Projet de thèse en Informatique
Sous la direction de Amel Bouzeghoub et de Naoufel Kraiem.
Thèses en préparation à Paris Saclay en cotutelle avec l'Ecole Nationale des Sciences de l'Informatiques , dans le cadre de École doctorale Sciences et technologies de l'information et de la communication (Orsay, Essonne) , en partenariat avec Télécom SudParis (France) (laboratoire) , ACMES (equipe de recherche) et de Institut national des télécommunications (Evry) (établissement de préparation de la thèse) depuis le 16-04-2018 .
Introduction of scientific issues and context of the study Social networking sites such as Twitter, Facebook, and YouTube are omnipresent in people's everyday lives. These sites originally served as a new way of socializing among friends, families and business people. However, in the last few years, some people have begun using these networks for purposes other than communicating with family and friends. These sites now often play a role in teen violence, cyberbullying and organized crime, which constitute a major threat to our security and in particular to the security of young people. Many statistics indicate a tremendous need for efficient approaches and systems that will be able to detect these violent behaviors and threats. The main long-term goal of this thesis is the creation of advanced approaches and systems that will be able to predict violent behaviors and threats in social networks, so that we can potentially limit the propagation of these behaviors among social network users, and significantly reduce the risks associated to social networking such as bullying or access to age inappropriate content. Such approaches and systems will be based on intelligent techniques e.g. Fuzzy logic and Genetic algorithm to produce easily interpretable knowledge, deal with a variety of structured and unstructured data types, and be efficient enough to handle large and complex data.
Detection of violent users and threats in social networks
Motivation Online Social Networks constitute an integral part of people's every day social activity. They provide platforms to connect people all over the world and share their interests. Recent statistics indicate for example, that every second, on average, around 6,000 tweets are tweeted on Twitter. Similarly, Facebook Messenger counted 700 million monthly active users, and the number of exchanged messages reached 800 million in January 2016. However, these network services also had many negative impacts and the existence of aggressive and bullying phenomena in such spaces is inevitable and therefore need to be addressed. Mining social network data to detect violent behaviors and threats is challenging for data mining, machine learning and artificial intelligence, for several practical reasons. First, different people have different ways of expressing the same violent behavior. Training an algorithm that works for everyone is difficult because of the variety of behaviors and the diverse ways in which behaviors are expressed, such as images, videos, and comments expressed in different languages. Second, the algorithms should have a way of detecting unseen potential violent behaviors and automatically adding them to the training set. Third, the multimodality and ultrahigh dimensionality of the structured and unstructured data available on social networking sites makes it challenging to develop data mining algorithms that will be able to extract relevant information useful for the detection of violent behaviors and threats. Finally, the algorithms must take into account the time-varying nature of networks to handle new users and links and automatically update the constructed models.