ES2341445T3 - EVALUATION OF THE POSSIBLE FRAUDULENT BEHAVIOR IN LINE GAMES. - Google Patents

Abstract

The present invention relates to a system and a method for evaluation of possible fraud behavior in online games, such as internet poker games, between at least two parties with bets including valuables. A polling means fetches a plurality of input attributes for a specific state in a game between at least two players, wherein the state demands a decision from the players. The attributes and decisions taken are stored in a play array for each player. A predetermined agent array is provided with attributes closest to represent the specific state for each player and a conventional decision taken for the specific state for each player. A map array is provided with a set of possible agent arrays for specific state arrays as attributes, subdividing the set in at least two subsets. Each subset determines a class for the play arrays and at least one subset classifies a play array as conventional behavior and at least one subset classifies a play array as fraud behavior. A difference is determined between the play array and the agent array for each player through a comparator, and the difference is mapped on the map array positioned within one subset. Each player decision is classified in a subset during play, determining an individual player pattern over said subsets. The pattern comprises every decision taken by said individual player and determines said player as fraud if the corresponding pattern fits to a predetermined pattern of fraud behavior within said map array.

Description

Evaluation of possible behavior fraudulent in online games.

Technical field

The present invention relates to a system and to a procedure to evaluate fraudulent behavior in online gambling games that include values, between at least two parties, comparing player actions during a state of game with the corresponding actions of a game agent standardized for said state, said agent being trained in standardized game tactics for a specific type of game by analyzing such games stored with previously through at least one neural network.

Prior art

The tradition of the game has historical roots. In areas of Central Europe, North America and Asia the casino culture has now grown to create entire cities just for that purpose. People are attracted to those redoubts of the game so much before the prospect of earning a quick and "easy" money as for the experience of the unique culture and environment that they bring with him. However, it often takes time, mobility and money in order to get access to such places of play. For people who come from afar, a trip to Las Vegas, for For example, it could be a very burdensome affair. In fact, few Game enthusiasts have the possibility to practice their hobby as often as they would like.

These inconveniences associated with the game of casino have probably been a major factor in the growing popularity of the various online internet casinos that have sprouted at an increasing speed over the past decade. With the help of a computer connected to the Internet, enthusiasts can Play now directly in the privacy of their homes. Of this mode expenses are cut, mainly incorporating the game Game losses and possible casino charges, and are eliminated completely the loss of time and money due to travel and planning activities

Most internet casinos allow play the person against the casino, for example when playing with digital coin machines, digital roulettes, black jack, etc. Y thus they involve only a part that has some means limited to fool the digital machine. Other games that they involve playing and betting between two or more parties online, such like card games, for example poker, they are much more susceptible  of deception and fraudulent behavior. The possibility of cheating in poker it is different if the game takes place around a Physical table or virtually online. In poker on the Internet, for For example, it is virtually impossible to predict or influence the cast of cards, what is possible in real casinos by using of decks of marked cards, bribes and espionage. On the other hand, the Internet poker allows a completely new variety of modes to cheat, since neither the casino nor the opponents have a Direct view of player activities. It's almost impossible also, for example, determine a collusion in a game of letters between physically remote participants.

Online poker in Internet casinos is frequently used today for doubtful activities, such as money laundering, which of course is not interest of casino managers who do not want a penalty to criminal behavior, neither of the interest of the players honest that could see spoiled the pleasure of the game and look economically affected by this phenomenon. Also, like casinos  of the Internet handle people's money in local accounts they assume The role of a credit institution. This implies being responsible for payments if it is discovered that a deposited money has been stolen Therefore, such deviant use of online poker could eventually produce, if allowed to remain unpunished, the end of the Internet card game as such.

Dumping, for example, is a type of laundering of money in which poker is played online with stolen money and "bleached" earnings are collected. In principle, fraud implies that stolen money, for example credit cards, It is transferred to an account in an Internet casino. The thief or dumper then deliberately loses to another account. This Receiving account is in possession and is played by the thief himself or by an accomplice, the so-called receiver. A special kind of dumping that incorporates avoiding losses from table charges adopts a very characteristic pattern, when the first round of bets consists of one or more bets / raises followed by the withdrawal of all participants except the receiver.

Dumping in Internet poker is used also sometimes by a player as a driving procedure multiple own accounts and can also be used as a tool to settle debts between associates.

Collusion is another type of behavior fraudulent and implies that players use the largest undetectable cooperation opportunities that allow casinos on the Internet compared to real casinos. Two or more players at a poker table they are able to exchange information about his hands of cards without his honest opponents being at So much of it. Such information could be used to determine the value of a hand of cards and contribute to raising bets to high levels.

