Sunday, January 23, 2011

How to title a paper

A colleague of mine once commented that the single most important factor contributing to whether or not a paper gets cited is its title. I recalled this comment this morning -- which is a Sunday morning when I had some interesting reading planned. Instead, I find myself brooding over possible titles of a conference submission -- exchanging mails full of "mights" and question marks with the first author.

Title is the paper's visiting card The paper title is a piece of metadata -- it will allow the paper to be found: discovered and also re-found. As such, titling is essentially choosing indexing features for your work. The title should tell the reader what the paper is about -- support the decision of whether or not to read the paper. But, as every good indexing feature, it needs to be not only representative, but also discriminative. The title must also help people to see the difference between your paper and other papers on a closely related, but different subject.

Predicting the future I'm fond of asserting (given the appropriate context) that indexing essentially involves predicting the future. When assigning indexing features, we try to guess what people will need to find in the future and choose features that will support their searches. When you write a paper, you consider not only the work at hand, but also the other work that you might in the future write about the subject. Select the title so that it will be able to flag the difference.

This process of forward-looking is very scary indeed. I find myself asking myself the question: If I turn out to be totally wrong about what I have shown in the paper, from which direction will that wrongness come from? How will I correct (or better, improve on -- or best yet: completely blow out of the water) my past methods? Contemplating potential wrongness or non-perfection is opening the door to thinking of something further -- something that really I should of thought of already and included in this paper. The process threatens to bring down the entire effort onto our heads -- and just at the moment when our every fiber is straining to the point in time that we can click on the submit button.

Pragmatic approach We can take another tack. What is the most immediate purpose which our title will serve? It gets put in the conference management system and is used by the reviewers to help pick the papers that they bid on and ultimately review. We want a title that will get our paper to the right reviewers. The right reviewers are the ones that know the topic area of our paper well -- especially the background literature. These reviewers will recognize immediately that we are citing the key contributions that have come before us and will also be able to point out anything that we missed.

The goal of getting to the right reviewers suggests that creating a descriptive, specific title is the way to go. But there is also something to be said for having an interesting title -- one that conveys that this is not another paper on xyz but a particular fine specimen of the genre. An interesting title signals to the reviewer that their reading effort will be richly rewarded.

Also, I worry about the title being straightforward. Many reviewers have a highly functional, but not particularly broad, mastery of English. As such, they are very competent in their areas, but are hard pressed to tell the difference between language use inconsistent with established grammatical convention and language use which simply diverges a bit from the formulations generally used in computer science research papers.

Titles and the art of getting cited If we create titles to be little advertisements for our papers, it's no wonder that my colleague sees a connection between a good title and a lot of citations. It can only help, if you also craft your title to ring with authority. It should be specific, but not so specific that it can be cited in only one particular context. People checking the citations should be able to look at the paper title and think, "Hmmm, yes, that's a plausible citation at this juncture." Of course, in an ideal world they would call up the paper and read it through in order to judge its appropriateness -- but in this day and age of information deluge that is nearly too much to ask.

The title finding process So, how to get to a title that fits the bill? Here, my main message is try to "Predict the future." However, there's descriptions of the process floating around out there which are perhaps a little more user friendly. One of the ones at WikiHow (that caught my fancy after a couple of searches) lists step 1 as "Get smart and think" http://www.wikihow.com/Find-a-Catchy-Title-for-Your-Paper/Essay It goes on with a reassuring "You can do it, you just don't know it." Right. Nothing there about brooding and write a blog entry -- better get back at it.

Coda If you catch me in a particular frame of mind, I will rail against the system that uses citation counts as the (exclusive) measure of academic productivity. But that rant is actually not all that much fun in the environment where I work, since the people around me generally accept that citation counts, although themselves quantitative, are only as good as the qualitative decisions that underlie them. For example, the decision on how to fairly account for cultural differences between fields that communicate via mainly via conferences vs. fields that communicate via journals. Or, the difference between quickly and slowly moving fields.

The real zinger remains this: we work in the field of information retrieval -- driven forward by our curiosity of all things search related. Isn't it the sign of a good IR researcher, that if you tell him that his output will be measured by Google scholar's calculation of citation count that the first thing that he will do is to set about developing a method to game the system?

I write papers to get my ideas read by the right readers -- people who will act upon what I have discovered. How many people this is and whether they are actually writing papers (and remember to cite the work they of myself and my colleagues that they have found influential) rather than developing software, systems and policy is of secondary concern. I tend to measure by the number of inquiries I get for more information about my work, e.g., along the lines of, "We are re-implementing your algorithm, do you mind giving us more information on aspect x."

But in the end, whether you are going for the cold hard cites or the multiplicity of other less quantifiable ways in which you know that your work is making a difference, the title of your paper is important. OK. Now I am back at it.