Distributed Entity Disambiguation with Per-Mention Learning

Entity disambiguation, or mapping a phrase to its canonical representation in a knowledge base, is a fundamental step in many natural language processing applications. Existing techniques based on global ranking models fail to capture the individual peculiarities of the words and hence, struggle to meet the accuracy-time requirements of many real-world applications. In this presentation, we propose a new system that learns specialized features and models for disambiguating each ambiguous phrase in the English language. We train and validate the hundreds of thousands of learning models for this purpose using a Wikipedia hyperlink dataset with more than 170 million labelled annotations. The computationally intensive training required for this approach can be distributed over a cluster. In addition, our approach supports fast queries, efficient updates and its accuracy compares favorably with respect to other state-of-the-art disambiguation systems.