Community Detection by Ensemble - Supervised model using Edge, Node and Graph Attributes (E-SEGNA)

Attributes (E-SEGNA) Govardana Sachithanandam Ramachandran Abstract Major social network sites such as Facebook, Twitter & Google+ offer manual classification of friends in the form of List or Circle. With average node degree on these site increasing & with user’s temporal interest evolves the manual classification & re-classification of users become laborious. The project baselines the effectiveness two of more popular existing approaches which uses network structure and nodes attributes to determine number of circles, node membership of these circles and aspects that influences the circle formation. With the baseline number the project analyzes shortcomings of these approaches and tries improvements on the existing approaches. The project is an adaption of problem “Learning Social Circles in Networks” competition in kaggle.com Dataset The dataset used for this project is published @ https://www.kaggle.com/c/learning-social-circles/data . It contains labeled social network data in Facebook, Google+ and Twitter. Properties Values Ego Networks 1,143 Circles 5,541 Users 192,075 Node Attributes 26 [Table:1 The dataset contains]