Characterizing User Behavior in a Music Navigation Application with Real-time Feedback

In a previous work, we proposed a direction-based navigation framework for large media collections. We implemented Mixtape (www.projectmixtape.org), a web application with a simple interface that allows users to navigate in a large music collection using online feedback and a geometric similarity space. In this work, we analyze user-generated data during an experiment on Mixtape. We collected all user actions on Mixtape over a course of two weeks, resulting in the participation of 800 users, generating a total of over 2 000 navigation sessions. We analyzed the trajectories that users followed within the music similarity space, and collected implicit and explicit feedback about user satisfaction. In particular, we compared geometric relationships, such as distance and direction similarity, between consecutive songs within individual and between different navigation trajectories. The navigation technique called Vector, which uses the concept of directions in this space of similarities, was able to keep the analyzed users for a longer time in the application. This may indicate a preference among users for navigating in their songs using the concept of directions.