Why we built the knowledge graph

Somewhere around the hundredth paper, most researchers stop being able to hold their reading in their head. You remember that a paper argued something relevant, that you wrote a note about it, and that a second paper seemed to disagree. What you can't easily tell is where each of those lives, how they relate, or which parts of your reading you never connected to anything at all.

The knowledge graph is our answer to that. It's a map of how the things in your library relate to each other, drawn from what you've already read and written. This post is about why we thought it was worth building, and what it's for.

The problem: connections live in your head

A reference manager knows your papers. A notes app knows your notes. Neither knows that the third paragraph of one note is really about page 214 of a particular paper, or that two of your notes are circling the same idea from different directions.

So the connections that make up a literature review, the "this builds on that" and "this is the evidence for that claim", stay in your memory. Memory is a poor place to keep them. It doesn't scale to a thesis-sized reading list, and it doesn't survive a three-month break.

Many of us used a notes app's graph view for exactly this reason. It was the closest thing to seeing how your thinking fits together. But it only sees the notes. It has no idea a note is really anchored to a passage in a PDF, or that the paper behind it cites five others in your library. We wanted the version where reading, notes and sources are all in one picture.

What it shows you

Open the graph from any paper or note and you get a map centred on that item, showing what's connected to it within a couple of steps: the papers it cites, the notes that link to it, the claims that cite it as evidence. A Levels control at the top widens or narrows the view, from one step out to three; two is the default.

Expand

Different kinds of things look different, so you can read the map at a glance. Papers, concept notes, literature notes, highlight notes, other notes, claims, authors and publications each have their own shape and colour. A small Key and filters button in the corner opens the legend, and it remembers whether you left it open.

Size tells you how connected something is: the bigger the node, the more connections it has. The line between two nodes tells you what kind of relationship it is. A note link you made yourself is a solid, coloured line, with the kind of link written on it. A suggested link, one the app has found for you but you haven't confirmed, is dashed. A citation between two papers in your library is a thinner grey line with an arrow pointing at the cited paper. Claims hang off their note on a dotted hairline. Each row in the key is also a switch: click Note links, Suggested links or Citations to hide that kind of connection, or click a node type to hide those items, and the map redraws. If you only want to see your own thinking, hide everything but your note links.

Why that's useful in practice

Seeing what a paper actually connects to. Before you write about a paper, open its map. Which of your notes already mention it? What does it cite that you've also read? Which claim in your notes rests on it? That's a much better starting point than a blank page and a vague memory.

Finding the gaps. The most useful thing on the map is often what's missing. An Unconnected panel on the map lists the orphans: papers with no links or citations, and notes with no paper link. Those are the things you read but never made use of, and the notes that float free of their sources. A thesis is easier to defend when you know where those loose ends are.

Moving through your reading by relationship. Folders force every paper into one place. A map doesn't. Click a node to re-centre the map on it, or double-click it to open the paper or note. The panel beside the map lists everything connected to the centred item, grouped by type, with Centre and Open buttons on each row, so you can follow a thread of an argument across your whole library instead of through a folder tree.

Checking a claim against its evidence. If you build claims on your concept notes, the graph shows which papers support each one. Thin evidence is visible as a claim with few lines coming into it.

What it doesn't do

The graph is built from your own library: your papers, your notes, your links, and the citations between the papers you've saved. Opening it doesn't use AI credits, and it doesn't send your library to anyone. It also isn't a verdict. The map shows connections, and it's still your job to decide what they mean.

The graph from a paper or note is the focused one. The Graph entry in the sidebar gives you the same centred map for whatever paper or note you have open, and falls back to a whole-library overview when nothing is open. For a very large library, that overview starts with just the main hubs and lets you expand outwards, with a Show full view button if you do want every node drawn. That keeps it usable instead of turning into a hairball. A search box beside the overview helps you pick an item to start from, and Explore connections opens the centred map for it.

Doing this in ScholarThread

Open any paper (in its details or in the reader) or any note, and use the graph icon in the toolbar; hover it and it reads Knowledge graph. The map opens in a workspace tab next to whatever you're reading, so you can keep the paper and its connections side by side. Click through to re-centre on a paper or note, hide the connection types you don't want, and check the Unconnected panel for anything you've read but never used.

If you're new to how notes and claims fit together, your note templates are an ontology explains the structure the graph draws on, and the literature review note-taking workflow shows how the notes get written in the first place. The graph is only as rich as the links you and your sources give it, and the more of your reading you connect, the more it has to show you.