Similarity, not citations
The edges in the graph are not citation links. Two papers connect when they share references and audiences, which is why the graph surfaces work your origin paper never cited.
The similarity graph
Roughly forty papers arranged by co-citation and bibliographic coupling: node size shows citation count, color shows publication year. One glance separates the old giants from the fresh outliers.
Prior Works
The ancestors most referenced across the whole graph, listed separately. When several graph papers all lean on the same earlier work, that is your field's must-read, whether or not you had heard of it.
Derivative Works
Surveys and follow-ups that cite many papers in the graph. These are the fastest route to a current overview, because someone has already done the synthesis you were about to attempt.
What you will do in the lesson
You answer and try things, step by step.
- Pick an origin paper. You compare graphs built from a seminal paper versus a recent survey and see how the choice reshapes everything. Origin selection is the one decision that matters most.
- Read size and color. Find the biggest node and the newest node and say what each means for your reading order. The visual encoding does real work once you trust it.
- Spot the clusters. You identify the subcommunities in the graph and name what separates them. The empty space between two clusters is often where the open question lives.
- Open Prior Works. From the ancestor list, you pick the single paper you would read first and justify the pick. Prioritizing is the skill; the list just makes it possible.
- Build a second graph. You spawn a new graph from an interesting outlier node and compare the two. If they barely overlap, you have just discovered you are looking at two fields, not one.
Before you start
Why is a paper I expected missing from my graph?
Each graph keeps only the few dozen most similar papers, and similarity is computed from shared references, not direct citation. If a paper you care about is absent, build a graph from that paper instead and compare.
Where does the data come from?
Graphs are built on demand from a large academic corpus, so coverage is broad across fields. A graph reflects the literature at build time, and you can rebuild it later to pick up newer work.
Do I need to install anything for this lesson?
No. Everything happens in the embedded lesson on this page, without signup. Connected Papers itself has a free way to try it, enough to build graphs for the topics you actually care about.
How is this better than Google Scholar's related articles?
Related articles is a flat list with no structure. The graph adds what a list cannot: clusters that reveal subfields, color that shows time, and Prior and Derivative Works that bracket the literature at both ends. Structure is the product.
