Application deadline: (a soon as possible)
Host organizations :
Academic advisor : Guy Melançon,
Université de Bordeaux / CNRS UMR 5800 LaBRI
Adresse : CNRS LaBRI UMR 5800
Campus Université Bordeaux I
351 Cours de la Libération
33405 Talence Cedex
Send your resume, intention letter and graduate studies record. Please clearly indicate why and how your profile is a good fit for the project. A description and pointers at past projects and realizations related to the topic is a plus.
Classified ads hide a turbulent activity to try and capture customers. Although ads originally are posted by individual users to hopefully find an interested reader, sellers organize into networks and implement a planned strategy. By looking at ads, we may hope to compare their content, build a network of cross-reference and hopefully read these strategies ultimately leading to the identification of the underlying organizations.
All these cross-referenced ingredients naturally builds into a layered network mixing different semantic entities such as places (towns), telephone numbers, services or goods that are being offered, pictures. Ads may compare because sellers use the exact same picture for different goods, a same phone number for a set of ads may indicate the presence of an organization (instead of individual sellers), etc.
The goal of this Master's degree project is to first implement a piece of work leveraging the content of the ads into a multilayer network. The layered network model needs to be both robust (so we can deploy and collect ads from different classified ads websites) and flexible so it can usefully support analysis. The second phase of the project is to try and build visualizations of these networks (most probably sub-networks obtained by extracting useful views), using multiple network views and allowing to simultaneously query/filter operations.
A small of the project will require to build/tune a web scraper. The core part of the project is an opportunity to extend Renoust et al.'s approach to multiplex networks in relation with searching the web. We expect the necessity to redefine/redesign usual network metrics (centralities) by adapting them to these layered networks.