Hierarchie

#Hierarchie

A structured display of topics from discussion surrounding the MH-370 disappearance.

Hierarchie provides hierarchical navigation of the different topics or themes from text. These topics are discovered using a technique called Hierarchical Latent Dirichlet Allocation (HLDA), and are represented in the visualization as lists of words.

For this example, we used a corpus of 1600 Tweets and 970 Reddit comments containing the keyword "MH370" in addition to 27 Daily Beast articles returned by a URL filter for any of the key words "malay", "370", "flight", "missing", "hijack", "radar", "pilot", "plane", "airplane", and "wreckage". These documents were collected during the first week of MH-370's disappearance. By exploring the visualization, it's possible to discern different topics and theories relating to the airliner.

Hierarchie was created by the data visualization team for DECISIVE ANALYTICS Corporation. This implementation of a sunburst was based upon Sequences Sunburst by Kerry Rodden and Zoomable Sunburst by Mike Bostock.

Further details can be found in our workshop paper, and we'll be presenting our work at the ACL 2014 Workshop on Interactive Language Learning, Visualization, and Interfaces.

You can view the visualization live here!

Interactivity

Hierarchie updates on mouseover and zooms on click, so it's best viewed on a laptop or desktop.

Known Issues & Suggested Improvements

• Exposing the documents underlying topics
• Search capabilities
• Improving responsiveness of website and visualizations
• Better cross-browser support (Center text appends using foreignObject will not work in IE)
• Needs tests!

License

Hierarchie is covered by the Apache Version 2.0 License. Please see LICENSE.txt for more details.