XMedia: Web People Search by Clustering with Machinely Learned Similarity Measures

In this paper we present an approach to person name disambiguation that clusters documents on the basis of textual features using cosine similarity and a machinely learned meta similarity measure. The approach achieves an F-measure of B-Cubed Precision and Recall of 0.741 on the Clustering Subtask for WePS-2. Such task consists of clustering a set of documents that mention an ambiguous person name according to the actual entities referred to that name.