• impud1cus about 1 month ago

    Discogs really needs to have better data around compositions on releases. This user need is greatest for classical and jazz works, I would think, but is applicable anywhere (including folk music, although this may require different authorities). And before I go any further I should say that the "tracks" feature is not what I am looking for! In this context it is useless at best, because it is very much about recordings of compositions, not about the compositions themselves.

    The problem
    Discogs has unique identifiers for people, together with "ANVs". These entities are linked explicitly to the releases to which they contributed, and users can navigate well across the site via these links. There are unique identifiers for "tracks" in Discogs too - sort of - being automatically generated using some combination of track properties. This works to an extent where the recordings are actually the same, because the differences in duration, artist and title may be minor. It does not work to identify compositions ("works") as opposed to tracks. Compositions are not recordings, and in most genres of music, compositions may well be recorded by different artists (whether the artist thinks about what they are doing as interpreting a traditional song, making a "cover version", or performing a classical composition, or improvising on a theme. In many situations (certainly in classical music) it's more important to be talking about the same composition than to be talking about the same recording. The names and translations given to compositions and parts of compositions vary as much as those of their composers - but they are also hugely repetitive because there are thousands upon thousands of "symphonies" out there, but even adding a composer and a number will do a very poor job of distinguishing between them. Anyway a text-based search is no substitute for a direct link to the right result - which is why release pages link to artist pages rather than to a search!

    Therefore we need a way to apply unique identifiers to composition on releases in Discogs. These will undeniably require a lot of work when they are input, but they will add enormous value when added and, additionally, can probably be automated to some extent as well (using the sources I will come to). First, some use cases to indicate ways that unique "work" identifiers could help.

    Use case 1: I am on my phone looking up a classical record prior to purchase and want to know if I already have the compositions on it on other LPs in my collection. I want to click on the composition name (NOT the composer or performer) and see a list of releases featuring the same composition. Ideally I will see how many of these are already in my collection, or be able to apply the same work-filter to my collection (but collection search is another bug-bear so supporting this is a whole other problem)
    Use case 1a: OK yes I know I already have 7 versions of Tchaikovsky's Violin Concerto, but do I have a recording by Ruggiero Ricci? In other words I want to unambiguously combine an artist ID with a work ID.

    Use case 2: I want to find a recording of a Bartók's Second Piano concerto. What is the most recent recording in Discogs? Who has performed that piece over the years? Which performance has the highest rating? The composition is referred to in many different languages and in multiple ways even in English, and a free-text search won't find all examples even whilst it returns lots of irrelevant results.

    Use case 3: The Damned wrote a song called "Love Song" (The Damned - Love Song). So did lots of other people. Who has actually recorded a cover of this "Love Song"? (here's one who did A House - Here Come The Good Times). It's not impossible to find this out with some clever searching but it would be a damn sight easier if each track was linked to the "work" it is based on.

    Proposed solution
    I would like to be able to attach a property an identifier from one of a number of authorities on works. This should be at the level of a track rather than a release. Following the link that would result I could then rapidly see what other releases contain that same work ID for any of the use-cases above. Ultimately it requires two new input fields, working similarly to the BOI field but at track level: a dropdown for the authority source; and a field for the URI or ID.

    The coverage of existing authorities is incomplete, but strong in certain areas. One place with excellent coverage is ISWC, which I have seen mentioned in other forum threads. Others are VIAF and the Library of Congress (links at foot of post). There is a white paper from Europeana that considered these and recommended the latter - their scenario being the aggregation of millions of records of musical recordings from across Europe.

    Here are some examples:

    Bartók's 2nd Piano Concerto:

    Discogs releases with differing forms of the name:
    "Klavierkonzerte Nr. 2" https://www.discogs.com/B%C3%A9la-Bart%C3%B3k-G%C3%A9za-Anda-Ferenc-Fricsay-Radio-Symphonie-Orchester-Berlin-Klavierkonzerte-Nr-2-Und-3/release/12788431
    "Second Piano Concerto" https://www.discogs.com/B%C3%A9la-Bart%C3%B3k-Second-Piano-Concerto-Sonata-For-Piano-1926/release/9105452
    "Concerto N°2 For Piano And Orchestra" https://www.discogs.com/B%C3%A9la-Bart%C3%B3k-Edith-Farnadi-Orchestra-Of-The-Vienna-State-Opera-Conducted-By-Hermann-Scherchen-Conce/master/1202016
    (Incidentally these have various auto-generated tracks that, inevitably, each only have one related release)

    The LoC Name Authority File (in the "FRBRwork" collection):
    The best data view of it is probably the RDF XML: http://id.loc.gov/authorities/names/no96055353.rdf

    The VIAF record:

    And here is the composer's own record which, incidentally, should also be possible in Discogs! Linked data, y'all
    Bartók, Béla, 1881-1945. https://viaf.org/viaf/89006617/#Bart%C3%B3k,_B%C3%A9la,_1881-1945.

    Here is the ISWC ID for the Concerto:

    When it comes to non-classical music I find VIAF and LoC to be more patchy, but ISWC is better

    Dave Vanian of the Damned is in LoC
    But I can't get an ID for "Love Song"

    However his creator ID on ISWC is 0073116007, and Love Song has ISWC ID T-010.475.320-9

    There is no getting away from the fact that finding the IDs is a chore and adding them as track properties would not be for everyone, but it would be made less onerous once Discogs started to contain IDs in similar releases, which we could refer to more easily than the VIAF, LOC etc. I know I would add them to all my classical collection because it would immediately make it so much easier to know what I had.

    Europeana Sounds white paper: https://pro.europeana.eu/files/Europeana_Professional/Projects/Project_list/Europeana_Sounds/Deliverables/EuropeanaSounds-D1.3-Ontologies-for-sound-v1.2.pdf
    ISWC: https://iswcnet.cisac.org/logon.do It's ugly as hell but not actually that hard to use. Provides IDs but no shareable URLs.
    LoC: search using the FRBR Work collection e.g. http://id.loc.gov/search/?q=john+field+piano&q=memberOf%3Ahttp%3A%2F%2Fid.loc.gov%2Fauthorities%2Fnames%2Fcollection_FRBRWork&q=cs%3Ahttp%3A%2F%2Fid.loc.gov%2Fauthorities%2Fnames
    VIAF: search here using the "works" field https://viaf.org/
  • hatfulofelt about 1 month ago

    Well worth bumping, thanks for the thread/info.
  • impud1cus about 1 month ago

    thanks for the support and the bump, hatfulofelt!
  • mjb 29 days ago

    unique identifier

    For what it's worth, ISWC (work/composition ID), ISRC (recording ID), and IPI (person ID) are not guaranteed to be unique. Each code only refers to one thing, but that one thing can have multiple codes. So the user interface would have to allow for that.
  • impud1cus 29 days ago

    thanks for that, mjb, good point. So they are unique in the sense that if you point at them you know that you're pointing at only one thing, but not unique in the sense that they are not the only way you could point at that thing. This is important, but less of a problem than the other way around because you can disambiguate those retrospectively (i.e. merge them) without changing the records of all the things that are pointing at the multiple codes for the same thing. Whereas the code turned out to represent multiple things you would have to revisit every record that used the code and make a decision!

    Anyway it's a very valuable point and one we'd need to consider in choosing which vocabularies would be favoured here, because it's not ideal at all if you have to pick between multiple ISWC IDs for the same work. I haven't looked properly at Wikidata either, which is fast becoming a linked data hub that can stand in for all sorts of other vocabs and can deal with this disambiguation too.

Log In You must be logged in to post.