Pharos: A Transparent and Steerable Visualization Recommendation System

Abstract

We propose Pharos, a novel visualization recommendation system that makes use of several provenance data sources and multi-perspective overviews to boost the transparency and steerability of the recommendation engine. Pharos helps a user understand recommendation contexts along with the user’s analysis progress in three complementary overviews. Pharos also serves categorized and scalable recommendations, either expanding or narrowing a user’s analysis scope. Based on provenance data and explicit user annotations (i.e., bookmarked or excluded visualizations), Pharos dynamically updates a recommendation list. According to the provided context, a user can steer the recommendation direction by filtering recommended candidates and rearranging them via the weight controller of the similarity measure on the recommendation engine. We showed how Pharos helped users understand and steer visualization recommendations through two comparative user studies.