Both types of fraud mentioned above ruin the legitimate chances of winning players honest and fun of the game, and also, as mentioned previously, they imply a monetary risk for casinos.

The known detection systems of fraudulent behavior have been based on analyzing data from player required by the Internet casino, such as data personal and contact, of which the validity is checked by the casino once the player has registered. The correction of this data along with the credit history and transactions of a player are weighed next together with the average of gains / losses per hand to form a partial functional model for the fraudulent character.

A problem with this system is that it is built on observations of a number of fraudster attributes, which in their design are both partial and statistical. A system built on a momentary image of the dumping character implies the risk of future fraudsters acting differently to a degree that does not fit the system profile. Various systems to monitor games played on remote computers are known from US 2004/0053690, US 2002/169018, US-A-5970143.

US 2004/0053690 discloses a system comprising means to simulate the results of a competition and process data from a game environment; the US document 2002/0169018 discloses a remote gaming system, in which the Winning exit of the game is generated by the computer; The document US-A-5970143 discloses a device of computer and a procedure to encode a message that corresponds to an output of a computer game, and a device of computer and a procedure to decode the message to Detect a fraudulent exit.

A further solution is therefore required objective and independent of time to detect fraudsters, in which more emphasis is placed on analyzing decisions and tactics of the players during a game, while that the data of the surrounding players are less important in the analysis.

Summary of the invention

The present invention relates to a system and to a procedure to evaluate fraudulent behavior in gambling games that include values between at least two parties in line in an internet casino. A standardized player model or agent is established for a certain type of game analyzing the actions or movements made by players in a state of game given in a multitude of games of the same type played with before, and the agent is used in the system to compare player actions in given states during a game with the corresponding actions according to the player agent standardized, as empirically derived from these states. The resulting difference for each action is added and normalized over all actions in the game to form a basis for evaluating and classify player behavior graphically as normal or conventional or fraudulent, depending on the degree of divergence from preset threshold values for a standardized game according to the agent, once the game It has been completed.

An object of the invention is proportional such system and procedure, allowing automatic evaluation and a visualization of fraudulent behavior in online games which include bets, which is easily understandable by a observer.

To achieve the purposes and objectives the The present invention provides a system for the evaluation of possible fraudulent behavior in games with bets that include variables, according to claim 1.

In an embodiment of the present Invention, the game is an online poker card game.

Another embodiment of the invention defines that the input attributes comprise at least one of some data of cards, betting round, dealer position and first player movements.

Still another embodiment of the invention defines that the map matrix is a SOFM (Self-Organized Map of Features) pre-trained, prepared with regions for classify the behavior of said player, in which at least a region corresponds to said agent matrix.

In a further embodiment of the invention, the agent matrix is updated periodically, by  the means of interrogation, adjusting the games that have most recently completed within the framework of behavior normal for an iterative training of the agent matrix in standardized game tactics.

A further embodiment of the invention defines that the map matrix is displayed on a digital screen with said regions displayed discrete.

In an alternative embodiment of the invention, the agent matrix comprises at least one neural network trained on multiple games previously played by backpropagation of the error.

A further embodiment of the invention defines that a first, a second and a third neural network, mutually separated and structurally identical, of the type of multilayer perceptron, direct feed, fully connected are used for said agent matrix, each of which takes a respective signal response corresponding to a degree of probability of not going, passing / seeing and betting / raising, respectively, for a given state of the game, in which the strongest signal of the three determines the movement of the matrix for said state.

In still an additional embodiment of the invention, the difference for each game action is added and normalized on all the actions in the game to form a basis for assessing and classifying player behavior as normal or fraudulent, depending on the degree of divergence with respect to preset threshold values for a game standardized according to said agent matrix.

Still a further embodiment of the invention defines that at least one of the a number of additional game attributes that include at least one net economic balance, the number of hands played and the number of highly divergent hands.

In addition, the present invention establishes a procedure to evaluate the possible fraudulent behavior in network games with bets that include values, according to the procedure claim 11.

The process of the present invention is capable of performing procedural steps of the modes of previous network realization according to claims Dependent on attached procedures.

Brief description of the drawings

Hereinafter reference is made to the figs. attached for a better understanding of the present invention and of its examples and embodiments, in which:

Fig. 1 schematically shows the card layout and names in a Texas Hold'em game between two players

Fig. 2 schematically illustrates an agent or player model with separate status information, agree with a preferred embodiment of the present invention.

