BERT

Tags

text-processing

BERT is an [see page 74, extension] of CWR which accounts for the context in which a word is used.

CWR approaches generally attribute the same meaning to the same word, regardless of context. BERT lets us capture the use of words in a context.

BERT was initially used for query re-ranking (assert document relevance using BERT representations). More recent efforts aggregate sentence similarity scores between sentences in a candidate retrieved document and the query.