Information Terra turns document collections into explorable narrative globes

Information Terra reimagines document exploration by placing an entire corpus onto a 3D globe, where the user chooses two endpoint documents to serve as the North and South Poles. The prime meridian is the great-circle geodesic between these endpoints on the embedding hypersphere, so latitude directly encodes narrative progress from one pole to the other, while longitude captures thematic deviation. Land features are derived from document density via kernel density estimation and automatically labeled by dominant themes, creating a visual terrain of ideas.

The method also constructs a narrative trail based on the underlying coherence graph, constrained to be monotone in geodesic progress, producing a clear storyline through the corpus. In their demo on a 540-article dataset about Cuban protests, the authors map a trajectory from Obama's 2016 visit to the 2021 International Aid during protests, showing how Information Terra reveals semantic shifts over time. The globe metaphor allows users to rotate the view, read antipodal relationships, and intuitively grasp the narrative structure. Accepted at IEEE VIS 2026, this work offers a novel way to visually summarize large document collections without losing semantic grounding.