. (a) is the S plan screen before the planned date and the number of units to be allocated are input, (b) is the S plan screen after the planned date and the number of units for setting 1 (2 units) are input, and (c) is the S plan screen after the number of units for all settings are input. [Figure 12] 10A and 10B are diagrams illustrating an example of a P plan screen, in which (a) is the P plan screen before a planning date is input, and (b) is the P plan screen after a planning date is input. [Figure 13] This is the S Plan screen when a specific month is set as the planning month. [Figure 14] This is the P plan screen when a specific year is set as the planning year. DETAILED DESCRIPTION OF THE INVENTION

[0007] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a schematic diagram of a gaming system according to one embodiment of the present invention. As shown in Figure 1, the gaming system of this embodiment is composed of gaming machines 1 installed in each gaming parlor (store), a hall computer 4 that can acquire, compile, and output gaming information related to the gaming machines 1, a server 6 that can acquire, compile, and output gaming information related to the gaming machines 1 installed in each store by communicating with the hall computer 4 etc. on a regular or irregular basis, and a terminal device 7 such as a smartphone or personal computer used by a store manager. In this gaming system, prediction information (predicted operation value) regarding the operation of the gaming machine 1 is output based on past gaming information (actual operation value) indicating the operation record of the gaming machine 1 installed in each arcade. Furthermore, the gaming system creates store business plans, such as operation plans for specific days or specific periods, by carrying out simulations based on the output operation prediction values.

[0008] As shown in FIG. 1, each store is equipped with gaming machines 1 (pachinko machines 1a, pachislot machines 1b). A rental machine 2 (2a for pachinko machines, 2b for pachislot machines) that rents gaming media to players is located adjacent to the gaming machine 1, and players play according to the following flow. For example, when starting a game, a player inserts cash or a predetermined card into the lending machine 2 located adjacent to the gaming machine 1 and performs a predetermined lending operation to lend gaming media. The lent gaming media are then inserted into the gaming machine 1. In the pachinko machine 1a, the player operates a launch handle to launch game balls, which are game media, into the game area, and when the launched game balls enter a predetermined winning slot, a predetermined number of game balls are paid out. In the pachislot machine 1b, when the player inserts a medal as the gaming medium and operates the start lever, multiple reels displaying predetermined pictures, letters, numbers, etc. begin to rotate, and the reels can be stopped by pressing multiple stop buttons corresponding to each reel at any time, and a predetermined number of medals are paid out according to the arrangement of the patterns on the stopped reels. A rental rate for the gaming machine 1 is set for each store. For example, some stores have pachinko machines 1a (1-yen pachinko) with a rental rate of 1 yen per game ball, and other stores have pachinko machines 1a (4-yen pachinko) with a rental rate of 4 yen per game ball. There are also pachislot machines 1b (5 yen slots) with a medal lending rate of 5 yen per medal, and pachislot machines 1b (20 yen slots) with a medal lending rate of 20 yen per medal. The gaming machine 1 and the rental machine 2 are communicatively connected to the hall computer 4 via a LAN 3 and a relay device not shown, and various gaming signals are transmitted from the gaming machine 1 and the rental machine 2 to the hall computer 4. The gaming signals include an out signal that is output in response to the number of gaming media inserted into gaming machine 1, a safe signal that is output in response to the number of gaming media paid out from gaming machine 1, and a lending signal that is output in response to a lending operation on lending machine 2.

[0009] The hall computer 4 is an information processing device that operates according to a program, and is communicatively connected to the gaming machines 1 via a LAN 3, and is communicatively connected to a server 6 via a public line 5 such as the Internet. The hall computer 4 is a gaming device installed in a location where it can be operated by the store manager, and is capable of acquiring, calculating, and aggregating gaming information regarding all gaming machines 1 installed in the store. For example, the hall computer 4 inputs game signals output from the pachinko machine 1a and the rental machine 2, and calculates and compiles game information indicating the operation of the pachinko machine 1a, such as machine out, start, base, BA, SA, T1Y, and customer occupancy rate, and inputs game signals output from the pachislot machine 1b and the rental machine 2, and calculates and compiles game information indicating the operation of the pachislot machine 1b, such as out, coin price, and payout rate. The breakdown and calculation method of each piece of game information is publicly known, so a description thereof will be omitted. The hall computer 4 can transmit the calculated gaming information to the server 6 in a manner that allows it to be compiled by gaming machine, PS category, model, model type, setting value, lending rate, and date. That is, the server 6 transmits past gaming information (operation performance information) of each store to the server 6. The server 6 may perform part or all of the calculation and compilation of the game information.

