Thi system works for me because:
- Encoded information is usually enough for me to identify the paper from filename only
- Filenames remain compact and UNIX shell friendly
- Filename collisions are nearly non-existent
aaaaaYY_tok1_jjjj.pdf
where:
aaaaa: Last name of the first author (variable length) YY: 2-digit year of publication (fixed length) tok1: first word of title, minus articles and other small words (variable length) jjj: abbreviation of journal or publication (variable length, but short)
For example:
Hochreiter, S., & Schmidhuber, J. (1997). LSTM can solve hard long time lag problems. In Advances in neural information processing systems 9. Cambridge, MA: MIT Press.
would be
hochreiter97_LSTM_anips.pdf
I then use the same without the .pdf as my BibTeX key (with an added :).
with the bibtex key hochreiter97:_LSTM_anips