MusicBrainz Summit / 11 / Session Notes

• Kuno Woudt (warp)
• Pavan Chander (navap)
• Rob Kaye (ruaok)
• Nikki
• Oliver Charles (ocharles)
• Jamie McDonald (jdamcd)
• Nicolás Tamargo (reosarevok)
• CatCat
• Per Øyvind Øygard (Wizzcat)
• Paul Taylor (ijabz)
• Mathias Kunter (mathiaskunter)
• Hilbert Woudt (monedula)

Sponsor representatives

• musiXmatch: Valerio Paolini
• Last.fm: Adrian Woodhead (massdosage)
• Google/Freebase: Micah Saul (micahsaul)
• Zvooq: Andrey Popp (andreypopp)
• BBC: Dave Evans (djce)

• They have had a lot of personnel changes over the last few years, but would like to re-establish a relationship with MB
• Are looking to switch to NGS schema by the end of the year
• They would like to use MBIDs internally to make communication between incoming data sets easier
• Will consider sharing partial label feed/data
• Might actually solve their artist disambiguation issue soon..ish

• They are a Spotify competitor in Russia that focuses on music released worldwide

• They are a lyrics database
• World wide license from Sony, Universal, EMI, Warner, BMG, Kobalt

• Freebase is a big data repository of various data sets covering movies, music, sports, people, locations, and others
• http://freebase.com

• They are looking to finally make the switch to NGS
• Their music news website now uses ws/2
• They outsource their album reviews and MB data entry to Unique Broadcasting Company

Goals

• Not storing plaintext passwords
• Not having knowable (i.e. reversible) passwords
• Not transmitting passwords in the clear
• Single sign on

Questions

• What specific password issues are we trying to solve?

Discussed proposals

• Implement OpenID
• Using digest authentication (still requires storing and transferring the clear text password)
• Using SSL (requires updating web service libraries)
• Using a separate LDAP server (password no longer in MB database and stored elsewhere, also allows for possible single sign on integration)

Conclusion: Use LDAP and phase in SSL to increase password security. Bonus: LDAP makes single sign on possible.

• Universal is considering handing over their entire cover art archive to us
• Labels actually don't own copyright on cover art
• There are potential messy legal issues to using cover art
• The Internet Archive functions as a library and can act as a 'cover art shelter' for us
• Possible process:
  • A release's MBID can be used to receieve a cover art image
  • If you know a release's MBID you can do a GET and receive a cover art image
  • Track cover art uploads by user and also use regular voting process
  • Images will be provided as a hi-res (~15 MB) and as a low-res (500 px)

Questions

• Does the user have to upload JPEG or can the server transcode?
• What status code will we return when a 'darkened' image exists but we're not allowed to display it?
• If we get cover art from Universal, how do we match each image up with a release?
• How do we handle a release group with many releases (i.e., do we use the same image?)
• How do we handle multiple images (e.g. front, back, obi, liner notes, cd faces, etc.)

Cover_Art_Archive

Cover_Art_Wishlist

• See Sunday

Goals

• Allow grouping edits together and bulk submitting them
• Allow editing an edit and resubmitting it without impacting the edit queue
• Allow editing via the web service, eventually

Bookbrainz

• Oliver's pet project and testing ground for future MB framework changes
• BB emulates git
  • it allows building a stack of changes and then submitting all of them together in one 'commit'
  • It takes a snapshot of the data at the time. We don't have that with historical edits so migrating old edits is a problem

Further reading regarding

• wikipedia:Inter-rater_reliability (especially the links at the end)
• http://www.mitpressjournals.org/doi/abs/10.1162/COLI_a_00074

• Roll out 3scale and move all commercial users over to a pay2play system with different packages
• Non-commercial users would use the free2play rate-limited system with the option of paying for better access

• We all hate PUIDs and we need to move forward
• Acoustid looks very promising, it's open source, file oriented, and has strong ties with MB
• http://acoustid.org/
• May be possible to bulk fingerprint some data sources

• Do we go with one provider or several?
  • Start with Songkick, but stay open to the option of different providers - especially to gain global coverage
• Do we concentrate on future events or archived events?
  • Initially link to Songkick for future events
  • Create a new setlist entity for past events
  • Create a new venue entity
• Need to consider Location, would be useful for artist as well as for events.

• Similar to the remaster issue
• Do we add further levels of abstraction?
  • No. We're already saturated with entities. We need better definitions
  • ...and we still haven't totally defined works
• Do we count silence as a divergence point?

• Announce the closing of trac (and all its tickets) and the deprecation of subversion
• svn.musicbrainz.org will remain as an interface for the search server
• Consider replacing gitweb with github in a more official capacity

• A new field that is to be used specifically for genres
• Features: autocomplete, canonical names,
• Micah is offering genre data based on wikipedia

This is not a complete or final model and not official!

