News for hackers. Stuff that matters.
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by londons_explore on Sunday September 05, 2021 @02:32AM (#28422035)
I really wish someone could build a better UI for this research internet.

Hyperlinks for all references would be a good start. Finding some way to make some automatic glossary of definitions of technical terms would make scientific papers substantially more accessible too.

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by sleepingsoul on Sunday September 05, 2021 @09:40AM (#28424276)
Have you come across scite (https://scite.ai) yet? We're also innovating in this space by extracting citation statements from full-text articles and classifying their intent.

So let's say paper A cites paper B. If you look at paper B, we show you:

- how many times it was cited

- the direct paragraphs from paper A where it was cited

- the sections from paper A where paper B was referenced

- ... and a lot more

You can also now search these citation statements directly to find evidence-based information pretty quickly.

- Short video to showcase that search: https://www.youtube.com/watch?v=JYjCn-4uMJk

- Website with a bit more details: https://citation.to/

You can also visualize citation networks similar to ConnectedPapers, set notifications for new citations on groups of papers you're interested in, and much more.

(Disclaimer -- I work here!)

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by KNrDajfZ on Sunday September 05, 2021 @10:12AM (#28424478)
No one cares. This thread is about free access to papers and not another paid service that forces you to pay monthly fees for something that could be a free service. In that sense you aren't any better than large online publishers. 8 bucks a month for a scientific paper search engine? Really?
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by ComputerGuru on Sunday September 05, 2021 @10:49AM (#28424738)
I vouched for your comment because you have a very valid point, but you could be more polite in making it. Welcome to HN.
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by KNrDajfZ on Sunday September 05, 2021 @10:58AM (#28424809)
Thanks. I am really cranky today. I'll do better next time.
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by sleepingsoul on Sunday September 05, 2021 @02:46PM (#28426658)
Hiya,

Well, I definitely agree with your sentiment in a normative sense that scientific papers should be free and readily accessible to all -- in part because a lot of it is funded through tax dollars!

But given the current state of affairs, we're looking at making that information accessible to people without having to pay exorbitant fees to access individual research. We also offer steep discounts for students or anyone in academia.

With that in mind I would push back a little that we're just a scientific paper search engine -- our system does a lot of work in extracting and classifying those citation statements, which makes it more powerful than traditional scientific search engines.

And besides just using our search, a huge time-saving value of our service is the report pages which helps you quickly build a qualitative understanding of how something was cited.

Even if all scientific papers were freely accessible, our report pages allow you to see the direct, relevant snippets from citing papers without having to manually read each and every single one. I think that is quite valuable!

I know I've gone on a little tangent from the original discussion about scihub, and having free and open access to papers, but I did just want to throw that in because I think it's an important distinction. And as much as we all want that free and open world to exist, I think it's also interesting to think about how we can open up that information for people in the interim.

Best,

Ashish

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by abcc8 on Sunday September 05, 2021 @05:32PM (#28428011)
The site sounds interesting, but signing up just to evaluate the results of my first query is a no-go for me.
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by geokon on Sunday September 05, 2021 @04:50AM (#28422595)
One thing I've recently discovered is this website: connectedpapers.com/

It builds a graph of referenced papers and makes it easier to narrow down which one are important/foundational for further research

I don't think a glossary would help. What you need to find is a "review paper". These act as a primer to the field for new researchers. They're usually well written, with less jargon and have tons of references for you to dig into. That said, I don't have a good method for finding them.. I just stumble across them haphazardly..

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by stevesimmons on Sunday September 05, 2021 @05:29AM (#28422758)
https://www.semanticscholar.org/ is very good too for reading paper abstracts, links to the references, citations and related papers. The full papers are included, if copyright allows.

The references and citations are tagged and filterable, making it easy to see which are the most cited, or review papers, etc.

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by elcritch on Sunday September 05, 2021 @02:54AM (#28422118)
DOI URIs work great and sci-hub understands them. Figuring out the DOI from the citations section is a bit more annoying still. A meta glossary would be fantastic. 80% of learning a new field is figuring out the jargon.
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by goodmachine on Sunday September 05, 2021 @05:25AM (#28422743)
> A meta glossary would be fantastic. 80% of learning a new field is figuring out the jargon.

True!

