Abstract

The Facebook social graph is growing. Currently Facebook has nearly 1.5B monthly active users who have produced 2T posts. Facebook Search is one of the pillars of this growth – by affording users to search for people they know in daily life and by allowing the discovery of new content from deep within the Graph. Often we might know what our friend is called, but not how to spell their name in order to connect with them. Daniel Bernhardt works on the Speller Team at Facebook London which is working on this problem. The talk will focus on how the team uses machine learning to build the Facebook Speller and what challenges lie ahead.

Daniel has been working on the Search team at Facebook London since August 2013. He is interested in search relevance, metrics and query rewriting. His recent work concerns the relevance stack for the Graph Search Speller as well as other NLP components including synonym expansions. His team has developed a number of machine learning components for ranking and classification as well as quality measurement. Before joining Facebook he was a software engineer in web search at Bing London, working on a number of projects including news classification, business taxonomy mining and query prediction. In 2009 he received a PhD in affective computing from the Computer Laboratory, University of Cambridge. His supervisor was Peter Robinson.