Resumen
A method is presented for generating training data for a bot detector module through human interaction with a mobile device. The method comprises receiving at least one signal generated by at least one sensor integrated into the mobile device, with the signal being at least one signal generated during the interaction of a human with the mobile device. The method further comprises calculating scalar and/or time variables of at least one sensor-generated signal that characterize human behavior, thus providing real training data and generating training data comprising at least the real training data. A bots detector module and a bots detection system are also presented.
Reivindicaciones
CLAIMS 1. Method for generating training data for a bot detector module through human interaction with a mobile device, the method comprising: receiving at least one signal generated by at least one sensor integrated in the mobile device, the signal being at least one generated during the interaction of a human with the mobile device; calculating scalar and/or time variables of at least one sensor-generated signal that characterize human behavior, thus providing real training data; and generate training data that comprises at least the real training data. 2. Method according to claim 1 further comprising synthesizing at least one signal that models human behavior to generate synthetic training data. 3. Method according to claim 2, characterized in that synthesizing the at least one signal that models a human behavior comprises a synthesis method based on the observation, analysis and characterization of the at least one signal generated by the sensor, the synthesis method comprising: receiving a plurality of signals generated by at least one integrated sensor; calculating a set of common features, the common features including at least one of a group comprising direction, duration, displacement, and angle, of the plurality of signals to characterize a human behavior; performing a statistical analysis of the common features to obtain a variability; generating synthetic training data comprising an initial data and successive data that differ from the initial data according to the variability, wherein the synthetic data is characterized by a synthesized direction, a synthesized duration, a synthesized displacement, and a synthesized angle; and generating training data comprising at least the synthetic training data. 4. Method according to claim 2, characterized in that synthesizing the at least one signal that models a human behavior comprises: providing a generative neural network configured to generate at least one synthetic signal, the generative neural network comprising at least two inputs; providing a discrimination algorithm configured to discriminate between real and synthetic data, and to generate as output a loss function that defines the error committed in discriminating, where the algorithm comprises at least three inputs; iteratively training the generative neural network and the discrimination algorithm in an adversarial mode by introducing as inputs to the generative neural network: the loss function generated by the discrimination algorithm; and at least one random noise signal to generate at least one synthetic signal; and by introducing as inputs to the discrimination algorithm: the at least one synthetic signal generated by the generative neural network, the at least one signal generated by the at least one integrated sensor and/or the real training data, and the loss function generated by the discrimination algorithm; until the absolute value of the difference in loss functions between successive iterations is between 0.0001% and 20%, thus obtaining a trained generative neural network; generating synthetic training data comprising at least one synthetic signal generated by the trained generative neural network, by introducing at least one random noise signal as input; and generating training data comprising at least the synthetic training data. 5. Method according to any of the claims 1 to 4, characterized by the fact that acquiring the real data of human interaction is carried out by tracking at least one of the following actions performed by a user: daily operation of a mobile phone, daily operation of a keyboard or touchpad, daily swiping actions on a screen of a user device and daily actions captured by a camera of a user device, holding the device or changing the device's location. 6. Method according to any of the claims 1 to 5, characterized by the fact that the at least one integrated sensor is a global positioning system, GPS, and/or a wifi system, and/or a Bluetooth device, and/or an accelerometer, and/or a gyroscope, and/or a magnetometer, and/or a pressure sensor and/or a keystroke feature acquisition sensor. 7. Training method of a bot detector module to detect whether a user action is a human or a synthetic action, characterized by comprising the step of training a detector module through an Automatic Learning method using the real training data acquired in accordance with any of the claims 1 to 6. 8. Method according to claim 7, characterized in that it further comprises training the bot detector module with a plurality of synthetic training data generated according to any of the claims 2 to 6. 9. Method according to claim 8 comprising: - providing the real training data acquired in accordance with any of the claims 1 to 6; - providing a plurality of synthetic training data generated according to any of the claims 2 to 6; - preprocessing the provided real training data and the provided plurality of synthetic training data, by selecting swiping gestures of more than 0.3 seconds on a screen; - normalizing such gestures to provide at least one of the following features: same origin coordinate and direction; - dividing the normalized data in three groups, each group comprising 60% of the normalized data for training, 20% of the normalized data for validation and 20% of the normalized data for testing; - training the bot detector module providing the 60% of the normalized data for training; - fitting hyperparameters with the 20% normalized data for validation and - testing the resulting classified data with the 20% of the normalized data for testing. 10. Bots detector module trained from the training data generated by the method according to any of the claims 1 to 6, the detector module configured to receive an input signal from a mobile device, and to generate an output comprising a percentage of confidence that the input signal is generated by a real human interaction. 11. Bots detection method comprising training a bots detection module according to the training method according to any of claims 7 to 9. 12. Bots detection method using the detector module according to claim 10, the detector configured to receive an input signal from at least one sensor of a mobile device, and to generate an output comprising a percentage of confidence, where the percentage of confidence comprises a probability that the input signal is generated by a real human interaction. 13. A system of detection of bots, the system comprising: a detector module in accordance with claim 10 at least one memory; and at least one processor to carry out the detection method in accordance with either claim 11 or 12. 14. A system for detecting bots, the system comprising: a detector module in accordance with claim 10: at least one memory; and at least one processor to carry out the training method in accordance with any of claims 7 to 9.
