Correlating Topic Rankings and Person Rankings to Find Experts

Expert search is about finding people rather than documents. The goal is to retrieve a ranked list of candidates with expertise on a given topic. The task is studied in the context of the enterprise track. We describe an approach that compares topic profiles and candidate profiles directly. These profiles are not based on unordered sets of documents, but on ranked lists. This allows us to differentiate between documents that are highly related to a topic or a candidate and documents that are only marginally related. The ranked lists for topics and candidates are obtained by simple retrieval queries. The correlation between the ranked list of documents for a topic and the ranked list for a candidate is used as an indicator of the candidate’s expertise on the topic. We study different ways to rank documents for the candidate profiles as well as various ways of comparing the document and candidate based ranked lists. Experiments show that starting from the right candidate profiles, reasonable results can be obtained. Furthermore, it seems important to take a correlation measure that takes into account the orderings of documents in both the candidate profile and the documents profile.