Visualising Latent Semantic Spaces For Sense-Making Of Natural Language Text

Latent Semantic Analysis is widely used for natural language processing, but is difficult to visualise and interpret. We present an interactive visualisation that enables the interpretation of latent semantic spaces. It combines a multi-dimensional scatterplot diagram with a novel clutter-reduction strategy based on hierarchical clustering. A study with 12 non-expert participants showed that our visualisation was significantly more usable than experimental alternatives, and helped users make better sense of the latent space.