I am planning to use S-Bert for asymmetric information retrieval purposes with highly technical data. Is there a best practice regarding how to manually annotate the data for fine-tuning? By annotation I mean to write the queries in the pair (query, document) for Multiple Negatives Ranking Loss.
I guess the query should be as close to the document paragraph as possible but of course not a copy-paste of it.
I am planning to use S-Bert for asymmetric information retrieval purposes with highly technical data. Is there a best practice regarding how to manually annotate the data for fine-tuning? By annotation I mean to write the queries in the pair (query, document) for Multiple Negatives Ranking Loss.
I guess the query should be as close to the document paragraph as possible but of course not a copy-paste of it.