Document Similarity Example

Each symbol represents a publication, coloured by its author (see legend below). Publications are clustered by their vocabulary similarity - the more similar the publication pair, the closer they attempt to be in the graph. Each publication is linked to its 6 most similar publications.

Instructions

• Drag publications with the mouse to see their stability. Clicking on a publication reveals a book summary and word cloud (word cloud is for Dickens only in this demonstrator)
• Holding shift key down and dragging zooms and pans across graph - left mouse button drag to zoom, right mouse button drag to pan.
• 'R' key restores a zoomed/panned view to the original overview.
• 'L' key to toggle linking lines on and off. Thick lines join each document to its most similar publication; thin lines to its five next most similar publications.
• Space key to toggle movement on and off. When movement is off, publications can be dragged without them returning to their original positions.
• Up and Down arrows (held down for a second or two) increase or decrease the minimum distance of separation between nodes. Can be useful for unclustering closely associated publications.

Changes since previous version

• Now uses database created 24th Nov with improved word counts
• Improved formatting of book summary window (ordinal rank and comma separated large numbers)
  
  
• Improved labelling (now spans multiple lines)
• Improved colouring of books by author (maximises colour contrast between labels)
• Now uses top six most similar titles to improve clustering appearance
• Links between nodes can now be shown with colour and shape indicating tension
• Slightly improved performance

You can also download this as a stand-alone application, which runs slightly faster and at higher quality:

• Windows zip file
• MacOSX zip file
• Linux zip file