Wearable wireless RFID devices provide novel ways to follow face-to-face interactions of people in various settings. Their application in pre-schools to record daily contacts and oral interactions of children can disclose how the social networks and linguistic skills of children co-develop over time during early time of schooling.
My Thesis addresses this challenge through the design, performance, and analysis of a large-scale social experiment carried out in a French pre-school. In this project we used wireless RFID sensors to collect proximity and voice data from over 200 participants (children and staff) during three years of observation period, for one week in each month. In parallel, we collected extensive ground truth data and using periodic questionnaires we followed the socio-demographic background and linguistic development of children.
The first goal of the thesis focused on the collection and processing of raw RFID sensor data using conventional data cleaning and signal processing techniques, with fueling of customised methods.
As a second goal I developed techniques to precisely reconstruct the interactions of participants as temporal networks, using advanced machine learning methods applied on sequential data. Using the reconstructed social networks and recorded linguistic and socio-demographic attributes of children, I conducted a multivariable statistical analysis to study the effects of homophily, inducing overrepresented social interactions between linguistically and demographically similar individuals.
Finally a visualisation technique of linage graphs is presented.
Gratuit
Disciplines