Recommendation Systems and Machine Learning: Mapping the User Experience

摘要

Human-algorithm interaction emerges as the new frontier of studies involving interaction design and interface ergonomics. This paper aims to discuss the effectiveness and communicability of streaming media recommendation systems, based on machine learning, from users' mental model point of view. We examined the content consumption practices on the Netflix platform, identifying some sensitive aspects of the interaction with recommendation algorithms. One-on-one semi-structured interviews were applied to a sample of students from three different universities in Rio de Janeiro, Brazil. We realised that interviewees had not correctly understood how the system works and have not formed an adequate mental model about tracked data and how it is processed to create personalised lists. Another issue concerns data privacy: Users have revealed a suspicion concerning algorithms and what might happen to usage data, not only in the Netflix platform but also in other services that use algorithm-based recommendation. Interviewees' responses suggested that there may be communication failures, and UX designers should strive to make it visible how the system tracks and processes user interaction data.