Some datasets that measure similarity among items, including as some in BEIR, are meant to ignore the item itself (i.e., discard results where the qid matches the docno) during evaluation. Adding this as a parameter of the measures would enable this type of evaluation and remove evaluation ambiguity. (e.g., rather than making it a global parameter)
Some datasets that measure similarity among items, including as some in BEIR, are meant to ignore the item itself (i.e., discard results where the qid matches the docno) during evaluation. Adding this as a parameter of the measures would enable this type of evaluation and remove evaluation ambiguity. (e.g., rather than making it a global parameter)