Fig. 3 schematically illustrates an agent or player model with amalgamated status information.

Fig. 4 schematically illustrates a Map Self-Organized Features (SOFM) that has regions predefined corresponding to normal and abnormal behavior for player hands in a poker game.

Fig. 5, according to an embodiment of the invention illustrates a deployment of a system to evaluate fraudulent behavior in online games with bets between at minus two parts

Fig. 6 shows a flow chart of system events to evaluate fraudulent behavior in online games with bets between at least two parties, all right with an embodiment of the invention.

Terms list

A neural network is a modeling technique based on the observed behavior of biological neurons and It is used to mimic the operation of a system. At context of this application, it consists of a set of neurons whose weights are initialized with random values, and which, in based on operational feedback, adjust so iterative towards the values required to generate the results required Multilayer Perceptrons and Self-Organized Maps of Features are two different types of neural networks used in the present invention.

A SOFM (Self-Organized Map of Features) It is a type of neural network capable of grouping and visualizing large amounts of data covering three or more dimensions simultaneously on a two-dimensional surface. A SOFM is routinely used to divide a data set into classes separate and describe how those classes relate to each other. In the solution according to the present invention, a SOFM is used instead to force new data into an existing class and graphically describe how well the new data vector fits in the class.

A perceptron model is built on a number of input signals which, together with a weight for Each signal is added together and transformed into a signal of output through a transition function, for example through of a formula such as:

one

in the which

2

Boards

Table 1, according to the invention, illustrates schematically a representation of a state of letters for a 52-card deck that has 104 values (1 or 0) in a game of Texas Hold'em

Table 2, according to a mode of embodiment of the invention, shows a state of letters vector for two closed cards in a Texas Hold'em poker game.

Table 3 shows a chart status vector for 5 opening cards in a Texas Hold'em game, all right with an embodiment of the invention.

Table 4 schematically illustrates a representation of a player state for a Texas game Hold'em, according to the invention.

Table 5 shows a first betting round in a poker game between two people, also called Mano a hand.

       \ vskip1.000000 \ baselineskip
    

Detailed description of preferred embodiments

The present invention raises a system and a procedure to evaluate possible fraudulent behavior in online games, such as poker games and card games similar on the Internet, enter at least two parties with bets that include values

To solve the problem of providing a functional system to evaluate fraudulent behavior in games online, means are needed to estimate the probability or normal actions or movements performed by a player during a game For a machine, such as a computer or a computer system be able to do this, a normal or probable poker game model, preferably in a piece of software component, hardware, or a combination of same. For a given state in a poker hand, the model should Estimate the probability of each possible movement. Comparing the estimation of the model with the real movements made by a player is possible to quantify the degree of normality of the player tactics. The divergence factor obtained by such Played hand analysis serves as a strong indicator of the honesty of the player in a subsequent analysis.

A neural network model based on a single layer of perceptrons is only able to approximate simple functions and classify linearly separable data sets. A superior complexity approach requires that the signals of output of a row of perceptrons are sent as input signals to additional rows of perceptrons. Such a neural network model, called a multilayer Perceptrons model, it can be extended to understand an arbitrary number of layers with the ability to approximate arbitrary functions. Also, the classification of problems that are not linearly separable becomes possible with this type of model.

In a preferred embodiment of the present invention, a player agent or agent matrix for a probable poker game, and specifically for a certain type poker game called Texas Hold'em, is built based on data stored in a database of a crowd, probably of the order of millions, from this type of game previously played This allows the construction of an agent of player or agent matrix describing a player standardized. The corresponding poker states and movements of stored games are extracted from or interrogated to the base of data, for an empirical training of the neural network of a set of multilayer perceptrons. Each state trains against a known target value for each individual set of multilayer perceptrons, according to the well known technology of supervised learning.

The player agent or agent matrix can be directed alternately to describe the optimal player, what which is theoretically achieved through network training neural based on prizes instead of games played above, for example using the well known TD algorithm (λ) of learning reinforcement. The objective of the agent of player trained through learning reinforcement can be directed towards maximizing the benefits in a game, since the most players are usually interested in winning large quantities of money. Alternatively, the training could be aimed at maximizing the frequency of results winners

The training is preferably done using a variant of an algorithm, well known for this purpose, called backpropagation of the error, although they can be also use other similar algorithms to train multilayer perceptrons. The backpropagation of the error uses the equally well known delta rule connected in series of the Last layer to the first. The delta rule allows the training of a perceptron to produce a certain output signal in response to a set of input signals, and therefore the perceptron It works like a classifier.

