tuhailong

/

cross-encoder-bert-base

like 1

From sbert cross encoder page, I find two train example, one, two. While one use CECorrelationEvaluator, two use CERerankingEvaluator.
I want use cross encoder to ranking the search items, which way should i use. From word meaning i think reranking maybe more suitable. But it's label only has 0/1. correlation's label could be number represent the relevance between query and doc.