The CODDDE (Evolving communities, diffusion and events detection ) project will be granted by ANR.
Partners are :
- Laboratoire d’Informatique de Paris 6, Paris, France
- Laboratoire de l’Informatique du Parallélisme, Lyon, France
- SME, RTGI SAS Linkfluence, Paris, France
Complex networks appear in many contexts: sociology, with friendship networks, collaboration networks, computer networks, with the Internet, the World Wide Web, blog networks, P2P networks, biology, with food webs, genetic regulatory networks, metabolic networks, epidemiology, energy and transportation with road networks, power grids, railways, airline routes, but also in economics, linguistics and many others.
The emerging and promising domain of complex networks study, and in particular social ones, has proposed a stream of studies aimed at identifying properties of complex networks, their causes and consequences, describing their evolution and capturing everything into relevant models. These properties are used as key parameters in the study of various phenomena of interest like robustness, spreading of information, ideas or viruses. These questions are very transversal since diffusion phenomena can occur on many different social networks (including online ones) in the form of diseases, ideas, innovations, news or rumor, on computer networks with computer viruses which can propagate directly or by mail, p2p exchange systems, etc…
Whether we consider the evolution of complex networks or the spreading of anything on them, we cannot expect it to be monotonic but rather to be composed of a background evolution with some specific events of prime interest. These events may correspond to abnormal activity on the network: for instance phone-call networks are over saturated for New Year’s Eve, news blogs can be very reactive to specific events and in some occasion one per thousand to one percent of all blogs can talk about a given subject during one day (see today’s highlights on blogpulse.com for instance). Events can also be related to sudden changes in the network structure, e.g. routing tables updates after a link failure. Being able to detect or predict such events is of key interest.
The CODDDE project aims at studying critical research issues in the field of real-world complex networks study:
- How do these networks evolve over time?
- How does information spread on these networks?
- How can we detect and predict unexpected changes in their structure?
In order to answer these questions, an essential feature of complex networks will be exploited: the existence of a community structure among nodes of these networks. Complex networks are indeed composed of internally densely connected groups that have few interactions with one another.
The CODDE project will therefore propose new community detection algorithms to reflect complex networks evolution, in particular with regards to diffusion phenomena and anomaly detection. These algorithms and methodology will be applied and validated on a real-world online social network consisting of more than 10 000 blogs and French media collected since 2009 on a daily basis (the dataset comprises all published articles and the links between these articles) correlated with a twitter dataset.
The consortium of the project comprises two academic partners from large research labs, with a strong experience in complex networks: the Complex Networks team from LIP6-UPMC (Laboratoire d’Informatique de Paris 6 of Université Pierre et Marie Curie), and the LIP-ENS Lyon team (Laboratoire d’Informatique du Parallélisme of the École Normale Supérieure de Lyon), and one industrial partner: the Linkfluence SME, who will be in charge of data collection. Moreover, the expertise of Linkfluence blogs analysts will be used during the results validation phase.