Datasets:

Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
ArXiv:
Libraries:
Datasets
Dask
License:
BerenMillidge commited on
Commit
d18591b
1 Parent(s): 3cee00c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +6 -0
README.md CHANGED
@@ -30,6 +30,12 @@ As such, we release a fully deduplicated version of DCLM in case it is of intere
30
 
31
  -/ TODO YURY
32
 
 
 
 
 
 
 
33
 
34
  ### Dataset Description
35
 
 
30
 
31
  -/ TODO YURY
32
 
33
+ ## Deduplication Details
34
+
35
+ -/ TODO Yury please add details on the exact process you used to deduplicate
36
+
37
+
38
+ We deduplicated DCM using the approximate minhash LSH method with the following parameters: minhash with signature size of 128 computed on character-based 25-grams signatures and split into 8 bands, giving roughly 85% Jaccard similarity threshold. We then constructed an undirected graph with nodes being documents and edges being duplicates, and found connected components in it, which provided us with clusters of duplicates. From each cluster, we selected the top document to keep and removed the rest.
39
 
40
  ### Dataset Description
41