Add SUMO Web Claims Summarization scenario

[yifanmai]

This is a summarization benchmark based on the climate subset of the SUMO dataset (Mishra et al., 2020).

[yifanmai]

Hi @ryokawajp, please review this pull request.

By approving, you acknowledge that:

• You will be listed as a co-author of the merge commit
• You agree to license your work under Apache 2.0

I have made the following changes have been made on top of your original commit:

• Fixed a bug where truncation happened before filtering, which would cause long examples to not get filtered if the truncation length is less than the filter length.
• Reorganized arguments to the scenarios to make their purposes more clear.
• Removed the round trip to CSV.
• Removed the use of Hugging Face datasets
• Moved column names to class constants.
• Moved configuration constants to args in get_sumosum_spec().
• Fixed _clean_and_truncate() to replace all sequences of whitespace.
• Added type annotations.
• Removed unnecessary extra call to _clean_and_truncate().
• Added scenario description and citation.

[ryokawajp]

Thank you, @yifanmai .
I suppose one needs pip install openpyxl because it depends on loading an excel file in SUMOSumScenario._load_dataset(). Otherwise, it cause an import error. The module is indirectly called from Pandas pd.read_excel(). Is it OK to add this to requirements.txt?

• https://pypi.org/project/openpyxl/

[ryokawajp]

Let me add @mtake to this thread, as he is the first implementer of this scenario.
Do you have any comments, especially on the changes @yifanmai made as above?

[yifanmai]

OK, I have added openpyxl as an optional dependency.

[ryokawajp]

Thanks! I confirmed that it is working.