Abstract
Researchers have developed Interactive Visual Facets (IVF), a new interface framework designed to make exploratory search—where users lack precise queries—feel more fluid. By introducing a "filter-swipe" technique, the tool allows users to rapidly preview intersections between different data categories without the jarring layout shifts of traditional hard filtering. Evaluations across email, social media, and clinical datasets demonstrate that IVF supports serendipitous discovery while maintaining performance comparable to traditional systems.
Introduction
Standard search engines are optimized for "lookup" tasks but struggle with exploratory search, where goals are ill-defined and users must learn about the data as they go. Traditional faceted search often suffers from two primary issues: a loss of context, where filtering hides the broader data landscape, and high interaction costs, where clicking filters on and off to test hypotheses becomes tedious. The IVF framework aims to achieve fluidity through smooth interactions and responsive graphics that minimize the cognitive load of switching between search criteria.
Methods and Interface Design
The IVF tool is built on two core design requirements: providing continuous contextual information (DR1) and supporting rapid transitions (DR2). The interface features coordinated facets, including a linear timeline facet where dots represent individual items and categorical tags for metadata like contacts or keywords.
A core novelty of the system is the "filter-swipe." Instead of clicking checkboxes to filter, users can grab a categorical tag and drag it across the timeline. As the tag passes over different time bars, the interface instantly highlights items at that intersection and dynamically updates other facets to summarize those specific items. This allows for "tentative" actions—low-stakes exploration that returns to the previous state upon releasing the mouse, avoiding the frustration of "zero-hit" filters.
Evaluation and Key Results
The tool was evaluated across three distinct domains: the Enron email corpus (~2,500 items), Twitter data for serendipitous discovery, and AML patient records for scientific analysis.
Key findings include:
- Comparable Performance: IVF showed speed and success rates comparable to traditional query-based baselines, suggesting that added visual complexity does not hinder simple tasks.
- User Preference: Participants significantly favored "selection without filtering," which keeps the global context visible while highlighting specific subsets.
- Discovery via Hovering: 44% of all suggestion usages were triggered by hovering over the timeline, confirming that providing temporal context successfully helps users discover relevant search terms they might not have otherwise considered.
Limitations and Future Work
While effective for thousands of items, the "dot-per-item" visualization may struggle with massive datasets (millions of rows) without further abstraction. Future work will explore semantic zooming to aggregate data across larger time scales and address "interaction fatigue" caused by sustained mouse-dragging. The authors also suggest that position-based encoding might be more effective than the current color saturation methods for human perception of order.
Summary updated