Collaborative filtering over evolution provenance data for interactive visual data exploration. Issue 95 (January 2021)

• Record Type: Journal Article
• Authors: Ben Lahmar, Houssem; Herschel, Melanie
• Is Part Of: Information systems. Issue 95(2021)
• Publication Date: 2021-01
• Subjects: Visual data exploration -- Provenance -- Recommendations; Database management -- Periodicals; Electronic data processing -- Periodicals; Bases de données -- Gestion -- Périodiques; Informatique -- Périodiques; Database management; Electronic data processing; Periodicals; 005.7
• DOI: 10.1016/j.is.2020.101620
• Languages: English
• ISSNs: 0306-4379

Abstract

Abstract: In interactive visual data exploration, users rely on recommendations on what data to explore next. EVLIN is a system that recommends queries to retrieve these data for the next exploration step, paired with suited visualizations. This paper extends EVLIN by combining its content-based recommendations with recommendations leveraging collaborative filtering to improve the effectiveness of recommendation-based visual data exploration. The recommendations rely on evolution provenance, which tracks users' interactions during interactive visual data exploration. As more users explore a dataset, the evolution provenance of individual user explorations is incrementally integrated into a multi-user graph, for which we present match and merge algorithms. To compute collaborative-filtering recommendations, we present a search algorithm and optimizations to efficiently search queries similar to a current user's query in the multi-user graph and give preference to queries that have been previously explored in an exploration step succeeding those similar queries. Our experimental evaluation studies the efficiency and effectiveness of the solutions proposed in this paper and demonstrates that using the full system with both content-based and collaborative-filtering recommendations enabled allows for effective interactive visual data exploration. Highlights: Proposal of a visual data exploration system with content and collaborative recommendations. Proposal of several merge techniques to aggregate users exploration sessions in a multi-user graph. Proposal of several optimizations to improve collaborative-filtering recommendation computation. Quantitative evaluation of collaborative-filtering recommendations techniques. Qualitative evaluation of users experiences when visually exploring data using our system.