• "Drug dealer" model - free the first time, get addicted, pay for easy further access
• Data dumps (twice a week)
  • Public $100* *suggested
  • CC-NC $250 (Paying for commercial use of NC use data)
• Live data feed ($/mth)
  • Twice-weekly $500
  • Daily $1500
  • Hourly $2500
• Web service calls (flat fee)
  • 10K $10
  • 25K $20
  • 50K $30
  • 100K $50
• Virtual machine
  • VM + Data $300
  • VM + Data + Search $400
• Tagger Affiliate Program
  • TBD: Clarification of the scope of the program
  • TBD: Web service referral kickbacks

• Lyrics from musiXmatch
  • daily updates, but will start with weekly ones
  • updates will include all MB/mXm matched lyrics
  • lyrics can be added also from edit interface
  • How do we best use their lyrics data?
    • Solution: Link to mxm via a lyrics icon in the tracklist and a proper link on the recording page
• See also Monday

• Videos are becoming increasingly common as a music release medium (e.g. iTunes)
• Will require major schema changes and looking at the long term goals of MusicBrainz
• Solution: Table the discussion for now, reopen in a different setting with developers

• There is a need to group multiple releases together when each release is the exact same - just released in a different country
• Due to tradition, different countries/regions issue releases on different days of the week
• Solution: Allow multiple release events per release when the label, barcode, and tracklist is the same

• Unknown end date (dead/disbanded, but we don't exactly know when)
  • Solution: Add a column to the date table to specifically state that the entity is dead/disbanded, but we don't know when
• Fuzzy dates (16th century composer edge cases)
  • Solution: Use a 'century' column

• User:Wizzcat/Data_Quality_Extension
• http://wiki.xabbu.net/Data_quality
• Current implementation of data quality has a bad name, is poorly defined, and isn't used
• What do we want to solve?
  • Explicitly state that a release has been reviewed/verified
    • Solution: +1 / -1 votes that decay in weight over some function of time
  • Protect against ignorance (The White Album vs The Beatles)
    • Solution: Add a 'Protected' flag (i.e. edits expire by default)
  • Measure of completeness
    • Solution: "Completed as per liner notes" checkbox that is accessible via the WS
• Conclusion: High quality is the protected flag, default quality is default, low quality goes away

• Currently, 'remix' and 'soundtrack' are at the same meta level as 'album' or 'lp'
• Conclusion: Postponed till a proposal can be drawn up

• Improve the explanation that is shown at the top of each reports' page
• Improve report flow (e.g. ability to hide items from reports)

• Allow marking an entry as 'done'
• Default report list should filter out all entries marked as done with more than X votes
• Allow viewing the report with the filtered out entries

• List all emails in a site inbox
• Create a dynamic list of subscribed artists with open edits

• As finances improve, employ a dedicated person that will lead the testing

• Filter on release group properties
• Use infinite scroll
• Be able to reorder, add, remove, and sort columns

• Switch from a hierarchical tree to attributes

• What information should we show?
  • http://www.discogs.com/
  • http://www.freebase.com/
  • http://www.soundunwound.com/

• Change from a tree to a graph
  • Flatten the graph into a tree and allow an instrument to have multiple parents
• Add model support to the instrument tree
• Importing freebase data
  • How often do we sync the data?
  • How do we reconcile differences in data?
  • How often do deletes/merges/changes happen?
• Going forward, if we need a new instrument we would add it to freebase

• "I am very happy to declare Universal's support for MusicBrainz and its community" - Innovation Manager at Universal Music Group International

• Default tracklist page shows the advanced view
• For new releases you see the add disc dialog
• The track parser moves into the add disc dialog
• There needs to be a way to reparse from the advanced view

• Remove unneeded extensions
• Update to Ubuntu's MediaWiki package
• Get the API working
• Install wiki at /wiki/Article and then redirect to /Article
• Write a wiki test suite

• Last.fm would like to create 'best of the decade' lists and filter out data such as the 2009 re-release of The Beatles
• Currently, release group dates match the date of the earliest release in that group, but in the case of re-releases we often only have data on the modern release and are missing (for example) the original '70s vinyl release
• Solution: Add an editable initial date field at the release group level
  • The date field will default to empty because anyone who wants the group date can guess via its earliest release (like MB does now)

• short description of musiXmatch expectations
• feedback from MB Editors on musiXmatch contributions
• Editors' willingness to help musiXmatch (IRC channel)
• musiXmatch will report unexpected Edit Interface behaviour (for example Split Artists while adding a Release)
• change usernames to make them easily identifiable (add customer name to username)
• provide guidelines for interactions between MB Editors and external Editors

• How do we properly link to different data sets? (e.g., musiXmatch, soundunwound, last.fm, etc.)
  • Solution: Build a generic framework that allows us to import any external data set and reconcile it with the data we have
  • Use a second "integration database" that contains all raw data from external sources (label feeds, partners, etc.)
  • Import data into the main database with a de-duplication script, but do not remove any of the original raw data (this allows further parsing in the future)
  • Also look into Google Refine for manual reconciliation: http://code.google.com/p/google-refine/
• A long term goal is to create an editing API that we can gradually open up to our data partners and the ecosystem
  • This will allow partners like Zvooq to edit data on their website, but feed the changes back to the rest of the MB ecosystem

Feature (votes):

1. Edit system (9)
2. Group multiple release events together (6)
2. Data quality (6)
3. 3^rd party data set integration (5)
3. Single sign on & password security (5)
4. Instrument tree (4)
4. Genres (4)
4. Medium attributes (4)
5. Release group attributes (3)
6. Music dashboard (2)
7. Tracklist/medium overhaul with video support (1)
7. Pagination (1)
7. Site notifications (1)
7. Report improvements (1)
8. Date improvements (0)
8. Auto-editor elections (0)
8. Full classical support (0)