Qutaber: task-based exploratory data analysis with enriched context awareness

Abstract

Exploratory data analysis (EDA) has emerged as a critical tool for users to gain deep insights into data and unearth hidden patterns. The integration of recommendation algorithms has enhanced its capabilities and further popularized its utilization. Most recommendation-based EDA methods concentrate on the extraction of pivotal insights from datasets, and the taxonomy of these insights is well-established. However, the support for further analytical endeavors to expand these initial findings remains constrained, as evidenced by the restricted scope of analytical intents that are tailored to specific scenarios. Moreover, these systems often lack sufficient context-awareness capabilities, failing to equip users with the necessary tools for a thorough exploration of extensive recommendations. To address these limitations, we introduce Qutaber, a task-based EDA system with enriched context-awareness. We first summarize six core analytical tasks tailored for EDA scenarios through literature reviews and expert interviews. Then, Qutaber integrates the use of small multiples, enhanced with a multi-metric re-ranking function, to enable a thorough and efficient examination of expanded charts pertaining to various analytical tasks. Furthermore, a machine learning method is leveraged to characterize the semantic features of these charts for a holistic landscape of recommended charts. Finally, a case study using a real-world dataset demonstrates Qutaber’s practical application, followed by a user study to further evaluate the usability of the proposed techniques. Our findings illustrate that Qutaber facilitates an effective and context-rich EDA experience for users.

Graphic abstract