LAUSR

Named Entity Disambiguation is the task of assigning entities from a Knowledge Base to mentions of such entities in a textual document. This article presents two novel approaches guided by a natural notion of semantic similarity for the collective disambiguation of all entities mentioned in a document at the same time. We adopt a unified semantic representation for entities and documents—the probability distribution obtained from a random walk on a subgraph of the knowledge base—as well as lexical and statistical features commonly used for this task. The first approach is an iterative and greedy approximation, while the second is based on learning-to-rank. Our experimental evaluation uses well-known benchmarks as well as new ones introduced here. We justify the need for these new benchmarks, and show that both our methods outperform the previous state-of-the-art by a wide margin.