There is an unfortunate tendency for people to make papers available on their websites and give them highly descriptive names like
types03.pdf or even the gem
tr.pdf. I have a directory or two full of files with names like that and despite the promises of Beagle and Spotlight it is very hard to find certain documents.
I know this is something Joshua rightful complains about, and I'm admittedly guilty of this to some degree myself. So I put it on my to-do list a while back and finally decided to take a few minutes today and sort out the filenames I've assigned to my papers. However, when getting started I was faced with the dilemma of just what the naming convention should be. I did some poking around with Google, but couldn't really find any information on choices that other individuals or organizations have made.
Clearly the year and some approximation of the authors last names should be used, but there are a lot of trade-offs.
One system that I do know about is the one used by the Church Project at Boston University. If I remember correctly the filenames and pages are generated from the BibTeX entries. For example, consider the paper »Type inference, principal typings, and let-polymorphism for first-class mixin modules« by Henning Makholm and J. B. Wells. This paper gets a web page named http://types.bu.edu/reports/Mak+Wel:ICFP-2005.html and a PDF file with a very long name. So long that I will instead just describe the format: the last name of the authors separated by the plus symbol, a colon, title of the paper, a colon, conference venue, a dash, and year. While very descriptive this scheme has a few problems. Firstly, colon is a reserved character in at least a few operating systems. Secondly, is that it is just really long name. I can easily imagine one papers with several authors that the name would exceed the common 255 character filename limit that many filesystems have.
So if anyone has suggestions for developing a consistent naming convention, let me know. While more descriptive filenames would be useful, perhaps what is really needed is the metadata equivalent of »ls«. Then we would be in the position to simply complain about the fact that people don't fill in the metadata fields of their documents.