Análisis de Textos en Dispositivos Android para la Detección de Grooming
Loading...
Download
Official URL
Full text at PDC
Publication date
2024
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
El avance tecnológico o la llamada tercera revolución industrial avanza a pasos agigantados y cada vez está más presente en nuestra vidas, con la popularización de los dispositivos móviles y el auge de las redes sociales, es cada vez menor la edad con la que se entra en contacto con la tecnología y a su vez es cada vez más temprana la edad en la que nos exponemos al riesgo del mundo digital, es bien sabido que en este mundo digital hay muchos perpetradores sexuales acechando a niños, a esta práctica se le denomina Grooming. El presente trabajo tiene como objetivo principal analizar y clasificar las conversaciones que tenga un niño a través de su dispositivo móvil con un posible perpetrador, con el fin de lanzar una alerta temprana a los padres y así lograr frenar el abuso en las redes sociales. Para lograr dicho objetivo se entrenó un modelo propio con la técnica de fine-tuning usando como base un pequeño modelo de lenguaje de gran tamaño (tinyLLM) llamado MobileBERT y un conjunto de datos llamado ChatCoder el cual contiene chats y logs entre abusadores y víctimas. Este modelo clasifica cada mensaje en cuatro posibles fases, las cuales son: amistad, relación sentimental, sexual y abordaje o conclusión. Cada fase determina cual es el nivel de relación entre el niño y su potencial perpetrador. El modelo entrenado se importó en una aplicación Android, la cual analiza cada mensaje escrito en una conversación, y hace un análisis para lanzar una alerta a los padres de detección temprana de depredadores sexuales o eSPD por sus siglas en Ingles.
Technological progress or the so-called third industrial revolution is advancing by leaps and bounds and is increasingly present in our lives, with the popularization of mobile devices and the rise of social networks, the age at which we come into contact with technology is getting younger and at the same time the age at which we expose ourselves to the risk of the digital world is getting younger, it is well known that in this digital world there are many sexual perpetrators stalking children, this practice is called Grooming. The main objective of this work is to analyze and classify the conversations that a child has through his mobile device with a possible perpetrator, in order to launch an early warning to parents and thus manage to curb abuse in social networks. To achieve this goal, a fine-tuning model was trained by ourselves using a tinyLLM model called MobileBERT and a dataset called ChatCoder containing chats and logs between abusers and victims. This model classifies each message into four possible phases, which are: friendship, romantic relationship, sexual, and approach or conclusion. Each phase determines the level of relationship between the child and the potential perpetrator. The trained model was imported into an Android application, which analyzes each message typed in a conversation, and makes an analysis to launch an alert to parents for early sexual predator detection or eSPD.
Technological progress or the so-called third industrial revolution is advancing by leaps and bounds and is increasingly present in our lives, with the popularization of mobile devices and the rise of social networks, the age at which we come into contact with technology is getting younger and at the same time the age at which we expose ourselves to the risk of the digital world is getting younger, it is well known that in this digital world there are many sexual perpetrators stalking children, this practice is called Grooming. The main objective of this work is to analyze and classify the conversations that a child has through his mobile device with a possible perpetrator, in order to launch an early warning to parents and thus manage to curb abuse in social networks. To achieve this goal, a fine-tuning model was trained by ourselves using a tinyLLM model called MobileBERT and a dataset called ChatCoder containing chats and logs between abusers and victims. This model classifies each message into four possible phases, which are: friendship, romantic relationship, sexual, and approach or conclusion. Each phase determines the level of relationship between the child and the potential perpetrator. The trained model was imported into an Android application, which analyzes each message typed in a conversation, and makes an analysis to launch an alert to parents for early sexual predator detection or eSPD.
Description
Trabajo de Fin de Máster en Ingeniería Informática, Facultad de Informática UCM, Departamento de Ingeniería de Software e Inteligencia Artificial, Curso 2023/2024.













