The Songs of Our Past

Advancements in technology have resulted in unique changes in the
way people interact with music today: Small, portable devices allow
listening to it everywhere and provide access to thousands or, via
streaming, even millions of songs. In addition, all played tracks
can be logged with an accuracy down to the second. So far, these
music listening histories are mostly used for music recommendation
and hidden from their actual creators. But people may also benefit
from this data more directly: as memory extensions that allow
retrieving the name of a title, for rediscovering old favorites and
reflecting about their lives. Additionally, listening histories can
be representations of the implicit relationships between musical
items. In this thesis, I discuss the contents of these listening
histories and present software tools that give their owners the
chance to work with them. As a first approach to understanding the
patterns contained in listening histories I give an overview of the
relevant literature from musicology, human-computer-interaction and
music information retrieval. This literature review identifies the
context as a main influence for listening: from the musical and
temporal to the demographical and social. I then discuss music
listening histories as digital memory extensions and a part of
lifelogging data. Based on this notion, I present what an ideal
listening history would look like and how close the real-world
implementations come. I also derive a design space, centered around
time, items and listeners, for this specific type of data and
shortcomings of the real-world data regarding the previously
identified contextual factors. The main part of this dissertation
describes the design, implementation and evaluation of
visualizations for listening histories. The first set of
visualizations presents listening histories in the context of
lifelogging, to allow analysing one’s behavior and reminiscing.
These casual information visualizations vary in complexity and
purpose. The second set is more concerned with the musical context
and the idea that listening histories also represent relationships
between musical items. I present approaches for improving music
recommendation through interaction and integrating listening
histories in regular media players. The main contributions of this
thesis to HCI and information visualization are: First, a deeper
understanding of relevant aspects and important patterns that make
a person’s listening special and unique. Second, visualization
prototypes and a design space of listening history visualizations
that show approaches how to work with temporal personal data in a
lifelogging context. Third, ways to improve recommender systems and
existing software through the notion of seeing relationships
between musical items in listening histories. Finally, as a
meta-contribution, the casual approach of all visualizations also
helps in providing non-experts with access to their own data, a
future challenge for researchers and practitioners alike.