[0010] The server 6 is an example of a gaming device of the present invention, and calculates operation prediction values ​​using AI (Artificial Intelligence), and creates a business plan (operation plan) by performing a simulation using the operation prediction values. The server 6 is an information processing device that operates according to a program, and is communicably connected to the hall computer 4 and terminal devices 7 via a public line 5 such as the Internet. As shown in FIG. 2, the server 6 is configured with an output unit 61, a communication unit 62, a storage unit 63, and a control unit 64. The output unit 61 outputs game information. For example, prediction information regarding the operation of the gaming machine 1 (operation prediction information) is output to the communication unit 62. The communication unit 62 is connected via the public line 5 to be able to communicate with external devices such as the hall computer 4 and the terminal device 7 of each store. As a result, the communication unit 62 receives past gaming information (operation performance information) of each store transmitted from the hall computer 4 of each store. The communication unit 62 transmits the operation prediction information output by the output unit 61 to the terminal device 7 and the hall computer 4. The terminal devices 7 and the hall computers 4 are equipped with software (browser) that displays the operation forecast information and operation plans output from the server 6 in a viewable manner. The communication unit 62 is also connected to an external device such as a server that provides a Web-API so as to be able to communicate with the external device. An example of a Web API is an SNS API that can provide SNS information such as text information posted on SNS (Social Networking Services) such as Twitter, Instagram, 5ch, TicTok, and Google Trends (including registered trademarks), the number of posts, and the number of searches. This allows the server 6 to acquire various types of SNS information. Other examples of Web APIs include a weather API that can provide weather forecast information, and a crowd count API that can provide the number of people in town or around a store.

[0011] The storage unit 63 is configured with a storage medium (for example, a ROM, a RAM, a hard disk, an SSD, etc.) capable of storing programs and data for executing various functions of the server 6. The storage unit 63 stores past gaming information (operation performance information) of each store received from each store (hall computer 4). FIG. 3 is a chart showing an example of past gaming information (operation performance information) of each store. As shown in the figure, the memory unit 63 stores, for the pachislot machine 1b, the out, coin unit price, payout rate, etc., and for the pachinko machine 1a, the machine out, start, base, BA, SA, T1Y, customer retention rate, etc., so that they can be managed by store, PS category, model, setting value, and loan date.

[0012] The storage unit 63 stores the feature information of the gaming machine 1 in addition to such past gaming information (operation performance information). The characteristic information includes information that is presumed to be correlated with the operation of the gaming machine 1. The characteristic information can be classified into model characteristic information indicating the characteristics and properties of the gaming machine 1 or the model, calendar information, and hall characteristic information indicating the characteristics and properties of the store.

[0013] FIG. 4 is a chart showing the breakdown of the machine characteristic information, calendar information, and hall characteristic information. As shown in the figure, the "machine characteristic information" includes the machine name, number of units, manufacturer name, PS classification, machine type, information on whether or not there is a stage setting, setting value, cabinet image information, introduction date, certification code, category name, SNS information, and chronological gameplay information. "Number of machines" is the number of gaming machines 1 of that model name installed in the establishment. "PS classification" is information indicating whether the machine is a pachislot machine 1b or a pachinko machine 1a. For pachislot machine 1b, the "machine type" includes normal type (A type), AT type, ART type, A+ART type, Okinawa slot type, etc., while for pachinko machine 1a, there are "sweet digital type," "max type," "middle type," "light middle type," "wing type," etc. The "stage setting information" is information indicating whether the gaming machine 1 has a function that allows the payout rate, including the probability of bonus occurrence, to be set in multiple stages (for example, six stages). For example, in the case of six levels, the "setting value" is numerical information from 1 to 6. It is not limited to six levels, and may be five levels or less, and does not have to be numerical information as long as the level is specified. The "cabinet image information" is, for example, image information of the cabinet of the gaming machine 1 (see FIG. 7). The housing image information mainly uses color still image information, but monochrome image information or moving image information may also be used. The cabinet image information may be information on an image captured in advance by a camera, or image information included in product information sent by a gaming machine manufacturer. "Category information" is information that indicates various categorization information (for example, the series name given to a series of models, the names of affiliated manga / anime or singers, original names when classified independently, etc.). "SNS information" is information about SNS for each model acquired from the SNS-API (see Figure 8). It may be information about one type of SNS or two or more types of SNS. "Time-series gaming information" is past gaming information used to identify operating trends, such as the previous day's actual operating results and each operating results value from the introduction date to the previous day.

[0014] "Calendar information" includes the date, day of the week, national holidays, weather, national events, and crowd numbers. "National events" is information showing events and happenings that many Japanese people are interested in, such as the Olympics and the World Cup. "Number of people" is information that indicates the occupancy rate of trains and the number of people in town or around stores. "Hall characteristic information" includes store-specific data and information indicating regional characteristics. The individual store data includes information indicating the store name. The individual store data also includes information indicating the operation of each store (for example, gaming information for each store, such as outs, sales, and gross profits for each lending rate). The information indicating the locality includes, for example, the address of the store.

[0015] In this way, the memory unit 63 not only stores programs, but also functions as a storage means for storing characteristic information (model characteristic information, calendar information, hall characteristic information) indicating the characteristics of the gaming machine 1 and past gaming information (operation performance information) indicating the operation performance of the gaming machine 1.

