Back in my student days, the younger brother of a good friend would ask me to recommend new music to buy. Those days my tastes were going through an awkward transition from the underground lo-fi movement of the early 1990's to the darker electronic beats of the dance scene. The lad didn’t know where to find the records I listened to and liked what he heard me play. He’d ask me to recommend 3 or 4 records every month and bought them on the strength of my recommendation. I adopted the role of Music Mentor and took pride in recommending records I knew he would like.
2006 and I’ve been replaced by so-called sophisticated recommendation engines such as Last.fm and Pandora. These engines recommend songs based on what I like or don’t like. They do this by looking at what members of the community with similar tastes are listening to. It could be seen as a naivety and arrogance about recommendation engines that they ignore the complexities and uniqueness of the human personality in recommending what we should listen to. How can recommendation engines be so blasé about our complex psychological processes?
Before recommendation engines arrived I was already happy to stand down from my role as music mentor, as the Internet began to offer a rich toolkit for users to explore the musical horizon: - streaming audio, online radio, p2p file sharing, free downloads, online reviews forums etc supported by the ever-increasing bandwidth.
But when a recommendation engine suggests Elton John based on a Sheep on Drugs record then I’m concerned – a random artist generator could make a more relevant recommendation.
'When there is a lot of choice, you need recommendations,' says Martin Stiksel, the brains behind Last.fm
That is true. But how accurate are recommendations based on social familiarities?
I am a complex & unique psychological being – a collection of emotional, thought and behavioural patterns. Social recommendation sidesteps the delicate issue of humanistic personality theories. Decades of research by psychologists and philosophers who have studied personality are being ignored as I become a number, a statistic.
If social networks make no attempt to encapsulate our personal intricacies then should I be insulted that I am no longer regarded as an individual. Maybe, but Recommendation Engines are not claiming to be the saviour but merely an avenue to explore music. They’re not claiming to be the real deal and get it right every time but merely an aid to filter our choices.
In that case how can recommendation engines become more intelligent?
Did anyone see How Music Works on Channel 4 on Saturday, where Howard Goodall deconstructs music and looks at the elements? It was a sort of Music Uncovered. Howard Goodall got me thinking again - are recommendation engines missing a trick and getting to caught up in the hype surrounding social networks whereas the answer may lay in the science of music.
It’s simple – we can deconstruct music to its basic elements. By breaking down the components in the appreciation of music; individual notes, beat, melody, lyrics, tempo, repetition, verse, chorus, instruments, time sequence, environmental sounds etc we are putting a song under the microscope to construct a mathematical formula which describes the track. We can extract all this information from the waveform. We can extract not only the tangible components but also the emotion? Researchers have attempted this before but until now we haven’t had the algorithms and computational powers to execute it. We do now.
‘Personality is a collection of emotional, thought and behavioural patterns unique to a person’ (Wikipedia). It’s about ‘that face that we put on in different situations’.
If we can then devise a formula to define our individual musical preferences at a point in time, taking in to account any environmental or emotional factors, then surely we make more accurate recommendations. Not all the elements that make up our preferences are tangible and ‘image’ or ‘the cool factor’ are difficult to define but we should still be able to get more relevant results than recommendations based solely on social communities.
This disjointed rambling has holes galore, but I suppose where I’m going with this is to raise the issue of Collective intelligence, otherwise referred to as Wisdom of the Masses. How much intelligence is there really in using social networks to make recommendations?
There are many critics of the Wisdom of Masses (WOM) theory. There are times when groups make better decisions than individuals but this requires certain conditions to be met. When the decision is dependent on your personality then group decisions are inevitably inferior. In such an emotive subject as music the WOM ignores my individuality.
These are early days and hopefully the intelligence will follow as we embrace the uniqueness of the human personality and include the human psyche in constructing more relevant recommendations.
The next evolution should begin when we stand up and say ‘what the hell happened to my individuality?’
In the meantime I’ll leave my recommendations to the expert reviews that I trust and can apply my own judgement to based on my personal preferences. If only someone could have bottled John Peel’s musical mind and produced a JP Recommendation Engine!
Recent Comments