Title : Link Prediction Using Time Series of Neighborhood-based Node Similarity Scores.
Orateur : Grégoy Martin (LACODAM)
Reading group presentation
Article: Ismail Gunes, Sule Gunduz-Oguducu and Zehra Cataltepe. “Link
prediction using time series of neighborhood-based node similarity
scores”. en. In : Data Mining and Knowledge Discovery 30.1 (jan. 2016),
p. 147─180.
The task of deciding whether an edge in a graph should be marked as present or absent is called a “link prediction” task. This decision is not trivial and can depend on the graph topology or on already known labels on nodes/edges. One common application of this task is about social networks, such as Facebook friendship network or co-authorship network of scientific papers. Research in this field is often focused on whether modeling the reality (e.g Facebook friendship network vs friendship network in reality) or predicting the link creation in the general future (e.g evolution of the co-authorship network). However, those approaches can not take temporal intervals into account, leading to the impossibility to predict if a link could appear at t+3 rather
than t+1.
This reading group is about the journal article “Link Prediction Using Time Series of neighborhood-based Node Similarity Scores” which presents a method in order to incorporate those time intervals in the link predict task through the use of times series modeled by ARIMA backed by topological measures on graphs.
Support de présentation