[0016] The control unit 64 is configured with a CPU and the like, and performs the following operations by reading and executing the programs stored in the storage unit 63. Specifically, the control unit 64 operates as a model generation means, and generates a model capable of outputting predictive information regarding the operation of the gaming machine by learning the relationship between the characteristic information (model characteristic information, calendar information, hall characteristic information) of the gaming machine 1 stored in the memory unit 63 (storage means) and past gaming information (operation performance information).

[0017] A "model" can be generated using AI, i.e., machine learning. Specifically, a large number of combinations (also called "data sets") of characteristic information of the gaming machine 1 and past gaming information (operational performance information) stored in the memory unit 63 are prepared, and these are input into a neural network (program) so that the relationship between the characteristic information of the gaming machine 1 and operational performance is learned autonomously, and as a result, a model is generated.

[0018] This article explains deep learning based on neural networks. Figure 5 is a schematic diagram of a neural network. As shown in the figure, a neural network can perform deep learning by having multiple intermediate layers between the input layer and the output layer. In each layer, functions that can weight input values ​​and output them are arranged as neurons. One neuron is arranged in the output layer. This is because one predicted value (continuous value) is output in this embodiment. The number of neurons in the input layer and the intermediate layer can be set arbitrarily. It is assumed that the weights of the neurons are set to predetermined initial values. The learning process in this embodiment will be described. During the learning stage, when "feature information" corresponding to a "question" is input from the input layer, it undergoes calculations in numerous neurons arranged in multiple intermediate layers, and an output value of the game information is output from the output layer. The weights in each neuron are autonomously changed (optimized) so that the error between this output value and the actual performance ("actual value") corresponding to the "answer" is minimized (see Figure 5). By performing this learning process (deep learning) on ​​a large number of data sets, a predictive model is formed that can accurately output unknown operational prediction values ​​(AI prediction values).

[0019] In this embodiment, a data set of feature information and past game information (achievement values) stored in the storage unit 63 is used as the learning target. For past game information (performance values), the performance values ​​of outs on the day are used. The feature information uses one or more pieces of information included in the machine type characteristic information, calendar information, and hall characteristic information (see FIG. 5). In other words, it is sufficient to use one or more of the following information shown in Figure 4: model name, number of machines, manufacturer name, PS classification, model type, information on whether or not a stage setting is present, setting value, cabinet image, introduction date, certification code, category name, SNS information, time-series gaming information (model characteristic information), date, day of the week, national holidays, weather, national events, number of people in attendance (calendar information), individual store data, and information indicating regional characteristics (hall characteristic information).

[0020] As an example, the case of learning a data set of the out performance value, cabinet image information, and SNS information for the previous day (i.e., the day before the current day) and the out performance value for the current day (i.e., the day after the previous day) for each model and each setting value will be explained in the order of (1) to (3) below. For ease of explanation, the models that have been introduced are models A to Z, each set to one of 1 to 6, and the actual out performance values ​​are stored in memory unit 63 from the introduction date (January 1, 2022) to May 5 of that year.

[0021] (1) For model A, as shown in FIG. 6(a), the actual out value, cabinet image information, and SNS information for setting 1 on the day of installation are input to output an output value, and a model is generated by learning so that the error between this output value and the actual out value for setting 1 on the day after installation (installation day + 1 day) is minimized. Next, the actual out value, cabinet image information, and SNS information for setting 1 on the day after installation (installation day + 1 day) are input into the trained model to output an output value, and the model is updated by learning so that the error between this output value and the actual out value for setting 1 on the day after installation + 2 days is minimized. This process is performed for each of settings 1 to 6 for up to 5 days. This completes the learning for model A, and completes the generation of an out prediction model that has learned the characteristics of model A.

[0022] (2) Add a learning process for model B to the out prediction model for model A. First, the actual out value, cabinet image information, and SNS information for setting 1 on the day of introduction are input into the trained out prediction model for model A to output an output value, and the model is updated by learning so that the error between this output value and the actual out value for setting 1 on the day after the introduction date (introduction date + 1 day) is minimized. Next, the actual out value, cabinet image information, and SNS information for setting 1 on the day after the introduction date (introduction date + 1 day) are input into the updated model to output an output value, and the model is updated by learning so that the error between this output value and the actual out value for setting 1 on the day after the introduction date + 2 days is minimized. This process is performed for each of settings 1 to 6 for up to 5 days. This completes the learning for models A and B, and completes the generation of out prediction models that have learned the characteristics of models A and B.

[0023] (3) The process in (2) above is executed for each of the models C to Z. This completes the learning for all models, and completes the generation of the out-prediction model M11 that has learned the characteristics of all models. That is, it is possible to generate an out-of-pocket forecasting model M11(5 / 6) that can output out-of-pocket forecast values ​​for the planned date: May 6th using out-of-pocket actual values ​​up to May 5th. That is, it is possible to generate an out prediction model M11 for each setting, which can output an out predicted value for the current day using the out actual value up to the day before the current day. The order of learning is not limited to the above (model A → model B → . . . → model Z), and any order may be used as long as learning is performed for all models.

