Add (mostly failing) test cases for ReferenceParser

[fzieris]

Author and title extraction don't work super-well right now.
This PR adds a small number of failing test cases and provides a fix for one simple case
(needless removal of trailing dots in abbreviated author names).

[fzieris]

@lidianycs Here are some of the issues I noted (run the tests to see concrete examples):

• For some (not all!) references with 3+ authors, the second author is missing in the output.
• For some references other authors go missing as well.
• For some references with "and" in the author list, the "and" becomes part of the output.
• For some references, the title will just be "and"

[lidianycs]

@fzieris many thanks for the update! We're definitely needing test cases. I'll look into it to release a new version.
I'm also looking for options to replace cermine as the default parser. I appreciate any suggestions!

[lidianycs]

Hi @fzieris, thanks again for the update! I'd appreciate it if you could keep working on it. Sorry for the delay in merging the PR, but I was only able to analyze it today. This week I'll be working on other updates in the tool, so I hope to be able to give a fast answer this time. 🙏🏾

[fzieris]

I'd appreciate it if you could keep working on it.

I did already. I've been playing around with GROBID as an alternative reference parser that people could run a Docker container and that Cerca calls via its REST API. The quality for both authors and titles is much higher (i.e., all test cases pass).

I'll create a PR for this soon.

[lidianycs]

@fzieris I really appreciate it! I've looked at GROBID too, but unfortunately the need to run a Docker container would be a downer in this scenario because it'd difficult the use by non Tech savvy users.

I'm considering creating a PDF parser and using an updated ML model to find the references but I'd prefer to keep it abstracted for the users.

[lidianycs]

Hi @fzieris I was looking into it again today and unfortunately what I was trying didn't workout. I read GROBID docs again to see if there's a way of using it without hassling the users but didn't find a way. Let me know if you think of something. Thanks!

[fzieris]

@lidianycs Check out my draft PR: #11

The PR follows the rules "make the change easy, then make the easy change":
The first few commits are for restructuring, the last one ("WIP") then uses GROBID, assuming it runs locally on port 8070 (e.g., with docker run --rm -p 8070:8070 grobid/grobid:0.8.2-full).

Feel free to cherry-pick bits and pieces from that PR.