Scite: Smart Citations for Better Research
scite.ai
scite.ai
If you're doing science and you don't know the literature, then you're doing a bad job.
Understanding a paper won't work by reading AI summaries. Finding surprising new papers in your field should be very rare even without tech.
[edit] Let me re-phrase the very negative statement above. There are a couple of important tasks in working with literature.
One is finding cumulative truth, i.e. figuring out the overall state of research. This should be done in a true meta-analysis of the data, since references are sometimes citing articles incorrectly (i.e. claiming something that's not in the paper) and the linguistic complexity is enormous. Meta-analyes are a lot of manual work, even though you can extract some data from standardized tables.
Another one is figuring out what new publications there are on a specific subject. Search engines like google scholar are immensely useful here, because you can follow the reference tree in both directions (citing and cited). This tool would help, but perhaps incrementally so (not a panacea).
A third task would be to identify novel interesting papers that you didn't know about, perhaps from a different field. This tool might help, but only if an unknown interesting paper cites something you're looking at (i.e. if the networks are not disjoint).
Apart from that, offering structural analyses and visualizations of science is something I believe should be done with open source and open data, because science is about trust and transparency - now more than ever.
Thanks for the thoughtful response.
Disclaimer: I work at scite.
I want to clarify one point in your comment because I know a lot of people read HN comments before clicking through: we do NOT offer AI summaries!
scite is fundamentally built on the idea that not all citations are equal. We offer disambiguation into citations by extracting the in-text contexts when citations happen between two papers. With that, you can qualitatively see how a paper has been cited by others by seeing the citation contexts from each citing paper.
As someone who used to be in research before at the National Cancer Institute, I certainly agree that scientists who are advanced in their field should know the literature inside out.
But the utility of scite is I think quite a bit more than that:
- If you're an expert scientist, you can configure alerts to know when new citations happen to papers that you care about (e.g. ones you published, or that are important in your field, etc). Except unlike Google Scholar alerts and such, you can now immediately see the citation contexts (as opposed to opening up each paper, and trying to locate the reference(s) and the in-text citations).
- If you're a student or an experienced researcher trying to ramp up in a new field (which is increasingly important given how interdisciplinary science is these days), this is a useful tool in your arsenal to digest and discover new papers.
- That last comment also applies if you are in the industry! e.g. someone in pharma doing drug discovery or evaluation and you're working on different drugs every <X> months.
We do have a preprint we published if you'd like to read more about how scite works: https://www.biorxiv.org/content/10.1101/2021.03.15.435418v1
I hope that was clear. I'm happy to answer any other questions. I'll try to come back and check periodically.
- Ashish
Well, that's a tricky question to answer! :)
At the moment we use DOIs to link papers / citations, and get our information on DOIs from crossref (https://www.crossref.org/).
So every DOI there is also in our system (and any citation coming from a source without a DOI currently won't be picked up).
From crossref we also have metadata on which publications a given paper references, so we can always show information on how many publications cite a given paper ("traditional citations").
As for our database of Smart Citations (the extracted in-text citation statements) -- the answer is that it depends on our ability to access full-text articles to actually do the citation statement extraction. We're constantly growing in this area through new indexing agreements with publishers and so on.
To your specific question about that DOI, here is the corresponding scite report for that paper: https://scite.ai/reports/a-unified-framework-for-coupling-dZ...
I see that we identified 371 distinct citing publications (and extracted 374 in-text citation statements from those publications).
Can you let me know where you saw it say "no citations"? That sounds like a bug... I'll take a look. You can email me here: ashish@scite.ai
- Ashish
When I have the results from an article in my screen and use the top search to find another, it only updates the upper part of the page with the paper title, not the stats and summary below it
I had searched for an article before that had no citations. That's why I continued to see "no citations" when I searched for the DOI I shared...
Thanks for that description, I managed to reproduce this and it is our fault not yours. You are using it correctly.
We'll get it fixed ASAP!
EDIT: Fixed now :)
Weird, I searched for that DOI moments before and it showed this result: "This publication either has no citations yet, or we are still processing them"
Let me look into others
I agree that those use cases are definitely valid. And I would also say that research is a technologically underdeveloped space. So any innovation here is great and will definitely help people!
The only worry I have (from past experience) is that easy tools have a differential impact: That they facilitate bad research more than they facilitate sound research. My perception as a scientist is that we primarily need to address actively harmful science (as in: spurious results, p-hacking). Fostering more science, on the other hand, is definitely a second or even third priority.
Along with our extraction of citation statements, we have a deep learning model that classifies each statement as "supporting", "contrasting", or "mentioning" of the claims made in the original paper.
Indeed there is a lot of nuance here, and we've tried to address this in a number of ways (including a manual flag process for review where someone believes it is incorrectly labeled), and we're constantly improving in this area. In general we would expect that research that is not sound will receive more contrasting citations rather than supporting citations because their findings won't be as well supported by future research.
I should say explicitly that a "contrasting" citation is not a bad thing per se, especially given the nature of scientific discovery. But a study that was published through p-hacking whose findings are not reproducible should be easier to discern through subsequent contrasting citations. In today's world, all we might know is that it was cited a bunch, with the nuance buried underneath.
- Ashish
Your statement on ‘finding cumulative truth, i.e. figuring out the overall state of research’ is a key point.To become newly acquainted with a research area (on one's own) is a daunting task, requires plowing through dozens and dozens of past research articles, review summaries, abstracts, texts, even conversing with authors of these works. But rapid progress is oft realized when people in two or more separate areas serendipitously interact in new projects, thus bringing expertise — heretofore missing or weak — to bear on a common problem. This is the intrinsically collaborative aspect of science that survives despite intense intramural competition.
We, at scite, actually designed scite to help folks see beyond the "one big claim per paper" model. Specifically, we designed scite to present show how specific results were either supported or contrasted by other papers, rather than showing that the publication itself was disputed or reproduced as a whole. If there was any feedback you could give on how we could present things so that folks don't get that impression we'd love it - feel free to DM me or let us know at hi@scite.ai as you test it out.
In an ideal world we would present the claims, results, and argumentation as structured information to the user but we are not quite at that stage yet!