[0024] The process for outputting the out-predicted value using the next day (May 6th) in the immediate future as the planned date will be described below. In this case, the out performance value, the case image information, and the SNS information on the day before the planned date (May 5th) are input to the generated out prediction model M11 (5 / 6). As a result, the out-of-stock value for the planned date (May 6th) is output. Here, when business hours end on May 6th and the actual out-of-stock value for that day is acquired (stored), at that point the out-of-stock prediction model M11(5 / 6) can be updated to the out-of-stock prediction model M11(5 / 7) that can output the out-of-stock value for May 7th. Specifically, the out prediction model M11(5 / 6) is updated to the out prediction model M11(5 / 7) by learning to minimize the error between the output value (out prediction value for May 6) when the actual out value for May 5, cabinet image information, and SNS information are input into the out prediction model M11(5 / 6) generated just before and the obtained actual out value for May 6. In other words, once the actual out-of-stock value for the business day has been acquired after the business has closed, it is possible to generate an out-of-stock prediction model M11 that can always output the out-of-stock value for the following planned date.

[0025] (Forecast values ​​for the following business days) It is also possible to output the out predicted value for the next day and beyond. As an example, a case will be described in which out-of-pocket forecast values ​​for each business day from May 6th to May 10th are output at the close of business on May 5th. In this case, the out performance value for May 5, the case image information, and the SNS information are input to the out prediction model M11(5 / 6). As a result, the out predicted value for May 6 is output. This out predicted value for May 6 is stored. Next, without waiting for May 6th to pass, the out predicted value for May 7th is output. Specifically, the out predicted value for May 6th, the case image information, and the SNS information are input to the out prediction model M11(5 / 6). As a result, the out predicted value for May 7th is output. This output value is stored. This process should be carried out until the out predicted value for May 10th is output. This allows the out forecast values ​​for each business day from May 6th to May 10th to be output at the close of business on May 5th. Therefore, it is possible to output each out-of-date forecast value over several days or the out-of-date forecast value several days later without waiting for the passage of business days. This allows each out predicted value over a long period in the future (for example, 180 days) to be output early.

[0026] By performing the same processes as those in (1) to (3) above using the coin unit price and the payout rate, it is possible to generate the coin unit price prediction model M12 and the payout rate prediction model M13. In other words, the coin unit price prediction model M12 can be generated by substituting the actual coin price value and the predicted coin price value in place of the actual out value and the predicted out value, and the payout rate prediction model M13 can be generated by substituting the actual payout rate value and the predicted payout rate value in place of the actual out value and the predicted out value. This allows the pachislot machine model M1 to be generated for each model and each setting value. In addition, for the pachinko machine 1a, by performing the same processes as (1) to (3), it is possible to generate pachinko machine models M2 for each model, such as a machine-out prediction model M21, a start prediction model M22, a base prediction model M23, a BA prediction model M24, an SA prediction model M25, a T1Y prediction model M26, and a customer retention rate prediction model M27. In this way, an operation prediction model that reflects operation trends over time can be generated.

[0027] The above-described process is a method for generating a model that learns the relationship between the performance value of the day before the current day and the performance value of the current day (the day after the previous day), but is not limited to this. For example, the model of the present invention may be generated by learning the relationship between the performance values ​​from three days before to the previous day and the performance values ​​of the current day. In this case, multiple actual values ​​such as the actual value from three days ago, the actual value from two days ago, and the actual value from one day ago can be input, and a representative value such as the sum or average of these actual values ​​can be input, and a model can be generated by learning to minimize the error between the output value and the actual value of the current day. In this way, for example, the model of the present invention can be generated by learning the relationship between the performance values ​​from the day of introduction to the previous day and the performance values ​​on the current day. The model of the present invention may be generated using a known recurrent neural network (for example, a recurrent neural network (RNN) or a long short-term memory (LSTN)) that can learn features related to time series.

[0028] (About the cabinet image information and SNS information) This embodiment is characterized in that the learning subject includes time-series gaming information, as well as one or both of cabinet image information of the gaming machine 1 and SNS information. The cabinet image information is input pixel by pixel as the cabinet image of the model (information indicating the board structure). For example, each vertical and horizontal pixel with a gradation (256 levels) of R (red), G (green), and B (blue) is input. For example, for a cabinet image consisting of 100 vertical pixels and 100 horizontal pixels, 100 (vertical) x 100 (horizontal) x 3 (RGB) neurons are placed in the input layer, and each pixel information is input to each neuron. As a result, information that can identify the external features of the gaming machine 1 (such as the features of the board surface) such as the shape, pattern, and color of the cabinet image and their positions is learned as feature information via the neural network. The reason why the external features of the gaming machine 1 are included in the learning object is that a correlation between these external features and the operation of the gaming machine 1 has been inferred or confirmed. In other words, the gaming machine 1 is often given a two-dimensional or three-dimensional design such as a shape, pattern, or color that is unique to each model on the cabinet, and the analysis results show that this can affect the level of operation independently of or in synergy with the playability of the model, for example, if the design is novel or attractive, it will make the player more interested in playing, while if the design is stale it will not make the player interested in playing.

