manupande21 commited on
Commit
5968f95
·
verified ·
1 Parent(s): 3c2afe6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +9 -1
README.md CHANGED
@@ -4,6 +4,14 @@ license_name: ms-marco-non-commercial-research-license
4
  license_link: LICENSE
5
  ---
6
 
 
 
7
  First we filtered queries.train.tsv to only contain queries which were available in qrels.train.tsv. This was done so that we only take care of queries of which we had a positive example of. Then we found top 200 ranked passages in collection.tsv for each query in filtered queries and then randomly sampled a passage whose rank lied in the range of 50 to 200 to produce our final hard negatives for each query.
8
 
9
- Use this as a training dataset. Test your model against qrels.dev.tsv for MRR@10 or a metric of your choice.
 
 
 
 
 
 
 
4
  license_link: LICENSE
5
  ---
6
 
7
+ The dataset has 3 columns : query, positive_example, negative_example
8
+
9
  First we filtered queries.train.tsv to only contain queries which were available in qrels.train.tsv. This was done so that we only take care of queries of which we had a positive example of. Then we found top 200 ranked passages in collection.tsv for each query in filtered queries and then randomly sampled a passage whose rank lied in the range of 50 to 200 to produce our final hard negatives for each query.
10
 
11
+ Use this as a training dataset. Test your model against qrels.dev.tsv for MRR@10 or a metric of your choice.
12
+
13
+ The original datasets used to generate this can be found at : https://microsoft.github.io/msmarco/Datasets.html#passage-ranking-dataset.
14
+
15
+ The author does not promise on the correctness of the dataset
16
+
17
+ For research purposes only.