An analysis of the structure and evolution of the scientific collaboration network of computer intelligence in games

Abstract

Games constitute a research domain that is attracting the interest of scientists from numerous disciplines. This is particularly true from the perspective of computational intelligence. In order to examine the growing importance of this area in the gaming domain, we present an analysis of the scientific collaboration network of researchers working on computational intelligence in games (CIG). This network has been constructed from bibliographical data obtained from the Digital Bibliography & Library Project (DBLP). We have analyzed from a temporal perspective several properties of the CIG network at the macroscopic, mesoscopic and microscopic levels, studying the large-scale structure, the growth mechanics, and collaboration patterns among other features. Overall, computational intelligence in games exhibits similarities with other collaboration networks such as for example a log-normal degree distribution and sub-linear preferential attachment for new authors. It also has distinctive features, e.g. the number of papers co-authored is exponentially distributed, the internal preferential attachment (new collaborations among existing authors) is linear, and fidelity rates (measured as the relative preference for publishing with previous collaborators) grow super-linearly. The macroscopic and mesoscopic evolution of the network indicates the field is very active and vibrant, but it is still at an early developmental stage. We have also analyzed communities and central nodes and how these are reflected in research topics, thus identifying active research subareas.