weishuhn

This looks super interesting! I'll take a look. Thanks.

Good feedback. Visualizations are definitely the most requested feature and something I'm working on. It's a hard problem to distill so many papers (5-150k) down into a usable graph. I've made big ones that look amazing but which are not that useful.

I like the idea of interactive filters to allow the person to explore it visually. I'm hoping to have something people can play with in the next few weeks. I hadn't really expected the site to take off the way it has and so it's not really feature complete yet :)

For that I use the Unpaywall API:

https://api.unpaywall.org/v2/10.1097/RHU.0000000000000563?em...

According to them it's closed but I definitely see it as open here:

https://journals.lww.com/jclinrheum/fulltext/2017/09000/omeg...

I'll look to see if I can find another data source for OA papers to add more coverage.

That is the most requested feature and something I'm working on. It's a fun (and hard) design/data problem. Which of the 5k-150k papers do you show in the graph? And then how do you render them in a way that is both visually appealing but also conveys the most import information?

If you have the chance, try the following:

1. Find a paper you like in a field you want to learn.

2. Use the keyword filters to filter down to papers that match your criteria.

3. Add a bunch of the interesting ones to a new graph using the purple "+" buttons.

4. On the "new" graph page, check out the similar papers section. If any of them are interesting, add those to the graph.

5. Repeat until you don't find anything else that is interesting.

The similar papers section uses a link prediction algorithm that basically says, if two papers cite a bunch of the same papers, rank them higher BUT if the paper they cite, is cited a bunch of times, don't give that connection much weight. The net effect of this is that it doesn't really matter if the paper was highly cited, only that it cites the same niche of papers as the ones you just chose. Also, because of the temporal nature of academic literature, the papers it brings up tend to be the newer and harder to discover papers.

The results are pretty great and it's as close to the "serendipity" slider that you'll get right now.

EDIT: Formatting

Eventually I'd like to move the site to open source, but right now the repo isn't in a place where I can do that. As for specific parts, it's pretty purpose built and this is my first Rust project and so I'm not sure which parts would be helpful to the community. And I doubt they would meet the communities standards just yet :)

I go through a lot of the details in my post on the Rust subreddit: https://www.reddit.com/r/rust/comments/kfiaqn/just_wanted_to...

But long story short, I end up doing most of the graph analysis by passing in the citations, using PyO3, to graph-tool in python then returning the data I need about each paper. I am planning on moving that over to Rust. But not being an academic I wanted to get feedback on the quality of the results before making it difficult to quickly test different types of algorithms.

I am planning on making the info in the BibTex files a bit more robust. Right now I'm just adding what I have readily available. But in general, if you are using Zotero or Mendely, the functionality to enrich the metadata on the entries does a good job filling in the missing info.

Thanks for the feedback. #1 might be some sort of bug. #2 Good call!

Much appreciated! I'll update it. I really hadn't anticipated much traffic. I was planning on doing an open beta but things have kind of taken on a life of their own (in a good way).

Creator here. It's similar in that it uses citations to make paper recommendations. But different in that I give you access to the entire paper graph rather try to distill it down to just a few. You can even write your own queries by clicking on the "SQL" button at the bottom of each table. I kind of view it as a Connected Papers for power users.

The database is about a month out of date right now. I am going to be updating it soon. You should be able to either put in the url or do arxiv:XXXX.XXXXX

Creator here! Wasn't expecting this to happen :) The site is definitely still in Beta so I appreciate any and all feedback. I just launched it a few days ago. It's been my COVID project and I finally got to the point where I felt comfortable having others use it.

The biggest hurdle was the speed of the graph creation. Basically taking a 250,000,000 paper/2,500,000 citaiton db and creating graphs that could be up to 200k papers and 3-4mm citations. For that I ended up learning/using Rust (which was a great experience).

The plan is to keep it totally free and hopefully get some institutional support once I get a better handle on demand and costs.

Ask me anything!

EDIT: As you are going through the site, be sure to use the purple "+" buttons to create your own graphs centered on the topic of your choice. That combined with the in-graph keyword filters are probably the most powerful ways to quickly zero in on the most relevant literature.