       \ newpage
    

A nonlinear transition function, such as logistics sigmoid function,

3

is used, in the mode of preferred embodiment of the present invention, to improve the approximation capacity of neural networks. The function sigmoid logistics, which is continuous, has an outbound domain that It is limited between 0 and 1 and is well adapted for use in both hidden layers, that is, intermediate layers, as in layers of signal of exit.

As the type of poker game called Texas Hold'em is currently one of the most popular poker variants, also in Internet casinos. This type of card game in specific line, which is played between at least two parts with bets that include values, is referred to continuation in the evaluation of the possible behavior fraudulent. However, other types of online card games, such as for example the "Seven open", are equally suitable for the evaluation of a possible behavior fraudulent with the solution according to the invention.

The Texas Hold'em poker game is played with an ordinary deck of 52 cards. A dealer is appointed player, whose job is to distribute closed cards to others players, and place open cards on the table. The letters are hand out clockwise, beginning with the player located directly to the left of the dealer. The distribution task also moves to the seated player directly to the left of the current dealer when it has been played the current hand. The game begins with the cast by the two card dealer to each participating player, these two letters are called "hand cards" and will remain in Secret for the opponents. From these letters, each player, again starting with the player to the left of the dealer, you can bet according to betting rules defined. Once the first betting round is completed, it three open cards are dealt to the table. These three letters are they are called "the first three community letters" and can be use in combination with the hand cards of each of the players to form a poker hand. Then follow one second round of betting, in which players have the opportunity to use the information that has been put to provision through "the first three community letters". Then a fourth open letter is dealt, "the fourth community letter ", and is put on the table still followed by A betting round Finally, the fifth and last letter open, "the fifth community card" is dealt to the table and The bet can be completed in a fourth betting round. The Distribution and name of the cards is shown schematically in fig. one.

When all the cards have been dealt, all the remaining players, in the order of clockwise, they can choose either to show their poker combination or admit the defeat. The best combination wins the game. The combinations, starting with the worst, they are the Highest Card, Couples, Doubles Couples, Threesomes, Stairs, Color, Full (more even trio), Four Equals (Poker), Color Ladder and Royal Flush.

Every move or action performed in a game Poker is based on the information or data available to the players at the time of movement. All this information or data together they constitute the state of the game before each movement. He game state, for example a Texas Hold'em game, is constituted by the player's open and closed cards, the betting round, the position with respect to the dealer, the previous movements of the player and the previous movements of opponents. The agent or player model will generate a estimation and a movement based on the unique game state for every moment given in the card game. A state vector, which It is unique to the given game state, it is established by both and describes all the relevant details of the game process Which leads to that particular moment.

In the preferred embodiment, a status representation for a Texas Hold'em player agent It is divided into two logical sections. The first section describes the status of letters and consists of 2 * 52 indicators (1 or 0), which they represent the player’s closed cards, called hand, and open table cards, called letters community As the number of cards available per player is constant regardless of the number of players, the vector of card status always consists of 104 elements and can have, of  according to an embodiment of the invention, an aspect such as the one shown in table 1 below.

The two closed cards: 2 of hearts and 5 of spades are described by a state vector according to the table 2 below.

The status vector for open letters is designates in the same way, for example as shown in table 3 Next for open letters: 4 of diamonds, 8 of clubs, ace of hearts, 3 of spades and ace of clubs.

The second state section, that is, the state of player, describes data on player position, round of bets and previous bets. The number of positions of player and the size of the bets depend on the number of players. Therefore, the size of the second status vector It varies with the number of players. The representation of the state of player, in which P is the number of players, can have then, in an embodiment of the invention, the appearance shown in table 4 below.

An example of how this state looks for the first betting round in a poker game between two people, also called hand in hand, it is shown in table 5 below. The player is the dealer and responds to the opponent by matching Your letters

The player agent or agent matrix comprising neural networks, trained as described above, responds to game states with signals corresponding to the tendency of the standardized player to choose not to go, pass / see or bet / raise. For game states, which were included in the training sets of the neural networks, the output signal is an approximation of the average of the corresponding target values in the training sets. The generalization capabilities of the neural networks thus achieved are used for other states to provide target values that correspond to similar states in the inter-group sets.
swimming

Fig. 2 schematically illustrates an agent of player or agent matrix according to the embodiment preferred of the invention, comprising three neural networks, this it is, three structurally identical three layer perceptrons, one for each movement of the game not to go, pass / see or bet / raise respectively, and is thus trained to respond with one of the respective movements corresponding to the movement of a Standardized player for a given state in a Texas game Hold'em, for example. The strongest output signal determines what movement will be executed by the agent or player model for the given game state. In order to achieve this, the three multilayer perceptrons hierarchically order the different movements based on a complete state vector or matrix of status comprising chart status data and status data of amalgamated player.

