Posted by Artificial intelligence is constantly evolving. We know that it’s still largely improvable and nowadays the results are not always perfect. But exactly for this reason, in this “middle age” where the level of precision is going to increase, it’s interesting to see the current results provided by the algorithms when they are asked to understand measures and qualities that belong more to the human dimension. Like that now-famous Gloom Index that a data analyst applied to Radiohead’s songs in order to discover their most depressing composition.
Following a similar measure system, the journalist Miriam Quick performed a similar investigation in order to identify the saddest (and also the happiest) songs ever arrived at the top of the Billboard singles chart, in a champion of over one thousand titles. The unit of measurement used is the same as the Gloom Index above, i.e. a pair of numerical parameters exposed by the Spotify Web API, indicating the “valence” and the “energy” of a song. In Spotify’s internal algorithms, valence measures how much the mood is positive or negative, while energy gives an indication of how lively and dynamic the song is. Spotify uses already these parameters to automatically build its playlists: “relax”, “energy”, “happy mood” and so on.
The distribution of the tracks according to these two parameters has then been transposed graphically, following the typical psychological quadrants model: low energy/low valence songs are considered “sad”, low valence/high energy ones are “angry”, those with high valence and low energy are “calm” and those with both high values are “happy”. The result is the distribution that you find below:
Not all song titles are visible in the chart, but you can already notice that the songs of the “Happy” quadrant are many more than in the other quadrants, a proof that a cheerful song has more chance to end up first in the charts compared to other moods. In any case, in her exhaustive article published on BBC the journalist identified the following five songs as the saddest ones returned by the Spotify algorithm.
The saddest song (above) can actually be defined as a romantic ballad, but its extreme slowness and the calm mood gave to Spotify the “feeling” of a sad song. If we bring human sensations into the game, perhaps the only really sad song among those ones is Elvis Presley’s one. In the other indicators the algorithm is more accurate, identifying songs like Outkast’s Hey Ya! or Macarena among the happiest compositions ever, and Don’t Worry (Be Happy) as one of the most calm ones.
Spotify has not yet revealed how those two parameters are calculated and what they are really based on, although they should be based on the average sensations perceived by humans. What is sure is that the algorithms are in a constant phase of improvement. And one day, when they become perfectly reliable, we will be able to unequivocally identify the saddest song ever made. Or the happiest one, depending on what you are most interested in.
