Thread by @npparikh

The first paper I wrote in grad school (12 years ago) just crossed 20,000 citations. To mark this satisfying but ultimately irrelevant milestone, I thought I’d share some thoughts on how to try to increase probability of more citations, especially since grad students follow me.

These are in no particular order.

Spend a great deal of time and effort writing and organizing the paper well. This paper went through hundreds of revisions, a number of them large.

To write well, you need to read other good stuff. So read widely, *not* just for content, but also for form and style. These aren’t the same. For example, I learned a lot from Fischer Black and Don Knuth even though they work in other areas.

journals.uchicago.edu/doi/abs/10.108…

This helps train your ear.

You need to be relentlessly clear about who your audience is and keep stepping back and rereading the paper and tearing it apart if it seems confusing.

Give the paper to other people to read. Give it to other students but also send it to people you may not know, even experts in the area (if appropriate depending on the scale of the paper). This both gets good feedback and makes them aware of the work at the same time.

Give lots of talks. Invite yourself if you can, politely. If you have a famous advisor, try to get them interested in it so they talk about it a lot. Mine had a gigantic megaphone so once he was interested enough to use it for this, he went around showing everyone.

How do you get the advisor and others interested? This is a difficult thing but it boils down to taste. You must develop and exercise good taste in choosing the “right” problems that are of interest and are relevant to people. A great paper on the wrong topic goes nowhere.

How do you develop taste? Read a lot, again. Go to lots of talks, in different seminars and departments. Read great old things. Learn the intellectual and social history of the field. Talk to senior folks and ask them about this.

Don’t entirely kill the problem, and certainly don’t pretend to if you didn’t. Leave lots of open threads explicit. Then people who are interested and looking for stuff to do have many avenues to follow up. We left many such comments in this paper. (I got this from David Donoho.)

Fit the form to the content, not the other way around. This “paper” was over 100 pages long. So, you can’t publish it in any ML conference. That’s OK. You write it the way that makes sense for it, then find an appropriate venue.

Be aware that you are doing marketing, a dirty word in academia. I was convinced we *must* have “statistics” and “learning” in the title of the paper to indicate to people this is very much about ML despite basically being an optimization paper. There were many convos about this.

Anyway, I’ll stop there. I don’t know if any of this is actually useful, but I suppose the overarching point is that all these things have to be very intentional, not just your experiments and stuff like that. There are all these layers around the “work” itself.

Bonne chance. :)

(Possibly of interest to #AcademicTwitter more broadly; cc @AcademicChatter.)