I felt that same overwhelm the first time I used it, especially with a broad query like yours. Starting with a known, specific paper as your seed is good advice I've seen here.
I'm curious, did you try adjusting the depth parameter to 1 instead of 2? That helped me cut the initial noise by a lot, and then I could expand specific nodes manually from the UI. It felt less like drinking from a firehose.
On the export, I'm wondering the same thing. If the API response isn't a standard graph format, writing a transformer script seems like extra work for something that should be basic. Has anyone gotten a straight JSON export from the UI yet, or is it all API-only?
One step at a time
>did you try adjusting the depth parameter to 1 instead of 2?
That's a great practical tip, and it's the workflow I ended up settling on too. Starting with depth 1 gives you a manageable set of core papers to evaluate. The key is resisting the urge to go wider until you've manually confirmed a few of those core nodes are actually relevant. It turns the tool from an automatic map generator into a guided exploration tool.
On the export, I haven't seen a UI export option yet. My workaround was to call the API from a simple Zapier HTTP request step and pipe that raw JSON into a Google Sheet to clean it up. It's not ideal, but it worked for a one-off. If you're doing this regularly, a two-line Python script to flatten the JSON is probably less effort than clicking around a messy UI export, if one even exists.
Automate everything.