You will develop a multi-objective EA algorithm to find the non-dominated (Pareto) set of communities in a multi-graph. A multigraph is a graph where multi-types of edges can exist between vertices. The multi-graph that you will be experimenting on is based on bibliographical data of researchers.
This project involves experience with multi-objective optimisation, using Evolutionary Algorithms, and how to interpret the solutions from the algorithms. This is an unsupervised learning task. Part of the challenge is to come up with ways to validate, interpret and justify the communities found.
There are many possible graphs that can be extracted from the semi-structured DBLP data. In this project, we focus on two relations between researchers:
a) co-authorship and b) co-publications at same conferences. Co-authorship is the number of papers that the two incident researchers have written together. Co-publications is the number of unique conferences where both researchers have published in. In addition, each of the researchers have the number of years they have been active, which is the difference between the years of their first and latest publications.
Formally, the multi-graph can be represented as G(V; A;U; Y ), where V denotes the vertex set, A denotes the co-authorship edge weights between the vertices (A : V V ! N1), U denotes the co-publication edge weights (U : V V ! N) and Y denotes the vertex weights of the number of active years (Y : V ! N). In this project, we concentrate on partitional communities, i.e., the communities completely partition the set of vertices and there is no overlap between communities. The set of communities is denoted by C, a community is denoted by Ci, where i is an index and C(vi) is a function that returns the community index for vertex vi.
A community is a group of vertices that generally form a cohesive group. Because this is a multi-graph, a \cohesive group" in one relation might not
be one in the other relation. There are many denitions and objectives for communities, hence to simplify the project both relations are to be grouped
according to the weighted modularity objective. In the weighted modularity objective, we want to find a set of communities that minimises some objectives for the co-authorship relation:
3 freelancers are bidding on average $333 for this job
I am an expert in Matlab and optimization. I can help you in time. i have many experiences in a multi-objective EA algorithm to find the non-dominated (Pareto) set of communities in a multi-graph.