[0029] FIG. 7 is a diagram showing an image of the cabinet of a gaming machine 1 known to be highly operated. FIG. 7(a) is an external view of model A, which has a bonus notification section with a jagged pattern a on the left side of the middle section of the cabinet. In addition to model A, it was inferred or confirmed that models with pattern characteristics similar to model A, such as models with jagged pattern a (including patterns similar to said pattern) on the casing, models with said jagged pattern a in the middle of the casing, models with said jagged pattern a on the side of the middle of the casing, and models with said jagged pattern a on the left side of the middle of the casing, tend to have higher operating rates. FIG. 7(b) is an external view of model B, which has a bonus notification section surrounded by a circle b in the center of the middle section of the cabinet. In addition to model B, it was inferred or confirmed that models with similar shape characteristics to model B, such as models that have a circular frame b (including a shape similar to the said shape) on the casing, models that have the said circular frame b in the middle of the casing, and models that have the said circular frame in the center of the middle of the casing, tend to have higher operating rates. FIG. 7(c) is an external view of model C, the entire housing of which is gold-colored. In addition to model C, it was inferred or confirmed that models with similar color characteristics to model C, such as models with gold or silver coloring on part or all of the casing, tend to have higher operating rates. FIG. 7(d) is an external view of model D, which has a distinctively shaped accessory on the top of the housing. FIG. 7(e) is an external view of model E, which has a special feature on the top of the housing that protrudes significantly from the housing body. In addition to models D and E, it was inferred or confirmed that models with similar shape characteristics to models D and E, such as models with distinctively shaped devices on the cabinet, models with devices that are so large that they extend beyond the main body of the machine, and models with such devices located on the top of the cabinet, tend to have higher operating rates.

[0030] In this way, it is inferred that there is a correlation between the external appearance characteristics of the casing and operation, and therefore, in the present invention, by learning the relationship between casing image information and actual operation values, it is possible to generate a model that outputs predicted operation values ​​for any model by inputting casing image information for that model. Furthermore, by including the housing image information in the learning object, it is possible to generate a model with higher prediction accuracy than when only other feature information is used as the learning object.

[0031] For SNS information, enter the SNS information for each model obtained via the SNS-API. On social media, people often post information about machines with high operation rates, such as "a godly machine," "tens of thousands of coins," "fun," and "amazing payouts" (high operation rate posting information). In other words, it can be inferred that models for which such posted information is posted have high utilization rates, and the more posts there are, the higher the utilization rate of the model. Furthermore, people who search for and view such posted information will become interested in the machine and be more likely to play it in the future, so it is expected that the more searches there are, the higher the utilization rate will be. On the other hand, on social media, people sometimes post information about machines with low utilization rates, such as "crappy machine," "removed," or "I won't play anymore" (low utilization information). In other words, it can be inferred that models for which such posted information is posted have low utilization, and the more posts there are, the lower the utilization of the model. Furthermore, since people who search for and view such posted information will not be interested in that machine and will be less likely to play it in the future, it can be assumed that the more searches there are, the lower the utilization rate will be. Therefore, in the present invention, as shown in Fig. 8, a score is assigned to each model depending on the number of characters in the high operation information and low operation information, the number of searches, and the number of posts, and the score is included in the learning target as SNS information. Note that instead of the score, posted character information can also be included in the learning target as SNS information. As such, it is presumed that a correlation exists between SNS information and operation information. Therefore, in the present invention, by learning the relationship between SNS information and actual operation values, it is possible to generate a model that outputs a predicted operation value for any model when the SNS information for that model is input. Furthermore, by including SNS information in the learning target, it is possible to generate a model with higher predictive accuracy than when only other feature information is used.

[0032] (Calendar information) The feature information may include other calendar information in addition to the housing image information and SNS information. The calendar information includes information indicating the date, day of the week, national holidays, weather, national events, and number of people. In particular, it is presumed that the "day of the week" is correlated with operation. For example, it is known that weekdays, Monday through Friday, are less busy than Saturdays and Sundays, and it is also estimated that Monday, the day after a busy Sunday, is less busy than other weekdays. Therefore, by generating a model that includes the "day of the week" as a learning target and inputting feature information including the day of the week of the predicted date into the model, it is possible to expect to output highly accurate predicted values. It is assumed that "weather" also has a correlation with operation. Therefore, by generating a model that includes "weather" as a learning target and inputting feature information including the "weather" expected on the predicted date into the model, it is possible to expect to output highly accurate predicted values. It is assumed that "national events" also have a correlation with operation. For example, during the Olympic Games, it is expected that utilization will be low because many people will be watching the games on television. The same is true during the FIFA World Cup. However, due to the time difference with the host country, it is expected that the situation will be different if the live broadcast is held at night or during the day. Therefore, if a "national event" such as the Tokyo Olympics or the Dubai World Cup is being held on the target date, information indicating that "national event" can be included in the learning target to generate a model. Also, if a "national event" is being held on the planned date, characteristic information including the "national event" can be input into the model, which is expected to output highly accurate prediction values. The information indicating "national events" can be automatically acquired from a publicly known Web-API that can provide calendar information. It is assumed that the number of people present also correlates with operation. It is assumed that when there are a lot of people in the city, there will be more activity than when there are fewer people. This tendency is particularly evident when there are a lot of people around the store. Therefore, by generating a model that includes information indicating the "number of people" as a learning target and inputting feature information including the "number of people" into the model, it is possible to expect the output of highly accurate prediction values.

