Sightis - Categorization Engine
Sightis® ... is a categorization engine based on Sight’Up Artificial Intelligence technology. It is self-learning and works on the principle of "Machin-Learning". Its main fields of application are e-commerce and e-mails management.
Innovative Technology : Sightis is founded on a unique combination of modelling technologies and new research coming from the biology world. Sightis extracts the numerical “signature” from a document and codes the unique "signature" of the principal concepts, thus allowing to a lots of operations to be automatically carried out.
The documents "are initially coded" according to their context in the form of chromosomes then the engine analyzes this data by using several parallel algorithmic chains. The final result is given by the combination of votes weighted by their relevance to the algorithms in the context of the corpus. This process is supervised by an independent system based on a "genetic" approach. This generates the following advantages :
- Reduction of learning effort, just a few tens of examples per category are enough to have quite a satisfactory result.
- Sightis is insensitive to the languages used (even Asian languages), to the environment or field of activity.
Categorization : Sightis assigns one or more categories to a new document according to its learning data.
Benchmarks carried out by our customers showed that Sightis is 4 times more powerful than traditional "lexical" or statistical approaches (eg bayesian).
There are currently three operating modes. The choice of mode depends on the expected precision-recall curves.
Sightis technology is already integrated into Sight’Up’s e-mail management software : MailRelation, which daily manages several thousand e-mails of more than 200 categories.
Sightis has been entirely developed in C in order to provide a large processing capacity and universal portability. It is supplied with its integration APIs and works in XML format and Unicode.
More details on Sightis
