Note: This bibliographic page is archived and will no longer be updated. For an up-to-date list of publications from the Music Technology Group see the Publications list .

Towards Time-aware Contextual Music Recommendation: An Exploration of Temporal Patterns of Music Listening Using Circular Statistics

Title Towards Time-aware Contextual Music Recommendation: An Exploration of Temporal Patterns of Music Listening Using Circular Statistics
Publication Type Master Thesis
Year of Publication 2010
Authors Resa, Z.
preprint/postprint document static/media/Resa-Zuriñe-Master-Thesis-2010.pdf
Abstract

Music is present in many situations of our daily life, it's a way of showing our personality traits, a way of reinforcing our social identity or even a way of inducing an emotional state or mood. These factors that are highly linked to our music consumptions and preferences, are highly changeable, they do vary over time and they can be assessed to be linked to physiological or natural rhythms. For this reason, it seems reasonable to explore the influence of these rhythms in music listening activity. This thesis is an attempt to explore temporal patterns in relation with the time of the day or the day of the week which can be observed when tracking specifi c artists or music genres. The final goal is to characterize music listeners based on the information extracted regarding the time of reproducing each artist and genre, assessing temporal patterns and thus, cyclic rhythms.

So as to detect these temporal patterns or rhythms, a circular statistical analysis is performed over a data-set containing the listening habits of almost a thousand users of an online radio - music recommender system. This analytical approach o ffers a wide range of strategies to examine "circular data", data where the period of measurement is rotationally invariant(as the daily hours which range form 0 to 24, 24 being the same as 0). This way temporal patterns are identi fied regarding the time of the day or the day of the week (respectively, a period of 24 hours and of 7 days). We show that for certain users, respectively for artist and genres, 20% and 40% of the found temporal patterns can be used to predict music listening selections with above-chance accuracy.

This fi nding has as a possible application personalized playlist generation and music recommendation based on providing user-specifi c suggestions at the "right" moment.