The problem of finding topic experts on social networking sites has been a continued topic of research. This thesis addresses the problem of identifying local experts in social media systems like Twitter. Local experts are experts with a topical expertise that is centered around a particular location. This geographically-constrained expertise can be a significant factor for enhanced answering of local information needs (What is the best pub in College Station?), for interacting with local experts (e.g., in the aftermath of a disaster), and for accessing local communities. I developed a local expert finding system - called OLE (online local experts) - that leverages the crowd sourced location-topic labels provided by users of the popular Twitter service. Concretely, I mine a collection of 108 million tweets for evidence of local topics of discussion occurring with user-mentions and location pairs; based on this collection, I developed a learning-to-rank approach that incorporates topic-location entropy and a local expert perimeter for varying the expertise focal window. In comparison with alternative expert finding approaches, I find that OLE is quite effective in finding local experts and achieves a 37.72% increase in mean average precision and a 16.8% increase in NDCG scores, across a comprehensive set of queries.
