More and more websites are using collaborative filtering recommenders to personalise their goods and services for you.  For instance, Amazon’s “Customers Who Bought This Item Also Bought,” uses collaborative filtering technology to let you know about other products that might be of interest to you.

Figure 1.  An example of collaborative filtering recommendation demonstrated on Amazon.com

Simply put, collaborative filtering recommenders allow a website to recommend stuff based on how similar your browsing behaviour is to that of other users.  These recommenders will often rely on some correlation threshold value to determine whether you do or do not share mutual interests with various other users.  A really good example of music recommendations based on collaborative filtering would be last.fm (www.last.fm)

One issue with collaborative recommenders is a result of a user's divergent goals - different goals that a user might have when using a particular interface.  For instance, say that a given user is a keen fan classical music (a Mozart aficionado), but regularly listens to Lady Gaga (and other contemporary pop music) when with certain friends because that user knows that Lady Gaga-type music will facilitate a belongingness with those friends.  When that user is alone, s/he wants to listen to classical music and be recommended only this music without having to sort through recommendations based on when Lady Gaga has been selected.

A solution that has been devised to help users with recommender noise that results from divergent goals has been to include product information (e.g., classical versus contemporary pop music) when the recommender filters your recommendations.  That way, our example user only gets recommendations for classical music when listening to Mozart.  Recommenders that include content information with collaborative filtering are called hybrid recommenders.

Another issue for any recommender system is what to do when a new user or new item comes along, commonly known as cold start. For the last four years, I’ve been looking at the relation between people’s music preferences and their personalities, which could be used as an alternative way to help resolve the cold start problem and improve music recommenders.  An associate of mine at Cambridge University, Dr. Jason Rentfrow, does a great job in describing the music preferences and personality research (see http://www.youtube.com/watch?v=29-xYiOOc8w).

Researchers like Dr. Rentfrow have identified relations between the genre of music that people listen to (e.g., rap or jazz) and personality characteristics that those people generally have (e.g., extroversion or openness to experience).  Still, genres can be really vague.  I mean, you and I might both love rock music, but are you going to necessarily love the same rock music as I do?  Instead, to help recommenders to their job, I identified audio features prominent in certain music genres and link these to personality characteristics.  So, instead of saying that extroverts like rap music, I say that extroverts really like music that has a lot of beats that happen quickly together… constantly.  This relation might apply mostly to rap music, but might also apply to certain rock songs, electronica songs, you name it.  Conversely, my research suggests that introverts like music that has few beats in the music, which is typical of classical music, but again, is not exclusive to classical.  As a result, identifying the relation between personality and music preference toward specific audio features can help improve both issues described above by identifying and sorting music according to more precise and objective audio features.

In sum, the work that I have described is still very new and there are a lot of challenges to see through before recommenders can truly become 'personalised' by learning and understanding users’ personality, but there are a lot of opportunities that may result from this type of personalisation as well.

So tell me, do you think your musical tastes describe your personality?  Have you made personality judgements about a person because of the music they listen to?

Also, do you think that there are similar aspects common in most or all of the music you listen to?  Or, do you tend to have a favourite instrument that you like when it’s played in a song?

Finally, what do you think about music recommendation based on personality?

Your comments (positive or negative) are welcome.