How to Do Research
Coming from Korea, I was used to cramming knowledge from teachers and textbooks rather than questioning or exploring ideas on my own.
My first real experience with research was during my master's program. My advisor would give me a few keywords, and I would simply dig into them without much dynamic thinking, backtracking, or breadth-first search (BFS). I wasn't really exploring the research space; I was just following the paths laid out for me.
Later, I came across a blog post (now offline) by Prof. Collin McMillan on how to write a research paper. It completely blew my mind.
You should write an introduction even before you start your research!
Here is the template he suggested:
- A description of the high-level domain/environment
- A key problem that happens in this high-level domain
- Current solutions to that key problem
- A problem with the current solutions
- A glimpse at how you solve the problem in (4)
- An overview of your evaluation/experiments results
I still have a copy of his original blog post, but I've decided not to share it. My concern is that people might simply feed it into an AI tool and let it do the thinking for them, missing the entire point of the exercise.
If you want to become a good researcher, you need to wrestle with these questions yourself. Think deeply, question your assumptions, get stuck, backtrack, and think some more. You should feel your head hurt a little before you finally feel your neurons growing (although I'm not sure that's scientifically accurate).
Ever since I started my PhD at UMass, I've used this template as the starting point for every paper and research project I've worked on. I still use it today.
I've also used this approach to mentor students, and together, we've published two papers at EMNLP.
The biggest lesson I've learned is that research doesn't start with finding answers. It starts with learning how to ask the right questions.