Citation counts lie a little
A raw count treats a throwaway mention and a foundational dependence as equal. Semantic Scholar's citation signals exist to pull those apart.
TLDR summaries
Model-written single sentences that state what each paper found, right in the results list. They are triage tools: perfect for deciding what to open, dangerous as a substitute for reading.
Highly Influential Citations
Citations where the citing paper genuinely built on the work, using its methods or results, rather than waving at it. Filtering by these separates load-bearing papers from merely famous ones.
Research Feeds
You rate a few papers as relevant or not, and the feed learns what your topic actually is. New matching work then arrives on its own, which beats re-running the same search every month.
What you will do in the lesson
You answer and try things, step by step.
- Triage by TLDR. You rank ten search results using only their TLDRs, then check your picks against the abstracts. The exercise shows both how fast TLDRs are and where they cut corners.
- Split a citation count. Open a paper's citations and sort them by intent: background, methods, results. The same number tells three different stories once you split it.
- Find what actually built on it. You filter to Highly Influential Citations and see how short the real list is. That short list is your reading queue.
- Train a feed. Create a Research Feed from a paper you like and rate a handful of recommendations. Two minutes of ratings buys you months of automatic discovery.
- Pressure-test a TLDR. You hunt for a TLDR that oversimplifies its paper and write down what it dropped. Once you have caught one, you will never fully outsource judgment to a summary again.
Before you start
Is Semantic Scholar really free?
Yes. It is run by Ai2, a nonprofit research institute, and both the site and its API are free to use. There is no premium tier hiding the good features.
How are TLDRs generated?
A language model writes them from the paper's abstract, introduction, and conclusion. They are usually faithful but compress hard, so caveats and conditions get dropped, which is why the lesson treats them as triage rather than truth.
Do I need to install anything for this lesson?
No. The lesson runs right here in the page, free and without signup, and Semantic Scholar itself needs no account for searching either.
Should I use this instead of Google Scholar?
Use both, for different jobs. Scholar tends to have the widest raw coverage; Semantic Scholar adds the judgment layer: TLDRs, citation intents, influence filtering, and feeds. A common pattern is Scholar for recall, Semantic Scholar for deciding what deserves your time.
