Re: [Tracker] data clustering and metadata storage



Peter Rullmann wrote:
Hi Jamie,

is it possible with tracker to do some sort of simple data clustering
to refine a query?
Tracker would examine the attributes of a resulting set of files and
offer a list of key-value pairs that help to refine the query. That
are keys that have different values (n/a is also an option) in the
resulting files.
To rank the keys by relevance for refining the query, one could prefer
the keys with the least different values. Or better: For each key,
count the number of files with the least used value and prefer the
keys where that number of files is the biggest.
But I am not an expert on information retrieval. I guess there are
theories for that :).

An example: If I search for "Frank", it could ask me for
type=music(number of hits), pictures(..) or text(..). After Clicking
on type=music, it would ask me if I mean Artist="Frank Sinatra"(..) or
"Frank Popp Ensemble"(..).

not easy to do that. You can apply < and > as prefixes to increase relevance of individual terms in the search query but I will need to modify my preformatting of search terms to allow this.



Another question:
How can I add persistent metadata to a file like a description to a .txt file?
If I add it only to the Tracker database, it will be lost at the next
rescan or when I copy the file to my usb stick for example. If I store
it in a Linux Extended Attribute, it won't be indexed by Tracker.
Whats your opinion on Linux Extended Attributes and MacOS X hidden
resource files?

At some point we will cache non-embedded metadata in EA's if they are available and you have write permission on the file. Some files like ogg can store arbitrary metadata in the actual file so thats another option.

AFAIK USB sticks do not support EAs (they mostly use FAT)

Data is not lost in a rescan. A rebuild of the tracker DB would lose it but I plan to backup metadata and keywords in some form before doing so (probably csv format or something).


How can I search for gnome emblems, that I have given to a file?

Not implemented yet - the preferred way is to get Nautilus to use tracker though Im getting strong resistance to that from the maintainer but hopefully that will change once tracker is officially in Gnome.




Thanks in advance and keep up the good work. I think it is excellent
that there is a fast, small and powerfull alternative to Beagle et al.

yes such stuff is better written in C IMO.


--
Mr Jamie McCracken
http://jamiemcc.livejournal.com/




[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]