Jonathan Vi Perrie, Master’s candidate
David R. Cheriton School of Computer Science
Over the years, hit song science has been a controversial topic within music information retrieval (MIR). Researchers have debated whether an unbiased dataset can be constructed and what it means to successfully model song performance. Often classes for modelling are derived from one component of song performance, like for example, a song's peak position on some chart.
We aim to develop target variables for modelling song performance as trajectory patterns that consider both a song's lasting power and its listener reach. We model our target variables over various datasets using a wide array of features across different domains, which include metadata, audio, and lyric features. We found that the metadata features, which act as baseline song attributes, oftentimes have the most power in distinguishing our dataset classes.
We also observed that the two chart dimensions from our target variables sometimes carried different information, which could only be represented with more complex target variables. Along with our model analysis, we also carried out a re-implementation of a related study by Askin & Mauskapf [1] and considered different applications of our data using methods from time series analysis.