Isaac Bloomberg Meets Michael Bloomberg

Abstract

This paper shows the implementation and evaluation of the Entity Linking or Named Entity Disambiguation system used and developed at Bloomberg. In particular, we present and evaluate a methodology and a system that do not require the use of Wikipedia as a knowledge base or training corpus. We present how we built features for disambiguation algorithms from the Bloomberg News corpus, and how we employed them for both single-entity and joint-entity disambiguation into a Bloomberg proprietary knowledge base of people and companies. Experimental results show high quality in the disambiguation of the available annotated corpus.

Named entity recognition with document-specific KB tag gazetteers

“…For example, the New York Times Annotated Corpus (Sandhaus, 2008) contains more than 1.5M articles "manually tagged by library scientists with tags drawn from a normalized indexing vocabulary of people, organizations, locations and topic descriptors". Names and types are also present in large quantities of financial news stories from Bloomberg (Bradesko et al, 2015), in the form of linked names of companies and people. Document-level tags may be quicker for annotators to apply than the usual method of marking spans in text, and are thus a cheap form of supervision.…”