D. Kowald, Elisabeth Lex, Markus Schedl,
"Modeling Artist Preferences for Personalized Music Recommendations"
: Late-Breaking/Demos of the 20th International Society for Music Information Retrieval Conference (ISMIR 2019), 11-2019
Original Titel:
Modeling Artist Preferences for Personalized Music Recommendations
Sprache des Titels:
Englisch
Original Buchtitel:
Late-Breaking/Demos of the 20th International Society for Music Information Retrieval Conference (ISMIR 2019)
Original Kurzfassung:
In our work, we model user listening behavior on the level of music artists to describe a user?s music taste.
Since a user?s music artist preferences may change over time [5], we take temporal drifts of a user?s music
listening habits into consideration. To do so, we utilize the Base-Level Learning (BLL) equation from the
cognitive architecture ACT-R [1] to model music listening habits. The BLL equation accounts for the timedependent decay of item exposure in human memory. It quantifies the usefulness of a piece of information
based on how frequently and how recently it was accessed by a user and models this time-dependent decay
using a power-law distribution [4]. In the present paper, we adopt the BLL equation to model the listening
habits of users in the three groups and predict their music artist preferences. We name our approach BLLu
and demonstrate the efficacy of BLLu using the LFM-1b dataset [6], which contains listening histories
of more than 120,000 Last.fm users, amounting to 1.1 billion individual listening events over nine years:
http://www.cp.jku.at/datasets/LFM-1b/.
Additionally, the dataset contains demographic data such as age and gender as well as a ?mainstreaminess?
factor, which relates a user?s artist preferences to the aggregated preferences of all users (i.e., the mainstream).
Based on this factor, we assign the users in our dataset (a subset of LFM-1b) to one of the three groups: (i)
LowMS, (ii) MedMS, and (iii) HighMS. Thus, the 1000 users with the lowest mainstreaminess are in the
LowMS group, the 1000 users with a mainstreaminess value centered around the median are in the MedMS
group, and the users with the highest values are in the HighMS group.