Australia's Discourse Explorer

Abstract storylines moving through a spatial data field

This project helped researchers trace multi-strand storylines across opinion articles and parliamentary speeches from 2006 to 2022.

Scale and approach

Worked with a corpus of 180 million words covering 23 disadvantaged groups and 33 social issues.
Built the pipeline behind two interactive metro-map variants: entity extraction, summarisation, temporal embeddings, clustering, and a narrative directed acyclic graph.
Pilot-tested the maps through story-reconstruction tasks to identify where the approach succeeded and where it broke down.

The work became Discourse Lines at the IEEE VIS 2023 NLP+VIS workshop and directly motivated the TTNG model.