Modelling music selection in everyday life with applications for psychology-informed music recommender systems

Music is a highly functional and utilitarian resource. It enables people to regulate emotions,
reduce distractions, stimulate physical action, and connect with others. However, with
technologically facilitated ubiquitous listening now commonplace, new problems have
emerged. The main problem is that of choice: how, given millions of songs to choose from,
should providers curate listening experiences? To resolve this, many online platforms employ
recommender systems, and there have been concerted efforts to orientate these systems in such
a way that they are responsive to the short-term, dynamic needs of listeners in everyday
situations. However, there is increasing scrutiny around the impact of automated recommender
systems in terms of interpretability and data usage. To this end, researchers have begun
exploring ways of integrating knowledge about user behaviours into the recommendation
process, rather than through purely data-driven approaches.

This thesis aims to bridge these strands of intrigue by exploring an approach to generating
situationally determined recommendations, based on an understanding of how and why
contextual factors influence music selection in everyday life. This is achieved through three
studies, in which contexts, functions, and content of listeners’ music selections are triangulated
to make inferences and estimates of situationally congruent musical characteristics. Firstly, a
psychometric structure of the functions of music listening is generated. Secondly, this is
triangulated with contextual factors and audio features of music selection. Finally, this is
supplemented with an exploratory approach to generating recommendations through the
explanatory model. These three studies result in both: a preliminary model of goal-orientated
music listening that can be deployed by recommender procedures; and provides an exemplar
methodology of how to construct behavioural models that can drive such systems. This thesis
therefore holds relevance to both psychological research and those interested in music curation
techniques.