Extraction of enriched social network for role analysing.
Summary
In this work, we automatically extract social network from forum debates on news website. We extract and create the social network based on the structural relation. The structural relation is based on the structure of the data, in our context the “reply-to” link that allows users to personally replies to another one. Then, we enriched the social network with two relations extracted from the content of the posts: the name citation relation and the text quotation relation. This enriched social network performed the interaction comprehension between users and help us to find interesting people inside their communities.
Team
Mathilde Forestier, PhD student and Julien Velcin, Assistant professor, ERIC Lyon 2
Topological classifier
Summary
In supervised learning, Ensemble Methods (EM) showed their qualities. One of the reference methods in this area is the Random Forest (RF). This is based on partitioning the representation space by boundaries parallel to the axes or oblique. the consequences in this way to partition the space of representation can affect the quality of each predictor. We felt that this approach could be improved if it is freed from this constraint in order to better fit the topological structure of the training set. In this work, we study new EM-based neighborhood graph whose performance, on our first experiments, are as good as those of RF.
Team
Diala Azzedine, PhD student, Fabien Rico, Assistant professor, ERIC Lyon 1
Rumors and propagation of information in social networks.
Summary
This research work analyzes and questions the cognitive and social processes related to information processing in “online social networks”. The first phenomenon on which this thesis will focus is the diffusion of information. This is motivated by the need to understand the paths that information takes in social networks. This could permit to predict the population reached by a given information, its spreading speed, etc. . Understanding this process will help us to detect phenomena like the spread of rumors that can, for instance, result from the exchange of conflicting information in the network. Finally, the third point this thesis will address is the quality of information. We thus intend to understand and distinguish real information and “misinformation”, on the basis of the properties identified in the two previous points. Thus, the goal of this thesis is to propose methods, techniques and tools able to extract a set of measures and facts related to information and its characterization in online social networks . Furthermore, it is expected that a tool incorporating these techniques and technologies will be implemented and validated at the end of this thesis.
Team
Adrien Guille, PhD student and Cécile Favre, Assistant professor, ERIC Lyon 2
