Semantic vectors at Shinshu University

Nakano, Okamoto and Maruyama laboratories at Shinshu University have worked on 3D representation for document retrieval on the Internet. They used semantic vectors defined as normalized vectors whose coordinates are the weights of the keywords used to define the basis of the vector space. These weights are computed from absolute and relative frequencies of the keywords in the document considered, compared with other documents of the database.

Those semantic vectors are used to represent:

• the semantic content of a document;
• a user's viewpoint.

They are also used to easily define the similarity between documents as the innerproduct of their semantic vectors. Simple technics (innerproducts, projections...) can therefore have a meaning on a semantic point of view, when they are applied to semantic vectors.