Here is my method to conduct (and automate) a literature review. Using Google Scholar,
@scite ,
@zotero and
@obsdmd. Tutorial with examples and best practices:
We start with a search. But: Without domain knowledge, it’s hard to tell what is relevant. We might collect too many. Instead: Find SEED papers and explore their relevant references/citations. → Fewer but more relevant papers. (I will show a tool for this)
Google Scholar is google but for scientific papers. But: – the domains are mixed up. – it ranks mostly by # of citations → older papers prioritized Use it to find SEED papers, not for an exhaustive search. Its strength:
@zotero and
@scite integration
Zotero is a BRIDGE between different tools in my workflow. To collect papers with a single click, install the Zotero connector (link in last tweet). Click its icon in the browser and select papers into a new collection. I don’t use it for annotations (why, in a second).
To identify RELEVANT papers, I use the
@scite browser extension (link in last tweet). New ideas aren’t accepted right away. The more supports/contrasts, the more novel/relevant the idea in the paper. Scite will add a box under each paper in Google Scholar telling you this!
This paper states: “Bio diversity improves eco system services”. But: 1. Biodiversity drivers are debated for ~70 years. 2. We don’t fully understand services ecosystems provide to us. → This idea is speculative & novel. We should read it and scite helped us discover this.
We now have a few SEED papers in Zotero. Let’s explore their references and citations with
. “Export Collection” from Zotero and import into Litmaps. Litmaps looks at the entire collection not single papers! Here is what you can learn:
Litmaps can visualize citations vs age. Older papers usually have more citations. You will notice a diagonal trend. → Impactful papers (recent + highly cited) sit far off the diagonal! We can SEE what is relevant. Continue to grow you collection this way.
To get the most “value for my time”, I focus on reading the: – 10% most cited – 10% most recent – 10% reviews The rest can be archived or postponed for now. Here is how you can do it in Litmaps:
80% of value is in 20% of the contents. By looking for relevance in every step we MINIMIZED our TIME. To MAXIMIZE INSIGHT we need to take good notes. I take notes with
instead of Zotero. You can pull your papers from Zotero to Obsidian:
To take GOOD notes, I don’t summarize. Instead I break it up into concepts and link those together. A summary is just paraphrasing. But linking concepts generates new ideas. Our brains are a network not a database. You can do this with Obsidian but not with Zotero.
Summary: 1. Find seed papers with Google Scholar 2. Filter with
@scite , add to
@zotero 3. Add collection to
@LitmapsApp and grow it 4. Prioritize top 10% by age, citations & reviews 5. Take notes in
@obsdmd (link to Zotero) (6. Ideally use a typewriter to write manuscript!)
Google Scholar: https://scholar.google.com
Zotero Extension: https://zotero.org/download/
Scite Extension: https://scite.ai/extension-install…
Litmaps: https://litmaps.com
Obsidian: https://obsidian.md
My Newsletter:
https://ilyashabanov.substack.com/
Processes like this are best shown. I have put together a 2-hr workshop around this flow: https://effortlesslitreview.carrd.co Priced resonably (I hope) at 15$. If not, follow
@artifexx as I deconstruct single steps of this process for more clarity here on Twitter in the next days.
https://effortlesslitreview.carrd.co/
One small update: For the online workshop the CEO of
@LitmapsApp is going to join us. Let’s chat with him live on March 18th. Really excited about this!
Some folks mentioned that ArxivGPT (in step 2) is not mentioned here. As it is not a 100% essential I did not mention it initially. But since you asked – here is how I use ChatGPT to screen papers by relevance:
