How I found my first research problem

When I joined ResearchDen, I thought the hardest part of research would be the math, or the coding, or the writing. It turned out the hardest part came first: deciding what to work on at all. This is the post I wish I'd read when I started.

Starting from interest, not from a paper

My first mistake was trying to find a problem by reading the most cited paper in a field and looking for a gap. That's backwards when you're new - those gaps are usually hard, contested, or already being chased by people with far more resources. Instead, my mentor pushed me to start from something smaller: a result that surprised me, or a claim I didn't quite believe.

A good first problem is one you can actually finish - small enough to fit a semester, real enough that the answer isn't obvious.

Narrowing down

We went through a few rounds of narrowing. Each week I'd bring three candidate directions, and we'd kill two of them. The questions that survived had three things in common:

Feasible - I could run a first experiment in a week with tools I already had.
Checkable - there was a clear way to know if it worked or not.
Interesting to someone - not just me; a reviewer somewhere would care.

The experiment that decided it

The project clicked when a quick baseline did something I didn't expect. That single result became the seed of the paper. Here's roughly the snippet that started it:

for seed in range(5):
    model = train(data, seed=seed)
    print(seed, evaluate(model))
# the variance across seeds was the whole story

That variance - which I'd assumed was noise - turned out to be the actual finding.

What I'd tell my past self

Pick something small and finishable. Talk to a mentor early and often. And treat your first surprising result as a gift, not a bug. If you're just starting out in research, that's exactly what this community is for.

Want to share your own experience? Every member can write here - reach out and we'll help you publish your first post.