GONDIM, D.; http://lattes.cnpq.br/1195739884342749; MÉLO, Daniel Gondim Ernesto de.
Resumo:
Often listening to music is not just a hobby or a leisure activity, but a way of reaching a certain emotional or psychological state, or even to better perform an activity, e.g. relax- ing. Understanding users’ preferences when listening to music to perform certain activities is primordial to improve context-aware music recommender. However, until now, studies in this area have focused on identifying general characteristics of music according to certain contexts, that is, they consider that a song, for example, to relax, has the same characteristics regardless of user. Thus, in order to explore possible different preferences and perceptions between users, when creating playlists for relaxation, this work carried out an analysis with approximately 91,000 playlists created by about 8,000 users on the 8tracks and Spotify plat- forms, which are widely used for sharing songs. In this study, we consider a user’s musical preference to be a representative model of the characteristics of the songs chosen by him in his playlists. These characteristics comprise high and low-level information of the songs and were obtained through the use of the Spotify API. High-level information refers to charac- teristics that human beings can recognize when listening to music, such as how danceable a song is, if it is instrumental, how energetic it is, etc. On the other hand, low-level information refers to characteristics of the audio signal, such as the timbre, which is used in this work. The results obtained with the analyses performed suggest that there are great differences in the perceptions of users when creating playlists for relaxation. It was possible to identify the existence of different groups of users, where each group has a specific set of relevant characteristics that define their preferences about relaxing music. Furthermore, inequalities between users were also observed when we compared the difference between their percep- tions of non-relaxing and relaxing music. Moreover, inequalities between users were also observed when we compared the difference between their perceptions of non-relaxing and relaxing music. Hence, these results suggest that there must be a specific treatment for each of these groups of users, for example, when recommending songs for relaxation. In addi- tion to the results obtained, this work contributes to the provision of a new dataset for future research on musical playlists for relaxation.