16

When I tag a file, I expect that all files that are similar (given some metric, say tf-idf) be automatically tagged with the same tag.

For example, if I tag a file with "foo" all files similar to it would gain an auto-tag "foo" which I could confirm (or infirm) next time I see the file in dolphin.

This would allow the generation of tag clouds for files, but also facilitate bibliographic research: I have lots of papers and publications accumulated over the years, and finding those relevant is not always easy.

The files corresponding to a tag would then have a signature. These signatures could (if the user allows it) be uploaded to a central server, and downloaded to other users computers, thus increasing the power of the auto-tagging system.

The major advantage of this is to have much more efficient search for files. If find used proxmities, you could get a better ranking of the results, but not significantly: you typically do searches with at most 4 words. However, with tags and relation between them, the documents could be classified in a much more clever way, and thus consistently put the wanted document on top.

In the meantime you should be able to do a search for files with those properties and then tag all of them.

This is not the point: you can only search file by their content. Now imagine this, you have these files:

a : "boating on the sea is cool"
b : "beaches are relaxing"
c : "during my holidays, I like doing things cool and relaxing"

A semantic system makes it possible that tagging "a" with "holiday", and finding that "c" is a possible candidate (for that tag). Next time I see "c" and confirm the tag, "b" becomes a candidate (the trick is that proximities between files are computed).

Now if I add file d : "playing beach volley is cool", it automatically becomes a candidate. And the profile for the tag "holidays" becomes more and more refined.

The difference between that and searching myself for all files is that searching myself simply doesn't scale. Also, as nepomuk/soprano depends on xapian, it should be able to do that...