How does dblp detect coauthor communities?

We use the following simple strategy to identify coauthor communities: Consider the neighborhood graph consisting of the target person and all (direct) coauthors of that person. Two nodes in this graph are connected by an edge if and only if they are coauthors (or co-editors) of a publication listed in dblp. Now remove the node of the target person and all incident edges. E ach component of the remaining graph is identified as a community.

We make use of the coauthor communities in our quality assurance process. If all coauthors seem to belong to the same community, then we are quite confident that the publication entries on this page do actually belong to a single person, as is intended. If the coauthor list is clustered into different communities, then this might be an indicator for a case of two or more homonymous authors being mixed up on a single person's page. In such a case, the coauthors can often be separated in disjoint groups that correspond to homonymous author entities who work in several distant research areas with unconnected colleagues.

However, this heuristic is not necessarily a reliable indication of the absence of homonymous author entities. One scenario we encounter quite often is the case when the coauthors index itself contains many homonyms. In that case, these "false witnesses" given by the homonymous coauthors forces the neighboring coauthor graph of a person into one big connected component. Unfortunately, this scenario happens quite frequently in the notoriously difficult case of Asian author names.

At the same time, there are two main reasons for a fragmented coauthor index to happen even in non-homonymous cases: If dblp only list a small sample of the publications of a person or her community, then our data on the coauthor topology is simply insufficient to yield conclusive results. This happens quite often in the case of senior researchers who are working in areas that are only partially covered by dblp. The other reason are changes in a person's biography, such as a change of the affiliation or a reorientation of a person's research interest, which may be accompanied by joining a completely new research community.

Therefore, checking the coauthor graph for indications of a homonym is always done manually by a member of the dblp team. It remains an open challenge to develop algorithms which indicate homonyms with a reliable precision.

maintained by Schloss Dagstuhl LZI at University of Trier