Memetic search for overlapping topics based on a local evaluation of link communities

Abstract: In spite of recent advances in field delineation methods, bibliometricians still don't know the extent to which their topic detection algorithms reconstruct `ground truths', i.e. thematic structures in the scientific literature. In this paper, we demonstrate a new approach to the delineation of thematic structures that attempts to match the algorithm to theoretically derived and empirically observed properties all thematic structures have in common. We cluster citation links rather than publication nodes, use predominantly local information and search for communities of links starting from seed subgraphs in order to allow for pervasive overlaps of topics. We evaluate sets of links with a new cost function and assume that local minima in the cost landscape correspond to link communities. Because this cost landscape has many local minima we define a valid community as the community with the lowest minimum within a certain range. Since finding all valid communities is impossible for large networks, we designed a memetic algorithm that combines probabilistic evolutionary strategies with deterministic local searches. We apply our approach to a network of about 15,000 Astronomy & Astrophysics papers published 2010 and their cited sources, and to a network of about 100,000 Astronomy & Astrophysics papers (published 2003--2010) which are linked through direct citations.