A visual and interactive data exploration method for large data sets and clustering

International audienceWe present in this paper a new method for the visual exploration of large data sets with up to one million of objects. We highlight some limitations of the existing visual methods in this context. Our approach is based on previous systems like Vibe, Sqwid or Radviz which have been used in information retrieval: several data called points of interest (POIs) are placed on a circle. The remaining large amount of data is displayed within the circle at locations which depend on the similarity between the data and the POIs. Several interactions with the user are possible and ease the exploration of the data. We highlight the visual and computational properties of this representation: it displays the similarities between data in a linear time, it allows the user to explore the data set and to obtain useful information. We show how it can be applied to standard 'small' databases, either benchmarks or real world data. Then we provide results on several large, real or artificial, data sets with up to one million data. We describe then both the successes and limits of our method