[0033] (Hall characteristic information) In the hall characteristic information, "information indicating regional characteristics (for example, hall address)" is presumed to be correlated with operation. This is because it is empirically and statistically presumed that the number of players entering and leaving a store will increase or decrease depending on the location and its relationship with neighboring stores. Therefore, by generating a model that includes "information indicating regionality" as a learning target, and then inputting "information indicating regionality" into the model, we can expect to output highly accurate predicted values. "Individual store data" (for example, gaming information for each lending rate (outs, sales, gross profit, etc.)) is store-specific operating information that is presumed to have a correlation with operating hours. Therefore, by including individual store data in the learning process, an operation prediction model for each store can be generated. For example, by performing learning including the actual outflow values ​​for each PS category of each store, it is possible to generate a prediction model (store) that outputs the outflow predicted value for each PS category for the planning date and planning month of any store. In addition, by performing learning including gross profit for each PS category of each store, it is possible to generate a prediction model (store) that outputs gross profit prediction values ​​for each PS category for the planning date and planning month of any store.

[0034] (How to create Model M) The method for creating model M will be explained. FIG. 9 is a flowchart showing the procedure for creating the model M. As shown in the figure, first, in the server 6, the feature information of the gaming machine 1 is stored in the storage unit 63 (S11). The characteristic information includes, for example, time-series game information, cabinet image information (FIG. 7), SNS information (FIG. 8), and the like. Next, in the server 6, past gaming information (FIG. 3) showing the operating results of each store is stored in the storage unit 63 (S12). Next, learning is performed (S13). Specifically, the data set of the characteristic information stored in S11 and S12 and the past gaming information (operation performance values) is learned. Specifically, each piece of feature information is input from the input layer of the neural network. As a result, the weights of each neuron are autonomously changed (optimized) so that the error between the output value output from the output layer after calculations in a large number of neurons arranged in multiple intermediate layers and the actual operating value is minimized (see Figure 5). This learning process (deep learning) is performed on a large number of data sets. This generates a model (S14). That is, a model M capable of outputting predicted operation values ​​on future planned dates is generated. For example, as a model M1 for pachislot machines, an out prediction model M11, a coin price prediction model M12, and a payout rate prediction model M13 are generated for each model and each setting value, and as a model M2 for pachinko machines, a machine out prediction model M21, a start prediction model M22, a base prediction model M23, a BA prediction model M24, a SA prediction model M25, a T1Y prediction model M26, and a customer retention rate prediction model M27 are generated for each model.

[0035] (Creating operation plans based on simulations) The server 6 can create an operation schedule (operation plan) through a simulation using the AI ​​predicted values. That is, the server 6 includes a plan creation means that can create an operation plan for a specific period based on the prediction information (AI predicted value) output from the model generated by the model generation means. For example, a store manager can create an operation plan for their store. There are two operation plans: the S plan for the pachislot machine 1b and the P plan for the pachinko machine 1a. FIG. 10 is a flowchart showing the procedure for creating an operation plan. As shown in the figure, first, the planned date is input (S21). FIG. 11 is a diagram showing an example of the S plan screen, and FIG. 12 is a diagram showing an example of the P plan screen. The S plan screen and the P plan screen can be displayed on the display unit 71 of the terminal device 7 via the Web. The S plan screen and P plan screen have a "Planned Date" field.

[0036] Explain the S plan. The store manager of store A operates the terminal device 7 to input the planned date in the "planned date" field, and inputs the number of units to be allocated for each set value for each model. For example, enter "Year, Month, and Day" in the "Planned Date" field and enter "2 units" in the "Number of Units" field for "Setting 1" of "Model A." The entered information is sent to the server 6. The planned date: △ year △ month △ day is the day after the current date. FIG. 11(a) is the S plan screen before the planned date and the number of vehicles to be allocated are input, and FIG. 11(b) is the S plan screen after the planned date and the number of vehicles for setting 1: 2 have been input. As shown in the figure, when the planned date and the number of vehicles to be allocated are input, the server 6 receives the planned date and the number of vehicles to be allocated, and inputs the characteristic information into the generated model M (S22). As a result, the server 6 outputs the operation prediction value for the pachislot machine 1b from the model M (S23).

