A demonstration: Navigation by meaning

Below you can find a visualisation of a set of documents from the web, as seen through the eyes of M-Brain's multilingual meaning-based technology. Each document is mapped to a very high-dimensional concept space. In this particular example, a medium amount of generalisation is performed for each mapping. This particular dataset is a random sample from a multilingual web slice containing the named entity "Barack Obama".

• Pointing at one of the Prominent concepts highlights related documents from the visualisation
• Pointing at a particular document displays the concepts it maps to under "Document concepts"
• Clicking a document opens the link

More on the visualisation

Visualisation layout is calculated according to conceptual distance between each document pair. Similar documents are grouped together and different ones are placed apart. X/Y axis bears no specific meaning, only the position of a document relative to other documents. Closeness means similarity.

Note that while the dimensionality reduction from a very high-dimensional space to two dimensions tries to preserve original distances as far as possible, some sacrifices typically occur and documents only remotely related end up close to each other, because they are even more distant from everything else. Also, sometimes imperfect boilerplate removal for a randomly found site leads the mapping astray. In the M-Brain process human intelligence will step in at this point to smooth out the rough edges for a polished finish.