A quantitative within-subject experiment study on the effect of different filter types on the consumer utility of music recommendation
Nowadays a large portion of music consumption is done via streaming. Platforms that enable music streaming have a hand in what is recommended and through which method, and therefore have a hand in what is consumed. This in turn results in undemocratic recommendation methods. Consumers are subsequently left largely unaware of the processes behind their song recommendations on music streaming platforms. This research has aimed to dissect the processes behind song recommendation via separating Content Based Filtering (CBF) and Collaborative Filtering (CF) mechanisms. Dissecting these mechanisms was necessary to assess whether the different filter methods resulted in different levels of consumer utility and to what extent. Consumer utility was assessed via looking at respondents’ likelihood to listen to filter recommended songs present in the within-subject experiment survey, and whether these songs fit within their taste. In order to test this a total of 302 participants took part in the survey of this study. Within the survey the respondents could pick one out of five music genre paths to go down, within these paths they came across five songs, two recommended via CBF, two recommended via CF, and one Randomly recommended. The CBF recommendations were made via attributing and cross referencing meta-data the those respective songs. CF recommendations were created via utilising prewritten code for the Million Song Dataset. Random recommended songs were recommended via generating a random number and corresponding that number with music chart positions. Results indicate that there is a difference in consumer utility between CBF and CF recommended songs. In fact, this research concludes that in general consumers have a preference towards CBF recommended songs. Alongside this the study found no significant results supporting the hypotheses that different listener types, heavy or light, would result in a preference towards songs recommended by a particular filter type. The higher consumer utility from CBF recommendations would infer that within the field of music, experts and curators are still much needed on the production side and much appreciated on the consumption side.
|Keywords||media, culture, society, Music recommendation, Content based filtering, Collaborative filtering, Algorithms, Consumer utility|
|Thesis Advisor||J. Ferreira Gonçalves|
|Series||Media, Culture & Society|
Z. Voyle. (2019, July 15). ‘Any suggestions?’. Media, Culture & Society. Retrieved from http://hdl.handle.net/2105/49333