Matching Heuristic causes Mismatches, e.g. "Deep Learning". Check for >=1 same author?

[lgalke]

Big thanks to the people behind rebiber. It is a very helpful tool.

I noticed some peculiarities with the matching heuristic (see example below).
TLDR: rebiber turned the entries for the Deep Learning book (Goodfellow) and the Deep Learning nature article (LeCun, Bengio, Hinton) into entries referring to a SIGKDD paper (Salakuthdinov) with the same title--without any warning.

Based on this behaviour, I assume the matching is done only based on title.

I suggest to check additionally for at least one common author. Or (less invasive): to emit an adequate warning when there is not a single same-author-name in a new entry compared to the original entry.

Input

@article{deeplearning,
  title = {Deep Learning},
  author = {LeCun, Yann and Bengio, Yoshua and Hinton, Geoffrey},
  year = {2015},
  month = may,
  journal = {Nature},
  volume = {521},
  number = {7553},
  pages = {436--444},
  publisher = {{Nature Publishing Group}},
  issn = {1476-4687},
  doi = {10.1038/nature14539},
  copyright = {2015 Nature Publishing Group, a division of Macmillan Publishers Limited. All Rights Reserved.},
}

@book{goodfellow_deep_2016,
	series = {Adaptive computation and machine learning},
	title = {Deep {Learning}},
	isbn = {978-0-262-03561-3},
	url = {http://www.deeplearningbook.org/},
	publisher = {MIT Press},
	author = {Goodfellow, Ian J. and Bengio, Yoshua and Courville, Aaron C.},
	year = {2016},
}

Cmd: rebiber -i main.bib -o main.bib -r editor -d True -s True

Output

@inproceedings{deeplearning,
 author = {Ruslan Salakhutdinov},
 bibsource = {dblp computer science bibliography, https://dblp.org},
 biburl = {https://dblp.org/rec/conf/kdd/Salakhutdinov14.bib},
 booktitle = {Proc. of KDD},
 doi = {10.1145/2623330.2630809},
 pages = {1973},
 publisher = {{ACM}},
 timestamp = {Tue, 06 Nov 2018 00:00:00 +0100},
 title = {Deep learning},
 url = {https://doi.org/10.1145/2623330.2630809},
 year = {2014}
}

@inproceedings{goodfellow_deep_2016,
 author = {Ruslan Salakhutdinov},
 bibsource = {dblp computer science bibliography, https://dblp.org},
 biburl = {https://dblp.org/rec/conf/kdd/Salakhutdinov14.bib},
 booktitle = {Proc. of KDD},
 doi = {10.1145/2623330.2630809},
 pages = {1973},
 publisher = {{ACM}},
 timestamp = {Tue, 06 Nov 2018 00:00:00 +0100},
 title = {Deep learning},
 url = {https://doi.org/10.1145/2623330.2630809},
 year = {2014}
}

The command line output of rebiber only state

• Converted. ID: deeplearning ; Title: Deep Learning
• Converted. ID: goodfellow_deep_2016 ; Title: Deep {Learning}

Desired output

a) Do not have the entries replaced, as there is not a single author name in common

or b) Emit a major warning on the command line, when replacing without a single author name in common

[lgalke]

Related: An unsupervised heuristic-based approach for bibliographic metadata deduplication (IPM 2011)

[yuchenlin]

Hey @lgalke thanks a lot for reporting this error and the suggestions. I feel like the waning is an easier option and we can let the users to deal with this manually. However, I'm not having enough bandwidth recently. If you'd like to create a PR to output a warning message, it would be much helpful! Thank you very much!