sisu 0.2.0

Performs extractive, hierarchical, summarization out of a corpus of documents.

Structured and Interactive Summarization

Performs extractive, hierarchical, summarization out of a corpus of documents.

• Free software: BSD license

• Documentation: https://balouf.github.io/sisu/.

Features

• Preprocessing tools.

Credits

This package was created with Cookiecutter and the francois-durand/package_helper_2 project template.

History

0.X.X (2021-XX-XX): TODO

• Run experiments and comparison on the flat summarizer

• Start converting the Wikipedia Animals dataset

• Start converting hierarchical summary notebook

0.2.0 (2021-03-14): Flat summarizer

• Main update: fully-functional flat summarizer!

  • Fully customisable;

  • Fully documented.

• Two tutorials

  • Building a Gismo for the Covid dataset;

  • Flat summarizer on the Covid dataset.

• Change of paradigm: start from the notebook and build the module cell by cell.

• Consequences: all non-converted modules from 0.1.2 are moved to the pit. They will be restored during the notebook transformation.

• sentence splitter optimized big time using nltk hidden features!

0.1.2 (2021-03-03): the pit

• Batch import of remaining modules in a temporary submodule (the pit). The pit will be dispatched afterwards.

• Fix import issues (e.g. spacy neuralcoref version incompatibility, Qt5, sknetwork…)

• submodule gismo_wrapper on death row (may never leave the pit)

• Embedding_idf OK

• Building summary: summarize and make_tree have been updated to work.

• Lot’s of cleaning remains (separating covid/generic, unified pre-proc and source convention,…)

• Take down neuralcoref for the moment. Does not build on github.

0.1.1 (2021-02-23): data_loader

• Finish import / transformation of the data_loader module.

0.1.0 (2021-02-23): First release

• First release on PyPI.

• Preprocessing submodule deployed