sakshath

apologies to you, too, for not meeting your expections! We’re aware of the sore spots you’re pointing out – they are our top priority. Please also keep in mind that Paulo tested our very first public beta version (0.5.7), and the current version is still quite early (beta 0.6.3), so there are major changes with every release.

Regarding the quality of the metadata extraction: There are great variations between different journals and layouts – many are parsed perfectly, and for some it doesn’t work at all. For the next major release (beta 0.7, due in approx. 6-8 weeks), author & title extraction should be improved significantly across the board; moreover, we’ll make use of the PubMed API to retrieve metadata for most bio/life science papers.

Regarding the crashes: All I can say is we’re very sorry – and all known bugs that lead to crashes (particularly during import) should be fixed in the next release (0.6.4) out early next week. If you should still experience crashes after that release, we’d be very grateful if you could send a bug report so we can trace it down. We do have Windows Vista and XP, Mac OS Tiger and Leopard, and many different Linux distributions running on our office machines, but of course we still can’t cover all possible configurations that lead to problems.

As for performance, we’ve so far focused our efforts on the handling of large libraries – the earliest versions got sluggish if you had more than 100 documents in them. We’ve tested beta 0.6.4 with libraries of more than 5,000 PDF documents (including cited references, that’s more than 100,000 metadata sets), and it’s working well, but it takes about a minute to start. Obviously, our goal for beta 0.7 is to support much larger libraries than that while still starting up much faster.

You were also mentioning the indexing speed. Indeed, if we were just indexing the full-text like Copernic or Google Desktop, we’d be much faster – it’s the metadata and reference recognition (which also involves server interaction) that slows it down. Admittedly, we haven’t had the chance to work on this since Paulo tested it, but we’ll be experimenting with a few ideas for improvements for beta 0.7.

I know this won’t solve your problems with the current release, but perhaps you can give the next ones another shot and see whether things have improved for you! Thanks in any case for trying,

best wishes,
Victor

The problem is that it rarely gets the metadata extraction right. So after quadrillions of cpu cycles wasted, and endless hours of thrashing away at the hard drive, you’re left entering your own metadata anyway! Moreover, I notice that Copernic Desktop Search (and Google’s) index the content of pdfs MUCH faster. But actually I suppose it is hard to compare because of the other stuff you mention that is simultaneously going on as it indexes the content.

I’ve been trying to add my folder of pdfs (thousands) but Mendeley keeps crashing on me. This is in XP.

I like the online integration. I hate the resource-heaviness, though I’m sure after the initial import it isn’t so bad. If only it would get to that stage! The database “structure”/handling is certainly much better than Zotero’s (BILLIONS of directories).

Using 50% of CPU resources is a lot, to be sure, but that’s only during automatic recognition (when all the heavy processing described above is done). Also, if it takes 5-10 seconds per PDF to do all the things I mentioned, then importing 700 files would take between 1-2 hours… But your point is taken – we’ll try to speed it up in future versions.

Cheers,
Victor

sorry to hear that you’ve had a less-than-desirable experience with our beta version. I agree that the automatic recognition you’re describing could be optimized – we’ll keep working on the speed and resource usage.

If I may, I’d like to add that Mendeley Desktop is doing a little bit more than just reading the PDF file. It extracts the full-text to make it searchable, it automatically tries to guess the correct metadata (author, title, journal, volume, issue, page numbers tc.) and extracts author-assigned keywords, and it parses each document’s cited references and turns them into machine-readable and exportable format. It also lets you share and synchronize all this data with other users.

Cheers,
Victor