Datasets:
Tasks:
Text Retrieval
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
10K - 100K
ArXiv:
License:
How are exactly calculated the IN and OUT values?
#3
by cbobed - opened
In the dataset, there are three values for the qrels domain (IN, OUT, NC); but I haven't found in the paper how the ndgc values are actually calculated. For example, in IN-domain evaluation, are the candidates filtered beforehand so only IN and NC documents are included in the candidate set or the candidate set contains everything but only the hits on IN-domain docs are considered (thus, OUT-domain docs would be counted as NC-ones - rel 0 ). I'm afraid I haven't been able to get this information anywhere.