Lack of documentation and quality code

[RoosBakker]

There are not many document relation extraction datasets, so thank you for introducing it.

However, I have several issues with the dataset, the code, and the documentation. I'll start with the documentation.

1. The README.md file does not describe the repository, but is more an abstract of the paper.
2. The README.md file for the code is too short, An overview of the code files and what they are for would be very useful.
3. The README.md file for the data contains the data format, which is useful. However, it does not explain the different elements of the data well enough.

For getting a basic understanding of the data and the code, I had to search in the paper and write additional scripts. This could be prevented by extending the readme files.

Regarding the data, it would be helpful if information is provided about the meaning of the different files. Especially more context about the relation info, and positions of the head and tail would be very useful, figuring this out is not trivial without diving deep.

Finally, the code is unreadable for an outsider. Variable names are often random letters, and there is literally no
usage of functions. In very few cases a comment is written. This results in long unreadable slabs of code with nested for loops galore.

The above points mean that reproducing the results is hard, making changes to the models and methods is harder, and extending the dataset is just not worth the hours. I'd love use this datasets and models to extract relations from a document about a different domain, but with the above points that takes too much time.

If the desire is there to keep this dataset useful and of value to the research community, I strongly recommend improving the documentation and quality of the code.