Fig. 3 schematically illustrates an agent of player or agent matrix, according to another embodiment of the invention, in which the underlying state of a movement has Been separated into two parts. The status of letters describes the Player's own cards and the cards that are available on table. The player status includes the game size or number of participants in the game, table position and movements and bets placed by the player and the opponents. Through this distribution the state estimate can be distributed in various multilayer perceptrons. In this mode of realization of the player agent status representation, a multilayer perceptron hierarchically orders the state of letters and the order is sent forward to three perceptrons additional multilayer. These three multilayer perceptrons order hierarchically not go, pass / see or bet / raise respectively to starting from the ordering of cards and player status. Like the Desired ordering values for individual states are unknown during training, this representation of status is only useful when combined with reinforced learning.

A player agent or agent based matrix in separate neural network has to be trained for each type of game, size, this is number of players participating in a game, and betting structure, for example bounce with limit, without limit, etc., requested, since both the appearance of the strategies such as the dimension and composition of the vectors of status or state arrays depend on these variables.

Instead of multilayer perceptrons, they can be alternatively use other mechanisms to approximate value, such as tile coding, in player agents or agent matrices. A heuristic model, based on rules specified manually could theoretically also serve as an alternative to models based on neural networks, given that they could formalize appropriate rules that describe the player standardized.

When poker hands are analyzed for potential dumpers / receivers, movements or actions of player are compared with the movements suggested by the agent as a player, for example, the three neural networks trained for specific type of poker game. In cases where player movements differ from agent movements, it calculate a scalar deviation as the difference between the signals of exit that correspond to the agent and the movements of the player. Deviations, accumulated on all movements of a hand and normalized on the number of movements, represent then a quantification of the probability of the strategy of player, in which a small deviation means that the hand is normal and vice versa.

A pair of divergent hands do not imply necessarily dumping or receipt. For better accuracy in the analysis, an embodiment of the invention comprises that the average of the difference of several hands combined with others factors such as a distribution of differences, for example, a few highly divergent or divergent hands constantly, an analysis of cash flow and how many There are opponents, their distribution and divergence.

In the preferred embodiment, the results of such analyzes or comparisons are visualized graphically using a pre-trained map matrix such as, by example, a SOFM. The map matrix or SOFM for this purpose has a graphic surface on which each point on the surface corresponds to a vector of attributes. The vector size of attributes corresponds to the dimension of the data set that is going to project on it. A pattern for the data set is matches the attribute vectors of the array of map or SOFM one by one. The attribute vector, which according to one dimension of the set represents the correspondence more next according to the pattern, represents the position of that pattern in the map matrix or SOFM. So, the map matrix or SOFM get the function of a ranked player ranking, that is, their hands, in different groupings or regions that correspond, for example, to normal or conventional behavior and abnormal for player hands in a poker game.

Traditionally, the training of a SOFM is carried out computing, for each pattern in a set of training, the corresponding position in the SOFM and doing iteratively that the attribute vectors in the area around of the position are more similar to the pattern. In order to apply The training algorithm to specific problems, should be specify similarity measurements and regional neighborhoods. For train the map matrix or SOFM used in the mode of preferred embodiment, a Euclidean distance is used as a measurement of similarity between attribute vector patterns and se use Gaussian functions to update the vectors of attributes in each region. The training sets of the map matrix or SOFM according to the invention are consisting of manually selected sets from player, in which both players are represented normal behavior as "criminal" of different types. The Map matrix surface or SOFM is thus prepared with regions or vector groupings each of which corresponds to players with different attributes, as exemplified in fig. 4, in which a player with, for example, normal and large game losses, is located in a region or grouping, while a player with a normal game and average winnings is located in another region or grouping.

Fig. 5 illustrates, in a block diagram, a embodiment of a system 100 deployed in accordance with invention, which uses an agent matrix such as, for example, software with an agent or player model pre-trained with a network neural according to the preferred embodiment described above, and a map matrix, for example software with a SOFM model pre-trained with neural network according to the mode of preferred embodiment described above, to detect and visualize fraudulent behavior in online games with bets, which include values.

