Aller au contenu. | Aller à la navigation

Outils personnels

Navigation

Voir le monde en interaction

  • logo-enssib.jpg
Vous êtes ici : Accueil / Agenda / Séminaires / Talk by Volodymyr Myr (EPFL): Anomaly detection in the dynamics of web and social networks using associative memory

Talk by Volodymyr Myr (EPFL): Anomaly detection in the dynamics of web and social networks using associative memory

Apart from being an essential part of modern life, social networks, online services, and knowledge bases generate a massive amount of logs, containing traces of global online activity on the Web. Most of this data is related to the standard activity of the users, however, the larger these logs become, the harder it is to detect deviations from the normal behavior in the network. Localization of this anomalies becomes even harder because of the continuous expansion and the dynamic nature of these networks. In this talk, we will present an unsupervised method for anomaly and event detection in large dynamic networks. We define an anomaly as a localized increase in temporal activity in a cluster of nodes. To demonstrate the performance of the algorithm, we apply it to the Wikipedia dataset. To detect anomalous patterns in user activity, we analyze the seven months logs of user activity on Wikipedia and its Web network structure (5K+ hours, 100K+ active pages, 6.5M+ links). We show that the anomalous spikes are triggered by the real-world events that impact the network dynamics. Besides, the structure of the clusters and the analysis of the time evolution associated with the detected events reveals interesting facts on how humans interact, exchange and search for information, opening the door to new quantitative studies on collective and social behavior on large and dynamic datasets.
Quand ? Le 09/11/2018,
de 10:00 à 11:00
Où ? Site Monod, M7.101
Ajouter un événement au calendrier vCal
iCal

Title: Anomaly detection in the dynamics of web and social networks using associative memory

Asbtract: 

Apart from being an essential part of modern life, social networks, online services, and knowledge bases generate a massive amount of logs, containing traces of global online activity on the Web. Most of this data is related to the standard activity of the users, however, the larger these logs become, the harder it is to detect deviations from the normal behavior in the network. Localization of this anomalies becomes even harder because of the continuous expansion and the dynamic nature of these networks.

In this talk, we will present an unsupervised method for anomaly and event detection in large dynamic networks. We define an anomaly as a localized increase in temporal activity in a cluster of nodes. To demonstrate the performance of the algorithm, we apply it to the Wikipedia dataset. To detect anomalous patterns in user activity, we analyze the seven months logs of user activity on Wikipedia and its Web network structure (5K+ hours, 100K+ active pages, 6.5M+ links). We show that the anomalous spikes are triggered by the real-world events that impact the network dynamics. Besides, the structure of the clusters and the analysis of the time evolution associated with the detected events reveals interesting facts on how humans interact, exchange and search for information, opening the door to new quantitative studies on collective and social behavior on large and dynamic datasets.

Volodymyr Miz, PhD candidate, EPFL, LTS2 (http://miz.space)

Joint work with Benjamin Ricaud and Pierre Vandergheynst