I've just asked a similar question with a limited scope and purpose, not of necessarily extracting a citation graph, but of simply making it easier to fetch a paper's citing and cited papers (which I call forward and reverse citations)

https://forums.zotero.org/discussion/77659/forward-and-reverse-citations-lookups?new=1

As I mention there, I think that a basic approach could be to use services like google scholar, citeseerx, and other services which already parse and extract data pertaining to a paper's local citation graph. For my use case I think it's also important to facilitate limiting the imported papers according to research questions that I pose during active review and annotation of the paper.

Citation Star Graphs

The citation graph structure I'm dealing with is just a single central node connected to handful of "forward" and "reverse" nodes which I find interesting, and without any other structure necessarily inferred between these nodes. This is like an attention-focused microscope view of the citation graph, something that we might call a citation star. Multiple stars can be assembled into a network, but they don't have to be (they're a useful research tool by themselves).

Citation networks

What you suggest is a bit more ambitious, and could lend itself to techniques like page rank and other citation graph mining techniques, where we import a large citation network, and then look for key papers possibly in a very automated way, and focus our attention on those papers which seem relevant to the structure of that graph.

I think the problem could be done, with a couple caveats:

1. We'd have to do some complex querying and data fusion to build a larger scale network
2. We'd need to add a custom table or database for this structure
3. It might be a challenge to implement the data mining algorithms

Two Approaches for problem 1: Find one (or more) large bibliographic dataset(s) - or - build a google scholar crawler

I would be very much for this kind of turbocharged citation network analysis, and will take some time to review the data source we might use here. Ideally we might be able to find entire multi-gigabyte datasets representing large fields with many subfields, and work directly with that (intimidating as that may initially seem). Barring that, it may be that it's feasible to "crawl" the citation graph - using above mentioned services - and build a local graph around a collection of "seed" papers.

Dev Mentorship

It's important to not get stuck, and it would be great if we can find a dev that will help sponsor this project with ideas and an occasional solution. If any devs reading this feel like this is not an ill fated disaster waiting to happen, and would like to spend some of their precious time guiding this project, that would drastically increase the project's chances of success!

Conclusion

I wanna see zotero be able to do space age citation network analysis, and that might mean pulling out the stops and doing some kind heavy engineering. However, I'm not /entirely/ sure what the scope should be, and keeping it simple will mean a higher chance of success. Lets brainstorm more about objectives, data services, and algorithms/techniques that can help us meet those objectives. And lets recruit a dev that can give us their blessing and guidance!

Hi, all! I'm submitting a proposal to develop a Zotero plugin that would add citations support using data from WikiData. Please see my comment here: https://forums.zotero.org/discussion/comment/365341/#Comment_365341