A Link-Based Memetic Algorithm For Reconstructing Overlapping Topics From Networks Of Papers And Their Cited Sources

In spite of recent advances in field delineation methods, enduring problems such as the impossibility to justify necessary thresholds and the difficulties in comparing thematic structures obtained by different algorithms leave bibliometricians with a sense of uneasiness about their methods. In this paper, we propose and demonstrate a new approach to the delineation of thematic structures that attempts to fit the methods for topic delineation to the properties of topics. We derive principles of topic delineation from a theoretical discussion of thematic structures in science. Applying these principles, we cluster citation links rather than publication nodes, use predominantly local information and grow communities of links from seeds in order to allow for pervasive overlaps of topics. The complexity of the clustering task requires the application of a memetic algorithm that combines probabilistic evolutionary strategies with deterministic local searches. We demonstrate our approach by applying it to a network of 14,954 Astronomy & Astrophysics papers and their cited sources.