Online players connect using a computer to an Internet gaming site, to play with each other a game of, for example, Texas Hold'em betting money or others values. Each player joins the game preferably at through your individual player account, which contains certain personal data requested from the player in a procedure of Registration before you are allowed to play on the site. The data Individuals are preferably checked by administrators of game to check its authenticity and against possible previous history of misconduct, before allowing / preventing The player participates in an online game with bets. The data of the individual player as well as the data of the poker hands plays through the registered account are stored in a base of 130 poker data. The poker hand data for each Participating player comprises at least card status data, player status data and data on decisions made for each state of the hand played arranged in a matrix of game, which is individual for each player. Therefore, each hand Played poker can be recreated and connected to player accounts individual to analyze a possible behavior fraudulent.

Such analysis is performed in the mode of preferred embodiment in one stage. In one embodiment Alternative is done in two stages.

The first stage of analysis includes that interrogation or retrieval means on server 140 of analysis retrieve data or game matrix from a hand played previously from, for example, Texas Hold'em from database 130 to recreate the game through the agent or agent matrix of Player using system software. The player agent or agent matrix is preferably pre-trained according to the steps described above for said type and size of game. The movements or actions of the agent or agent matrix of player for the game so recreated are then compared to each game state, by means to compare, with those of the game matrix, which includes the movements or actions real held by players in each game state, and a resulting scalar difference is computed, by means computing, for each action or movement. The average of deviation and distribution of all movements in the hand are computes for each player and fits by mapping according to a pattern in regions prepared on the map matrix, for example such as a SOFM, preferably according to the steps described previously, in which at least one region defines a fraudulent behavior and at least one region a behavior normal. The analysis is then stored in the database 150 analysis

The map matrix, for example the SOFM of according to the embodiment shown in fig. 4, how do I know discussed above, thus receiving the function as a classifier of data that is able to classify players in the region of surface of the map matrix or SOFM that has the vector pattern of attribute data that most closely resembles the pattern of the set of data that originates from the comparison / analysis above described of a really played hand with that of the womb player agent or agent.

The map matrix or SOFM, according to a mode of embodiment of the invention, distinguishes graphically fraudsters of honest players, for example digitally on the screen of a digital display device, and it is also able to visualize possible relationships between players, such as the relationship between a dumper and a receiver. The game administrators are therefore able to detect visually accounts, this is players, who show a obviously fraudulent behavior in an online game of, by Texas Hold'em example, through their positions computed in the map matrix, and are therefore able to take measurements appropriate against these to avoid such behavior of dumping / reception and collusion.

Stage two of the analysis includes combining also the results of stage one with other data such such as cash flow data, opponent data and any analysis result of a previous entry of each player recovered from database 150 by means of interrogation in the analysis server 140 to compute so more precise the position of each player in the map matrix, of according to the steps described above.

Human administrators can inspect at any time the results to continuously analyze individual players Connecting a client instance 120, such as a web browser on a computer, with server 110 of applications, a network connection between the application server and its customers allow interactive inspection of players analyzed by a graphic representation of the SOFM with the analyzed players represented on the surface in their computed positions, as illustrated in fig. Four.

Fig. 6, according to another mode of embodiment of the invention, schematically shown by a flowchart system events for continuous evaluation of fraudulent behavior in a Texas Hold'em online game with bets between at least two parties, according to a mode of embodiment of the invention.

The poker database is interrogated, to through the system software, about a 210 hand just move. If a new hand is not found, system 290 stays "asleep" or inactive for a set period of time and then interrogate the database once more. In another case, the new hand found is recovered from database 220 by of means of interrogation. Then the game is rebuilt. with agent matrices or players 230 agents corresponding and the game state advances 240. If there are any player actions, a state vector or matrix of status for the current player 250. The player agent is exposed then to the state vector and 260 the probability is computed of actions not to go, pass / see and bet / raise. The difference scale between the action preferred by the agent or model and the actual movement carried out is computed 270 using a comparator and the game model is updated with movement 280 led to out really. The game state advances 240 again. If there are no left player actions, the game ends and the average of deviations and distribution are computed for each player 300 through the means of computing the system. Cash flow and the relationships between the participating players are computed 310 by means of said means and the accumulated values are combined with the analysis of previous entries of each 320 player and positions in the map matrix, for example SOFM, they are computed by means of said computing means for each player 330. The analysis is finally stored in an analysis database.

