Download Latest Version Bump min. deps + fix Rouge evaluation bugs source code.zip (264.6 kB)
Email in envelope

Get an email when there's a new version of Automatic text summarizer

Home / v0.13.0
Name Modified Size InfoDownloads / Week
Parent folder
Bump min. deps + fix Rouge evaluation bugs source code.tar.gz 2026-08-12 230.0 kB
Bump min. deps + fix Rouge evaluation bugs source code.zip 2026-08-12 264.6 kB
README.md 2026-08-12 2.4 kB
Totals: 3 Items   497.0 kB 0
  • INCOMPATIBILITY: cosine_similarity, unit_overlap, AbstractSummarizer.__init__, ItemsCount and the ROUGE evaluation functions raise TypeError instead of ValueError for invalid argument types.
  • INCOMPATIBILITY: Rouge-L (Summary Level) scores change to match the official ROUGE-1.5.5.pl.
  • INCOMPATIBILITY: ROUGE-L functions raise ValueError, not ZeroDivisionError, when either input has no words.
  • INCOMPATIBILITY: nltk>=3.9 is now required, since 3.8.2 withdrew the punkt pickle over CVE-2024-39705. See https://github.com/miso-belica/sumy/issues/216
  • INCOMPATIBILITY: Gave every dependency a lower bound, so installing sumy alongside an old release no longer silently produces a broken environment.
  • FIX: Excluded nltk==3.10.1, which fails to import when the venv sits inside the working directory. See https://github.com/nltk/nltk/issues/3730
  • FIX: Raised the setuptools floor to 70.0.0 (78.1.1 on Python 3.9+) to fix two advisories, GHSA-cx63-2mw6-8hw5 and GHSA-5rjg-fvgr-3xxf. Python 3.8 stays on 70.0.0 and remains exposed to the second, as 75.3.4 is its last installable setuptools.
  • FIX: Capped setuptools<82, since breadability imports pkg_resources, which v82 removed. See https://github.com/miso-belica/sumy/issues/235
  • FIX: Fixed ZeroDivisionError in rouge_l_sentence_level and rouge_l_summary_level when the summary and reference share no words. See https://github.com/miso-belica/sumy/issues/128
  • FIX: Fixed ZeroDivisionError in rouge_n for a reference shorter than the n-gram size; it now raises ValueError.
  • FIX: Fixed Rouge-L losing repeated reference words, which capped even a perfect summary below 1.0.
  • FIX: Fixed Rouge-L (Summary Level) scoring above 1.0 for a repetitive reference; hits are now clipped to the candidate's word counts, as ROUGE-1.5.5.pl does.
  • FIX: Fixed PlaintextParser treating an ordinary sentence in a caseless script as a heading. See https://github.com/miso-belica/sumy/issues/110 and https://github.com/miso-belica/sumy/issues/215
  • FIX: Fixed KeyError in SumBasicSummarizer from inconsistent word processing order. See https://github.com/miso-belica/sumy/pull/240
  • CHORE: Removed leftover Python 2 compatibility code.

Full Changelog: https://github.com/miso-belica/sumy/compare/v0.12.0...v0.13.0

Source: README.md, updated 2026-08-12