Point at the Triple: Generation of Text Summaries from Knowledge Base Triples

This repository accompanies our article entitled: "Point at the Triple: Generation of Text Summaries from Knowledge Base Triples" which has been published at the Journal of Artificial Intelligence Research. It contains the code along with the datasets that we used in our experiments.

Datasets

We used two datasets of aligned knowledge base triples from DBpedia with snippets of text.

• Biographies: triples aligned with Wikipedia biographies
• Full: triples aligned with open-domain Wikipedia summaries (incl. biographies)

The first is the D1 dataset, which has been provided by https://www.sciencedirect.com/science/article/pii/S1570826818300313. It can be downloaded by following the instructions at: github.com/pvougiou/Neural-Wikipedian.

In order to download and un-compress the Full dataset in its corresponding folder Full, in a Unix shell environment execute: sh download_datasets.sh. The dataset folder contains two sub-folders:

• data contains the aligned dataset in binary-encoded pickle files. Each file is a hash table. Each hash table is a Python dictionary of lists.
• utils contains the dataset's supporting files, such as hash tables of the instance types and the labels of the entities. All the files are binary-encoded in pickle files.

dataset.ipynb is an iPython Notebook that allows easier inspection of the Full dataset. It provides also further details regarding the structure of the dataset and the functionality of its supporting files.

BibTeX

Please cite the following paper should you use this repository in your work.

@article{Vougiouklis2020,
  author    = {Pavlos Vougiouklis and
               Eddy Maddalena and
               Jonathon S. Hare and
               Elena Simperl},
  title     = {Point at the Triple: Generation of Text Summaries from Knowledge Base
               Triples},
  journal   = {J. Artif. Int. Res.},
  volume    = {69},
  pages     = {1--31},
  month     = sep,
  year      = {2020},
  url       = {https://doi.org/10.1613/jair.1.11694},
  doi       = {10.1613/jair.1.11694},
}

License

This project is licensed under the terms of the Apache 2.0 License.