According to the previous teachings, the The present invention comprises that interrogation means retrieve a plurality of input attributes for a state specific in a game between at least two players, in which the state demands a decision from the players. The attributes and decisions made are stored in an array of Game for each player. An agent array is available default with the closest attributes to represent the specific status of each player and a conventional decision taken for the specific state by each player. An array of map is predisposed with a set of possible agent arrays for specific state matrices as attributes, subdividing the set in at least two subsets. Each subset determines a class for game matrices and at least a subset classifies a game matrix as conventional behavior and at minus a subset classifies a game matrix as fraudulent behavior A comparator determines a difference between the game matrix and the agent matrix for each player and the difference is mapped on the map matrix positioned in a subset. Each player decision is classified in a subset during the game, determining an individual player pattern about such subsets. The employer understands each decision made by said individual player and determines said player as fraudulent if the corresponding pattern fits a pattern of Default fraudulent behavior in the map matrix.

The input attributes can comprise the minus one of some card data, betting round, position of the dealer and previous movements of the player.

The map matrix, for example such as a SOFM pre-trained, can be prepared with regions to classify the behavior of said player. At least one region can then correspond to the agent matrix.

The agent matrix may comprise at least a neural network trained on multiple games played with Previously by retropropagation of errors.

Through the means of interrogation, the matrix Agent can be trained alternately in game tactics standardized using training sets in which most recently analyzed games fit within the framework of normal behavior in the map matrix or SOFM, thus achieving a periodic update and a fine adjustment of the tactics of standardized set of agent matrix.

A digital screen can be used preferably to show the map matrix with regions and Discreet displayed players. Alternatively, the regions of the map matrix is displayed on another medium, such as paper or Similar printed medium.

First, second and third neural networks, mutually separate and structurally identical of the type of multilayer perceptron fully connected, direct feed, can be used for said agent matrix, each of the which takes out a respective signal response that corresponds to a degree of probability of not going, passing / seeing and betting / raising respectively for a given state in the game, in which the signal strongest of the three determines the motion of the matrix for the state.

The difference for each game action can be added and standardized on all actions in the game to form a basis for assessing and classifying the behavior of players as normal or fraudulent, depending on the degree of divergence from preset threshold values for a standardized game according to the agent matrix.

At least a number of game attributes additional ones that include at least the net economic balance, number of hands played and number of highly divergent hands can be considered in the evaluation or analysis.

The means mentioned herein description can be software means, hardware or a combination of both.

The present invention has been described with non-limiting examples and embodiments. What it describes all possible embodiments for a person skilled in The technique is the set of claims attached.

Boards

       \ vskip1.000000 \ baselineskip
    

       \ vskip1.000000 \ baselineskip
    

TABLE 1

       \ vskip1.000000 \ baselineskip
    

4

       \ vskip1.000000 \ baselineskip
    

       \ vskip1.000000 \ baselineskip
    

       \ vskip1.000000 \ baselineskip
    

       \ vskip1.000000 \ baselineskip
    

TABLE 2

       \ vskip1.000000 \ baselineskip
    

5

TABLE 3

       \ vskip1.000000 \ baselineskip
    

6

       \ vskip1.000000 \ baselineskip
    

       \ vskip1.000000 \ baselineskip
    

       \ vskip1.000000 \ baselineskip
    

       \ vskip1.000000 \ baselineskip
    

TABLE 4

       \ vskip1.000000 \ baselineskip
    

7

TABLE 5

       \ vskip1.000000 \ baselineskip
    

8

Claims (20)

1. A system for the evaluation of a possible fraudulent behavior in online games, with bets that include values, said system comprising: a means of interrogation, to retrieve a plurality of input attributes for a specific state in a game between at least two players, claiming that state a decision by said players, and storing said attributes and decisions made in a game matrix for each player; a default agent array means that have attributes closer to representing that state specific to each player and a conventional decision taken to said specific status for each player; a map array means, having said map matrix a set of possible agent arrays for specific state matrices as attributes, subdividing said set in at least two subsets; determining each subset a class for said game matrices, at least a subset that classifies a game matrix as conventional behavior and at least one subset that classifies a game matrix as behavior fraudulent; a means of comparison, to determine a difference between said game matrix and said agent matrix for each player, the difference being mapped on the matrix of map and being positioned within a subset; Y means to classify the decision of each player in a subset during the game, determining a pattern individual player on said subsets, said said employer each decision made by said individual player, and determining that said player is fraudulent if the employer corresponding fits with a default pattern of fraudulent behavior within said map matrix.

