Detección de suplantación en redes sociales.
Archivos
NO SE HA AUTORIZADO la consulta de los documentos asociados
Fecha
2021-11-15
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Jaén: Universidad de Jaén
Resumen
El auge de redes sociales como Twitter, combinado con la disponibilidad de grandes volúmenes de información
publicada por los usuarios, se convierte en un entorno atractivo para los ciberdelincuentes. Estos actores
maliciosos, utilizan técnicas malintencionadas para suplantar la identidad de la víctima. Debido a este problema,
es conveniente identificar si un autor ha sido víctima de este fenómeno, haciendo uso de la estilometría para
determinar la autoría de un texto según el estilo de escritura del autor. Con este objetivo se creó TweetLib: una
herramienta diseñada para tareas comunes del PLN, como preprocesamiento del texto, extracción de
características y clasificación. Los resultados experimentales obtenidos, mostraron el gran potencial de las
estrategias de extracción de rasgos basadas en el aprendizaje profundo. Algoritmos como BERT, duplicaron (2x)
el rendimiento obtenido con respecto a las estrategias tradicionales de aprendizaje automático estudiadas en el
trabajo.
The rise of social networks like Twitter, combined with the availability of large volumes of text-based information published on social media by users, has become an attractive scenario for cybercriminals. These attackers use malicious techniques to impersonate the victim. Due to this problem, it is desirable to identify if an author is being targeted by these attacks, using stylometry for authorship attribution tasks of a given text, based on the author's writing style. With this goal in mind TweetLib was created: a tool designed for common NLP tasks such as text preprocessing, feature extraction and classification. The experimental results obtained, showed the high potential of feature extraction techniques based on deep learning. Algorithms like BERT, were able to double (2x) the performance obtained by traditional machine learning approaches studied in this work.
The rise of social networks like Twitter, combined with the availability of large volumes of text-based information published on social media by users, has become an attractive scenario for cybercriminals. These attackers use malicious techniques to impersonate the victim. Due to this problem, it is desirable to identify if an author is being targeted by these attacks, using stylometry for authorship attribution tasks of a given text, based on the author's writing style. With this goal in mind TweetLib was created: a tool designed for common NLP tasks such as text preprocessing, feature extraction and classification. The experimental results obtained, showed the high potential of feature extraction techniques based on deep learning. Algorithms like BERT, were able to double (2x) the performance obtained by traditional machine learning approaches studied in this work.
Descripción
Palabras clave
Seguridad Informática