comunidad integral, alumno integral’)
2. PREGUNTAS DE INVESTIGACIÓN Y OBJETIVOS
3.1 Aproximación al estudio
In this investigation, described in detail in Chapter 7, we did not find an upward trend in the number of new genres up to the social media era. In our composite datasets, the rate of genre proliferation increases, up to the end of the cassette era, and then drops-off during the digital era. The mean, in both sets of data, falls to less than half of the peak value in the Internet era, and rapidly drops off in the social media era. See Table 10.3.1 below, which is a copy of Table 7.4.1 from Section 7.4, for an illustration of this.
ΩY Category Total Genres Category-Genres Mean
1899 Pre-recording (pre-rec) 22 22 0.047 1920 Phonograph (phono) 83 61 2.905 1954 Radio (radio) 308 225 6.618 1971 Microgroove (micro) 593 285 16.765 1987 Cassette (cass) 946 353 22.063 1997 Digital (dig) 1130 184 18.400 2007 Internet (net) 1214 84 8.400 2015 Social media (sm) 1227 13 1.625
Table 10.3.1: Genre inception (date-corrected dataset).
In earlier eras, the genre numbers all tend to increase, implying perhaps that tech- nologies and an ever-developing recorded-music industry are responsible for this. Also implicated is the introduction of consumer recording equipment; cassette tape in the case of The Echo Nest. This era marks the peak in genre numbers in those data.
The post-Internet drop is surprising, because Echo Nest staff and clients have the op- portunity to ‘seed’ new genres into the system (McDonald, 2013). An increase of 119 genre-categories in The Echo Nest over the duration of this study implies a degree of artificial proliferation caused by this seeding. These same interventions may also be a driver of genre fragmentation: a client-seeded genre may be for a single artist (a micro-genre). In cases such as these, not only would a single artist be responsible for all genre activity, thus making that genre less visible in the larger data, but our pro- cess would discard that genre since it could not generate a 2-artist cluster. This would indicate that genres proliferate through fragmentation as opposed to new genre in-
ception, and our low rates of genre proliferation in later eras are due to this.
Our work with Every Noise At Once (ENAO) is a further indication of genre fragmen- tation. When considering the data from Spotify, it seems doubtful that 1554 genres have been ‘born’ between April 2016 and May 2019. When we consider 2018 and 2019 only, where we have results for 17 months, the number of genres increases by 1498: an average of 88 new genres per month. Table 10.3.2 (a partial copy of Table 7.4.4 from Section 7.4.5) shows these increases.
Date Genres Increase over last
Dec. 31 2017 1539 43 Jul. 26 2018 1896 357 Sep. 30 2018 2073 177 Oct. 24 2018 2193 120 Nov. 30 2018 2368 175 Dec. 29 2018 2474 106 Jan. 19 2019 2538 64 Mar. 31 2019 2877 339 Apr. 23 2019 2935 58 May 14 2019 3037 102
Table 10.3.2: ENAO genre increases.
The increasing numbers of genres appearing in Spotify in 2018 and 2019, when con- sidered alongside the apparent drop in proliferation in other data, demonstrate that genre fragmentation is the case, and indicate that streaming systems are implicated in this process.
As discussed in Chapter 8, the out-degree centrality values (examined in the EN/MB datasets) tend to decrease in the date-corrected networks in each era until the digital era. In the minimal dataset, out-degree centrality drops in every era. In terms of mean out-degree, the value rises in every era in both datasets. The centrality values show that the fraction of genres that are descendants of a given genre drops in both datasets, but then rises (in the corrected EN/MB data) from 1997 onwards. Table 10.3.3 below, a copy of Table 8.2.2, shows these values for the date-corrected data.
ΩY Mean Out-Degree Centrality Average Out-Degree 1899 0.103896103896 2.1818 1920 0.0562738759918 4.6145 1954 0.0212149414104 6.5130 1971 0.0146700241557 8.6847 1987 0.0110741971207 10.4651 1997 0.0112073492871 12.6531 2007 0.0115912051078 14.0601 2015 0.0115967405481 14.2176
Table 10.3.3: Out-degree characteristics (date-corrected).
Increasing mean out-degree values could indicate increased genre fragmentation; this in made more likely because fragmentation into micro-genres implies genre-invisibility in our processed data. The decreasing centrality values can be explained by the in- creases in overall genre numbers. In the latter eras the increase in genre numbers stalls and proliferation-through- fragmentation overtakes proliferation-through-inception. The MusicBrainz user-tag derived dataset offers a slightly different picture: genre in- ception continues to increase into the digital era, with proliferation receding in the Internet era, where it again drops to less than half of the peak value, and more then rapidly decreasing in the social media era (see Table 10.3.4, copied from Table 7.4.3 in Section 7.4.4).
ΩY Category Total Genres Category-Genres Mean
1899 Pre-recording (pre-rec) 4 4 0.010 1920 Phonograph (phono) 8 5 0.238 1954 Radio (radio) 56 48 1.412 1971 Microgroove (micro) 161 105 6.176 1987 Cassette (cass) 286 125 7.813 1997 Digital (dig) 391 105 10.500 2007 Internet (net) 441 50 5.000 2015 Social media (sm) 445 4 0.500
Table 10.3.4: Genre inception (MusicBrainz tag-derived dataset).
with an emphasis on physical recordings: there being more music on CD represented in such a system is to be expected. It is also likely that the ‘folksonomic’ nature of MusicBrainz5creates a different temporal distribution of genres to that found in Echo Nest.
There are a number of possible explanations for these findings as they relate to genre inception and proliferation. As discussed, It is possible that proliferation has slowed because genre fragmentation has occurred, whereby numerous micro-genres have splintered from larger parents. Many of these are too small (in terms of member- ship and activity) to be processed by the Echo Nest system, to be recognised by Mu- sicBrainz editors, or be included in our clustering process. Alternatively, new genres may not be recognised by systems because they are exhibiting ‘tag-lag’: they have not
yet generated sufficient activity to be recognised, or (as in the case of MusicBrainz
or Wikidata) for editors to have added the artists to the database. Alternatively, user- edited tags (such as those in MusicBrainz) may not be recognised as genres at all.