metadata

license: mit

Wikipedia Summary Dataset 128k

This is random subsample of 128k entries from the wikipedia summary dataset, processed with the following code:

import pandas as pd

df =  pd.read_parquet('wikipedia-summary.parquet')
df['l'] = df['summary'].str.len()
rdf = df[(df['l'] > 300) & (df['l'] < 600)]
# Filter out any rows 'topic' that have non-alphanumeric characters
mask = rdf['topic'].str.contains(r'^[a-zA-Z0-9 ]+$') == True
rdf = rdf[mask == True].sample(128000)[['topic', 'summary']].copy().sort_values('topic').reset_index(drop=True)
rdf.to_csv('wikipedia-summary-128k.tsv', sep='\t', index=False)

Citation

@mastersthesis{scheepers2017compositionality,
  author  = {Scheepers, Thijs}, 
  title   = {Improving the Compositionality of Word Embeddings},
  school  = {Universiteit van Amsterdam},
  year    = {2017},
  month   = {11},
  address = {Science Park 904, Amsterdam, Netherlands}
}