text_summarizer_czech

This project implements 3 different article summarizers for the Czech language:

1. Feature-based summarizer with RBM

   • file: summarizer_features_rbm.py
   • based on this article: https://arxiv.org/pdf/1708.04439.pdf
   • uses a combination of several features extracted from the text and an RBM to enrich those features

2. Matrix-based TextRank summarizer

   • file: summarizer_nlphackers_textrank.py
   • based mostly on this article: http://nlpforhackers.io/textrank-text-summarization/
   • uses an implementation of the TextRank algorithm from the text, but with Okapi BM25 similarity function and several other improvements

3. Graph-based TextRank summarizer

   • file: summarizer_machovec_modified.py
   • based on a thesis by P. Machovec at Masaryk University: https://is.muni.cz/th/359331/fi_m/Diplomova_prace.pdf
   • uses Petr's implementation of the TextRank algorithm, with several modifications

Usage

Python 3.6+ needed to run the scripts.

Install requirements:

pip install -r requirements.txt

Run one of the 3 summarizers without any arguments to see how it performs on a test set of several articles.

Run one of the 3 summarizers with 1 argument, a path to a file to have it print out a summary of the text in the file. The file has to contain just simple text (no xml, ...).

For example:

python summarizer_nlphackers_textrank.py my_article.txt