Researchers at the University of Bristol have developed state-of-the-art algorithms that allow a specialized software to detect if a song will become a hit. The discovery has already drawn massive interest from worldwide media, with many taking interest in it for obvious reasons.Just to be clear, what the algorithms do is predict the chances a song has of going viral, and climbing at the top of the charts. The finding was presented at an international workshop last week, the team says.
The meeting was the MML 2011 4th International Workshop on Machine Learning and Music: Learning from Musical Structure. It was held in conjunction with the 25th Annual Conference on Neural Information Processing Systems, and took place on December 17, in Sierra Nevada, Spain.
The work was conducted by experts at the UB Faculty of Engineering Intelligent Systems Laboratory (ISL), who were led by researcher Dr Tijl de Bie. In order to develop the algorithms, the team analyzed top 40 charts in the United Kingdom for the past 50 years.
In order to have a few objective factors by which to analyze the songs, researchers selected simple attributes such as song duration, loudness, tempo and time signature for their algorithms. However, that proved to offer an incomplete, inaccurate description of their likelihood to become hits.
In order to fix this issue, the team factored in some more complex attributes, including the “noisiness” of the sound, the complexity of the chord sequences being used, as well as the complexity of the harmony on which the entire song was based.
Based on these factors, the team was able to compile a hit potential equation. In all, 23 features are taken into account when assessing a new song. For every new entry, the system uses the chart data the researchers computed in when they created it.
The equation proves to be successful, the team says, since it is able to predict about 60 percent of the songs that will make it as a hit. A hit is defined as a song that will climb somewhere in the top 5 positions, while a flop is a song that will never climb beyond position #30.
“Musical tastes evolve, which means our ‘hit potential equation’ needs to evolve as well. Indeed, we have found the hit potential of a song depends on the era. This may be due to the varying dominant music style, culture and environment,” De Bie explains.
As an interesting side-note, the system had difficulties predicting hits for the late 1970s and early 1980s, suggesting that indeed these periods were characterized by innovative music and composition styles. Conversely, the algorithms worked very well for music put on the market beyond 2000.
A website based on the tool can be visited here.