schmaett

7 ms·

1. 1.▲
   by schmaett 3y ago

   My results suggest that the main reason their entire thing works is because they compare substrings between some query text and all texts in the corpus. By precompiling the list of substrings that occur in the corpus, I can get their algori
2. 2.▲
   by schmaett 3y ago

   I can match the performance in their paper with lz4, and then by just explicitly comparing character-level n-grams between a query document and all documents belonging to the same topic.
3. 3.▲
   by schmaett 3y ago

   IMO their results have nothing to do with compression: When I replace gzip with lz4 (lower compression ratio, just does the "substring replacement" part of gzip), I get the same performance. When explicitly just comparing the subs