Talk:Converting ID3 tags to RDF
From Music Ontology Wiki
Well, I put my comments here. I will not repeat all the discussion which has occured on the Google group.
There is a debate which opposes two viewpoints concerning what the ID3 tags actually represent. On the one hand, Yves Raimond argues that the ID3 tags basically apply to mo:Track and mo:Record, while Antoine Zimmermann argues that it merely identifies a mo:Signal (but in both cases, an mo:Track or mo:Signal can be instantiated with an anonymous resource, related to the identified ones).
Arguments in favour of mapping ID3 tags to mo:Signal:
- The tags usually cannot discriminates between several publication of the same Signal (e.g., the tags are not enough to determine either a CD release or a iTunes MP3 release).
- User:YvesRaimond - What about the date, the track length, the track number? Even it is impossible to discriminate, one MP3 file correspond to one unique release, not several - the mapping is unique. owl:sameAs linking to further release databases can perfectly be done in a further process.
- There are also files which are not related to an identifiable release because they are just a bunch of files available online.
- User:YvesRaimond - I would argue a bunch of files available online _are_ a release :)
- On the other hand, the Signal is usually identifiable with only the title, the album name and the artist name.
- Alice has an MP3 file with id3:TITLE "Sunday Bloody Sunday", id3:ARTIST "U2", id3:ALBUM "War" and id3:YEAR "1983". She can identify the signal (the one of the album War by U2) but not the exact Record (is it the UK CD release? the US Vinyl release? The cassette release? etc. Bob has a file with id3:TITLE "Sunday Bloody Sunday", id3:ARTIST "U2", id3:ALBUM "War" and id3:YEAR "1983". If Alice and Bob map their tags to a Release, they cannot deduce they have a title in common (except by comparing character strings, which they could have done without RDF). If Alice and Bob map their tags to a Signal, they know it is the same Signal (admitting that there is no other recording of a song called "Sunday Bloody Sunday" on another album called "War" by a band called "U2" and released or recorded on 1983).
- User:YvesRaimond - When mapping to a release, they can perfectly relate their two releases using an instance of mo:Signal (if they're unsure they got the MP3 from the same release). (But be careful! As you mention, it might be the case another album is really similar to this one, but not the same (not in that case, though - but especially true if you lack quite a lot of ID3 tags). Better to refer to external editorial database and good interlinking algorithms for that IMHO.)
:alicerecord a mo:Record; dc:title "War"; foaf:maker :u2. :bobrecord a mo:Record; dc:title "War"; foaf:maker :u2. :sig a mo:Signal; mo:published_as :alicerecord, :bobrecord.
I just think such editorial information should be attached to what is edited :-) (foaf:maker, in that case, is just a really weak property referring to the "big name" on the edited version).
Arguments in favour of mapping ID3 tags
- mo:Records can be identified on existing knowledge bases. One can use this information. [Counter Argument: the information about a Record is containing the information about the underlying Signal. So the information about the Signal is there too.]
- User:YvesRaimond - I don't get that last point. The information is about an edited version (track listing etc.) - look at the two releases of "The Process of Belief" on Musicbrainz, for example. However, the information to join these two records up is not available atm on Musicbrainz.
...I'm tired, I will work on it later.
User:YvesRaimond :-D