[0037] Specifically, for the out predicted value of model A, the out actual value of setting 1 of model A on the day before the planned date, cabinet image information, SNS information, etc. are automatically entered into out prediction model M11, and as a result, the out predicted value of setting 1 on the planned date is output from out prediction model M11, and the out predicted value for the two units (output out predicted value x 2) is calculated and automatically entered into the relevant column (out column for setting value 1 of model A). This makes it possible to show the predicted out value for model A installed in store A when setting 1 is set to two machines (see FIG. 11(b)). The same applies to the other setting values ​​(settings 2 to 6). For example, for setting 2, if "3 units" is entered in the number column for setting 2, the actual out value, cabinet image information, SNS information, etc. for setting 2 of model A on the day before the planned date are automatically entered into out prediction model M11, which causes the out prediction model M11 to output the out prediction value for setting 2 on the planned date, and the out prediction value for those three units (output out prediction value x 3) is calculated and automatically entered into the relevant column (out column for setting value 2 of model A) (see Figure 11(b)). Similarly, if you enter "4 units" in the number field for setting 3, the out prediction value for 4 units for setting 3 (output out prediction value x 4) will be automatically entered in the corresponding field; if you enter "3 units" in the number field for setting 4, the out prediction value for 3 units for setting 4 (output out prediction value x 3) will be automatically entered in the corresponding field; if you enter "4 units" in the number field for setting 5, the out prediction value for 4 units for setting 5 (output out prediction value x 4) will be automatically entered in the corresponding field; and if you enter "1 unit" in the number field for setting 6, the out prediction value for 1 unit for setting 6 (output out prediction value x 1) will be automatically entered in the corresponding field. This makes it possible to show the predicted out-of-range value for each setting value when each setting value is assigned to 17 units of model A installed in store A (see FIG. 11(c)).

[0038] The same applies to the predicted coin price and payout rate for machine A, and the same applies to the predicted operation values ​​for other machines. In other words, in this way, predicted operation values ​​for each setting value of all models of pachislot machines 1b on the planned date of store A are output (S23), and an operation plan (S plan) is created by simulation (calculation) based on the predicted operation values ​​(S24).

[0039] Explain the P plan. For example, when "Year, Month, Day" is entered in the "Planning Date" field on the P Plan screen, information indicating the entered planning date and number of units is sent to the server 6. FIG. 12(a) is the P plan screen before the planning date is input, and FIG. 12(b) is the P plan screen after the planning date has been input. As shown in the figure, the server 6 receives information indicating the planned date and the number of units, and the characteristic information is input to the generated model M (S22). As a result, the server 6 outputs the operation prediction value for the pachinko machine 1a from the model M (S23).

[0040] Specifically, for the predicted out-of-stock value of model M, the actual out-of-stock value of model M on the day before the planned date, cabinet image information, SNS information, etc. are automatically input into the out-of-stock prediction model M21, and as a result, the predicted out-of-stock value on the planned date is output from the out-of-stock prediction model M21, and the predicted out-of-stock value for that number of units is calculated and automatically input into the relevant column (the out-of-stock column for model M). Regarding the predicted start value of model M, the actual start value of model M on the day before the planned date, cabinet image information, SNS information, etc. are automatically entered into the start prediction model M22, and as a result, the predicted start value on the planned date is output from the start prediction model M22, and the predicted start value for that number of units is calculated and automatically entered into the relevant column (the start column for model M). The same applies to other operation forecast values ​​(base forecast value, BA forecast value, SA forecast value, T1Y forecast value, customer occupancy rate forecast value) and operation forecast values ​​of other models. In other words, in this way, the predicted operation values ​​of all models of pachinko machines 1a on the planned date of store A are output (S23), and an operation plan (P plan) is created by simulation (calculation) based on the predicted operation values ​​(S24).

[0041] In this way, the server 6 of the present invention can output predicted operation values ​​for a future planned date (for example, the next day) and create an operation plan through a simulation using the predicted operation values. The planned date may be any day after the next day. In this case, the server 6 can output the predicted operation value for the planned date and the predicted operation values ​​for multiple days up to the planned date using the method described above in "Regarding predicted values ​​for business days from the next day onwards."

[0042] The server 6 can create an operation plan for a specific period by performing a simulation using the predicted operation values. Figure 13 shows the S Plan screen when June is the planning month. For example, the operation plan for June can be created and displayed for each business day, including out (average), gross profit (total), achievement rate (today), and achievement rate (cumulative). "Out (average)" is information calculated as the average value of predicted out values ​​per day for all pachislot machines 1b in store A. "Gross profit (total)" is the total gross profit of all pachislot machines 1a in store A. Gross profit is calculated by "sales - exchange amount." A predicted value (gross profit forecast value) can be output for "gross profit / day" per pachislot machine 1a using Model M, and "gross profit (total)" can be calculated by adding up the predicted gross profit value for the number of machines. "Achievement rate (today)" is information that represents, for example, the ratio of the planned value for that day (for example, the forecast value for that day of outflow or gross profit) to the actual value, expressed as a percentage. "Achievement rate (cumulative)" is information that, for example, expresses as a percentage the ratio of the cumulative planned value up to the current day (the cumulative value of the out-of-pocket forecast value and the gross profit forecast value from the first day of the planning month to the current day) to the cumulative actual value (the cumulative value of the out-of-pocket actual value and the gross profit actual value from the first day of the planning month to the current day). According to such an operation plan, the store manager can refer to the operation plan for the planned month by day.