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by IshKebab on Sunday September 05, 2021 @03:25AM (#28422242)
Yeah it's called Web of Science. It's very good but it's not free unfortunately. And it doesn't go as far as hyperlinking references in PDFs or defining terms. I agree those would be great, but unfortunately there's not much incentive for paper authors to do that.

https://en.wikipedia.org/wiki/Web_of_Science

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by Al-Khwarizmi on Sunday September 05, 2021 @05:36AM (#28422793)
Hyperlinking references in PDFs is trivial for authors if they use LaTeX and the journal/conference template supports it. It's just a matter of ensuring that the bibtex entry has an URL or a DOI, and most bibtex entries copied and pasted from curated sources already have them.

If you are finding many papers without hyperlinked references, it's probably just because they're published in journals whose templates don't support it. In my particular research field, most publication venues' templates starting supporting those links around 3-4 years ago, so my papers from, say, 2015 have no hyperlinks in references, while those from 2019 do. This didn't require any significant extra effort on my part, in fact in general it requires less because well-curated bibtex entries are easier to come by now than some years ago.

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by IshKebab on Sunday September 05, 2021 @09:14AM (#28424106)
It's not adding the links that is hard. It's choosing the destination URL. Where do you link to? I guess you could link to doi.org. Probably better than nothing but still not ideal because it doesn't actually take you to the PDF.

Can you show an example paper with links?

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by foobarbecue on Sunday September 05, 2021 @10:05AM (#28424431)
DOI is the correct thing to link to because the author or publisher has chosen this as their canonical URL for the object. The DOI could and sometimes does point to the pdf, it's just conventional to point to an html version of the paper. It would make a lot of sense to me to standardize a field in the DOI metadata containing the PDF URL. (source: I manage the DataCite membership of a large organization)
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by jltsiren on Sunday September 05, 2021 @11:41AM (#28425148)
Linking directly to the PDF is usually the wrong choice. When you find a new paper, you often want to get the citation metadata, which the PDF document rarely contains in a convenient form. There are often multiple versions of the same paper, and you may want to determine which version you managed to find. Is it a preprint, the final authors' version, the published journal paper, an early version published in conference proceedings, or an unpublished extended version of the paper?
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by ZeroGravitas on Sunday September 05, 2021 @10:13AM (#28424495)
Some of the background data for this is being collected by wikiCite (https://meta.wikimedia.org/wiki/WikiCite) and Scholia https://scholia.toolforge.org/ projects

Part of this involves turning text data of authors into linked data that used to navigate between texts: https://author-disambiguator.toolforge.org/

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by beckman466 on Sunday September 05, 2021 @06:29AM (#28423080)
Have you heard of Alexandra Freeman's Octopus? https://www.science.org/careers/2018/11/meet-octopus-new-vis...

Edit for direct link to Octopus: https://science-octopus.org

Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by generalizations on Sunday September 05, 2021 @11:25AM (#28425020)
Seems like a large part of what's needed is just being able to make the pdfs machine-readable, by making decent plain-text versions of the text content. Right now, IIRC, there's no hands-off way to get the text of a pdf. Especially if there's weirdness like multiple-columns (sometimes happens with this stuff).
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by zapataband1 on Sunday September 05, 2021 @11:45AM (#28425183)
Like you said the hard parts are the unstructured data/images/tables. There are pretty-good(80% of the way there) solutions tho. But nothing that could handle millions of paper without error
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by zapataband1 on Sunday September 05, 2021 @11:44AM (#28425176)
Currently working in this field and this is actually the cutting-edge(!!) but it will be 100% possible/robust within the next year or so I believe. Really cool ML techniques being used for htis.
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by sleepingsoul on Sunday September 05, 2021 @02:48PM (#28426677)
How well does it work with old OCR'd PDFs? :)
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by generalizations on Sunday September 05, 2021 @10:33PM (#28429681)
I had no idea. Is there anything that's open source, or is it all still being kept proprietary?
Re:Today Sci-Hub is 10 years old. I'll publish 2M new articl... by jpeloquin on Sunday September 05, 2021 @03:56PM (#28427240)
It wouldn't take much to significantly improve things. We don't even have full-text search for paywalled articles. The paid search engines like Web of Science just do title, keywords, and abstract. Even considering the subset of open access articles, Google Scholar and Semantic Scholar do ok but don't offer much in the way of search refinement (e.g., "DTAF" NEAR "collagen"). They're good at finding _something_ related to your query but not good for systematic review.