Converse
Converse is a search engine based on conceptual vectors. A request is translated into one (or several) vector(s) which are compared to article indexes. An index is also by itself a conceptual vector. The classical angular distance (the arcos of the similary defined as the scalar product) is used as a mean of comparison between the request vector and the article vectors.
Playground
It is possible to browse hrough documents. Request are from French only words. Several ontologies are available : French terms (term) and geography (geo). Enter any number of words and the the number of desired articles n:
Text bases
Le Monde 2200 : Articles are from the newspaper Le Monde. The set is of roughly 2200 articles from 1999 to 2001. Not all subjects are represented.
Several ontologies
A given text could be indexed agains several ontology, i.e inside several vector spaces. A request is translated (when possible) into a conceptual vector for each vector space. The n liste of candiadtes retrieved are then collated to formulate a unique result list to the user.
Request vector
Several methods can be used to compute the request vector from the request terms. Generaly a lexicial desambiguation is performed through the various terms. For instance : bank+river will not produce the same vector than bank+money.