         \ vskip1.000000 \ baselineskip
      

2. A system according to claim 1, in which said game is a game of poker cards in line. 3. A system according to one of the claims 1-2, wherein said attributes of entry comprise at least one of a letter data, round of bets, betting speed and position on delivery man. 4. A system according to one of the claims 1-3, wherein said map matrix is a pre-trained SOFM (Self-Organized Characteristics Map) prepared with regions to classify separate behavior of said player, in which at least one region corresponds to said agent matrix. 5. A system according to claim 4, wherein said map matrix is displayed on a digital screen  with said regions displayed discrete. 6. A system according to one of the claims 1-5, wherein the agent matrix, through periodic interrogation through the means of interrogation, is trained with the games over recently that fit within the framework of normal behavior for a standardized game setting of the agent matrix. 7. A system according to one of the claims 1-6, wherein the agent matrix comprises at least one neural network trained on multiple games previously played by retropropagation of errors. 8. A system according to one of the claims 1-7, wherein at least one first, a second and a third neural network separated from each other and structurally identical to the multilayer perceptron type Fully connected, direct feed, is used to said agent matrix, each of them obtaining a response from respective signal that corresponds to a degree of probability for the movements of not going, passing / seeing and betting / raising respectively for a given state of the game, in which the strongest signal of the three determines the movement of the agent matrix for said state. 9. A system according to one of the claims 1-8, wherein a difference for each game action is added and normalized over all actions in the game to form a basis for evaluating and classifying player behavior as normal or fraudulent, depending on the degree of divergence from threshold values preset for a standardized game according to said agent matrix. 10. A system according to one of the claims 1-9, wherein at least one of a number of additional game attributes that includes at least one net economic balance, a number of hands played and a number of highly divergent hands are taken into consideration in the evaluation. 11. A procedure to evaluate a possible fraudulent behavior in online games with bets that include values, which includes: retrieve a plurality of input attributes for a specific state in a game between at least two players, demanding said state a decision on the part of said players, and storing those attributes and decisions made in a matrix of play for each player; provide an agent array means defaults that have closest attributes to represent said specific status for each player and a decision conventional taken for that specific state for each player; provide a means of map matrix, said map matrix having a set of possible matrices of agent for specific state arrays as attributes, subdividing said set into at least two subsets; determine each subset a class to said game matrices, at least a subset that classifies a game matrix as conventional behavior and at least one subset that classifies a game matrix as behavior fraudulent; determine a difference between said matrix of game and said agent matrix for each player, mapping the difference over the map matrix and being positioned in a subset; Y classify the decision of each player in a subset during the game, determining a player pattern individual on said subsets, said pattern comprising each decision made by said individual player, and determining that said player is fraudulent if the corresponding pattern adjusts in a predetermined pattern of fraudulent behavior within said map matrix.

         \ vskip1.000000 \ baselineskip
      

12. A procedure in accordance with the claim 11, wherein said game is a card game of online poker. 13. A procedure according to one of the claims 11-12, wherein said attributes input include at least one of a letter data, round of bets, betting speed and dealer position. 14. A procedure in accordance with one of the claims 11-13, wherein said matrix of map is a SOFM (Self-Organized Map of Features) pre-trained prepared with regions to classify the separate behavior of said player, in which at least one region corresponds to said agent matrix. 15. A procedure in accordance with the claim 14, wherein said map matrix is shown in a digital display with said regions displayed discrete. 16. A procedure in accordance with one of the claims 11-15, wherein the matrix of agent, through periodic interrogation by means of question marks, is trained with the games finished more recently that fit within the framework of a behavior normal for a standardized matrix game setting of agent. 17. A procedure according to one of the claims 11-16, wherein the matrix of agent comprises at least one neural network trained on multiple previously played games through the backpropagation of mistakes. 18. A procedure in accordance with one of the claims 11-17, wherein at least one first, a second and a third neural network separated from each other and structurally identical to the multilayer perceptron type Fully connected, direct feed, is used to said agent matrix, each of them obtaining a response from respective signal that corresponds to a degree of probability for the movements of not going, passing / seeing and betting / raising respectively for a given state of the game, in which the strongest signal of the three determines the movement of the agent matrix for said state. 19. A procedure according to one of the claims 11-18, wherein a difference for each game action is added and normalized over all actions in the game to form a basis for evaluating and classifying player behavior as normal or fraudulent, depending on the degree of divergence from threshold values preset for a standardized game according to said agent matrix. 20. A procedure in accordance with one of the claims 11-19, wherein at least one of a number of additional game attributes that includes at least one net economic balance, a number of hands played and a number of highly divergent hands are taken into consideration in the evaluation.