SCCS Colloquium - Aug 29, 2019

Date:	August 21, 2019
Room:	02.07.023
Time:	15:00 - 16:00

Mohamed Farghal: Finding suspiciously dense components in dynamic edge-attributed graphs

This is a Master's thesis submission talk. Mohamed is advised by Prof. Thomas Huckle.

How can we find anomalies in complex interaction graphs? Many authors have shown that dense subgraphs are often some types of anomalies or generally interesting patterns in the graph. To model real-world data as graphs we often need to have complex edge labels as part of the graph, for example a social review graph is formed among users who review some businesses, in this graph we would expect edges to hold the text of the review, timestamp and possibly some rating value. Given such complex graphs, we aim at finding interesting dense subgraphs which we wouldn’t able to find without incorporating those edge labels. In this thesis (1) we propose new type of modelling edge-attributed graphs (2) we propose a novel density definition in line with our graph representation (3) accordingly, we present an algorithm that finds dense subgraphs.

Keywords: Graph Mining

Language: English

We are looking for a second talk! Please use the Colloquium registration form.