[0043] Figure 14 shows the P Plan screen when 2020 is the planning year. As an operation plan for 2020, for example, gross profit (total / month) by rental rate and operation (units / day) by rental rate can be created and displayed monthly. "Gross profit by lending rate (total / month)" is the predicted total gross profit for each month for 4 yen pachinko and 1 yen pachinko. "Operation by lending rate (units / day)" is the predicted total number of 4 yen pachinko machines and 1 yen pachinko machines in each month. According to such an operation plan, the store manager can refer to the operation plan for the planned year by month. It is also possible to specify a planning period of multiple days, multiple months, or multiple years.

[0044] As described above, the server 6 (gaming device) and gaming system of the present invention are gaming devices capable of outputting information relating to gaming machines 1 installed in an amusement facility, and are equipped with a storage means capable of storing feature information indicating the characteristics of the gaming machine 1 and past gaming information indicating the operating performance of the gaming machine 1, and a model generation means capable of generating a model that outputs predictive information relating to the operation of the gaming machine 1 by learning the relationship between the feature information and the past gaming information, and the feature information includes image information relating to the gaming machine's cabinet. The image information includes information that can identify one or more of the shape, pattern, and color, and the feature information includes SNS information related to the gaming machine. According to the present invention, simulations are performed not simply using actual values, but using AI predicted values ​​obtained through machine learning, which is expected to result in more accurate predicted values. In particular, it is possible to conduct simulations that take into account novel information such as images of the case and social media information, making it possible to derive new predicted values ​​(simulation results) that were not previously possible. The system also includes a plan creation means that can create an operation plan for a specific period based on the prediction information output from the model created by the model creation means. In particular, for new models, even before they are introduced, an operation prediction value can be obtained that reflects the relationship between the operation records and characteristics of other similar models, and even immediately after introduction, an operation prediction value can be obtained that takes into account the relationship between the operation records and characteristics of the new model over a short period of time (for example, three days) after introduction. This allows long-term operation plans to be drawn up early on, even for new models, so store managers can quickly decide whether to increase or decrease the number of units of the new model, or whether to maintain the status quo without increasing or decreasing the number of units. In contrast, conventionally, simulations are performed simply using actual results, which often results in a discrepancy between the simulated values ​​and the values ​​after the business hours have ended. According to the present invention, it is possible to solve some or all of these conventional problems.

[0045] Although the preferred embodiments of the present invention have been described above, it goes without saying that the present invention is not limited to the above-described embodiments, and various modifications can be made within the scope of the present invention. For example, sales and the like may be included in the gaming information. Furthermore, since the coin price and payout rate can be calculated based on the out, the predicted coin price and payout rate may be calculated based on the predicted out value without using the model M. Furthermore, the configuration and method for generating a model M capable of outputting a predicted operation value for any model using actual operation values ​​for all models in all stores has been explained, but it is also possible to generate a model M capable of outputting a predicted operation value for a target gaming machine 1 (a gaming machine of a specific store, model, manufacturer, model type, rental rate, etc.) using actual operation values ​​for the target gaming machine 1. The present invention can also be applied to so-called enclosed gaming machines that can play games using pseudo gaming media in data format, without using actual gaming media such as medals or gaming balls. [Explanation of symbols]

[0046] 1. Gaming machines 4 Hole Computer 6 Server 61 Output section 62 Communications Department 63 Storage section 7 Terminal Equipment 71 Display section M Model

Claims

1. A gaming device capable of outputting information about gaming machines installed in a gaming facility, a storage means capable of storing characteristic information indicating characteristics of the gaming machine and past gaming information indicating the operation record of the gaming machine; a model generation means for generating a model that outputs game information relating to the operation of the gaming machine as prediction information in response to input of the feature information by learning the relationship between the feature information and the past game information; The characteristic information includes the number of characters of SNS information related to the gaming machine. A gaming device characterized by:

2. The number of characters in the SNS information includes: The number of characters in positive and negative posts about the gaming machine is included.

2. The gaming device according to claim 1.

3. The characteristic information includes information indicating a national event.

3. The gaming device according to claim 1 or 2.

4. The characteristic information includes address information of the amusement facility.

3. The gaming device according to claim 1 or 2.

5. A gaming system including a gaming machine installed in a gaming facility and a gaming device capable of outputting information about the gaming machine, a storage means capable of storing characteristic information indicating characteristics of the gaming machine and past gaming information indicating the operation record of the gaming machine; a model generation means for generating a model that outputs game information relating to the operation of the gaming machine as prediction information in response to input of the feature information by learning the relationship between the feature information and the past game information; The characteristic information includes the number of characters of SNS information related to the gaming machine. A gaming system characterized by:

Images