Acousticbrainz full size csv2rdf

csv2rdf/acousticbrainz/ab_reconciliation_details.md

@@ -1,9 +1,72 @@
 #   Properties:
+
+##  Genres:
+
 The reconciliation of AcousticBrainz properties are not specific. 
 For genre properties, if the cells in that column is all literal objects, then the property uses the AcousticBrainz documentation page for that genre. Since AcousticBrainz documentations include links and detailed description for feature extraction of genres, I decide to use the source description of the property.
 For the wikidata link objects, I have no choice but to use the wikidata property for "genre" (P136). Since the AcousticBrainz genre feature-extracted and is specified in much more details, this causes some loss of meaning in the detail of the genres. 
 The same principle applies to the columns for mood.
 
+##  Machine Versions:
+
+Since there is no properties in Wikidata and Schema.org, I choose the official download website for the feature extraction machines.
+Build_sha and Git_sha are made-up URIs. It uses the same URI as essentia and gaia followed by a fragment identifier.
+The probability for all features are also made-up URIs. It uses the same URI as the feature followed by a ".probability".
+
+##  Testing Purpose:
+
+An acoustic_brainz_reconciled_short.csv is used for testing with less records but same columns. It contains the first 3000 records of the full CSV.
+
+##  Made-up URIs:
+
+metadata.tags.engineer
+metadata.tags.mixer
+metadata.tags.compilation
+metadata.tags.notes
+metadata.tags.ensemble
+metadata.tags.replaygain_reference_loudness
+metadata.tags.acoustid_id
+metadata.tags.mp3gain_album_minmax
+metadata.tags.mp3gain_minmax
+metadata.tags.replaygain_album_gain
+metadata.tags.replaygain_album_peak
+metadata.tags.replaygain_track_gain
+metadata.tags.replaygain_track_peak
+metadata.tags.ripping tool
+metadata.tags.retail date
+metadata.tags.supplier
+metadata.tags.wcop
+metadata.tags.woaf
+metadata.tags.www
+metadata.tags.custom1
+all that contains map3 gain, replay gain
+metadata.tags.djmixer
+metadata.tags.trash
+metadata.tags.acoustid_fingerprint
+metadata.tags.gracenotefileid
+metadata.tags.gracenoteextdata
+metadata.tags.remixer
+metadata.tags.eaclog
+metadata.tags.accurateripcountalloffsets
+metadata.tags.accurateripcrc
+metadata.tags.accurateripcount
+metadata.tags.accurateripid
+metadata.tags.accurateriptotal
+
+##  Not accurate URIs:
+
+metadata.tags.rip date
+metadata.tags.white label: schema.org > MusicRelease, sell under multiple brands/labels
+metadata.tags.contentgroup: wikidata > member of, the content is a member of
+all album artist: wikidata > performer, since there's no album artist in wikidata and schema.org, so we use "artist", which is performer in wkd.
+all artist credit: wikidata > performer, since there's no artist credit in wikidata and schema.org, so we use performer with the same reason.
+[metadata.tags.artistwebpage,
+metadata.tags.paymentwebpage,
+metadata.tags.publisherwebpage,
+metadata.tags.radiostationwebpage]: schema.org > WebPage, no specification for artist's webpage.
+metadata.tags.jamendo-track-id: wikidata > Jamendo album ID, no track ID in wikidata. We use the Jamendo album ID.
+
+
 #   Instances:
 For testing, the gender and genres are reconciled against Wikidata.
 Albums, artists, etc are reconciled against their MusicBrainz links since they already have MBIDs.