Although it isn’t on our blogroll, I sometimes read the FemaleScienceProfessor blog and it had an interesting post (and comments) about keeping up with the scientific literature, e.g. by browsing journal issues when they come out. Although I have argued that there is a Google effect among researchers, it seems a good number of people still do browse rather than search and in fact enjoy it.
I’ve recently been thinking about a problem that arises in the crowdsourcing of image processing and perhaps because of that was flipping through the IEEE Transactions on Image Processing and came across the paper “This is SPIRAL-TAP: Sparse Poisson Intensity Reconstruction ALgorithms—Theory and Practice” which has interesting content but is perhaps more notable for its title. I seem to remember that a signal processing class problem that your former undergrad advisor (who also happens to be the TPC chair for ICIP2012) once wrote referenced the same cult classic.
Now that I have been thinking about scientometrics a little bit, it does not surprise me that the question of naming scientific papers has been studied empirically. One recent paper looks at titles in certain social sciences and puts forth a taxonomy including a subclassification of compound titles called Metaphor-Topic, which is defined as having a metaphorical expression of the research concern given first and then the research concern is unveiled; the example provided is Slicing the Onion Ethnographically: Layers and Spaces in Multilingual Language Education Policy and Practice. This classification seems to fit the sparse Poisson paper described above, but I am very much struggling to understand the metaphor. Does putting things up to eleven have anything to do with intensity reconstruction?
What are your thoughts on the naming of papers? Is it good to be catchy, informative, both, or neither? Does it matter who the author is? Does it matter which article metrics one is interested in? I’m sure there are Big Data ways of finding out the answer, but would automatic text analytics algorithms really be trained for cultural references?