LIP6 2000/005:
THÈSE de DOCTORAT de l'UNIVERSITÉ PARIS 6 LIP6 /
LIP6
research reports
213 pages - Janvier/January 2000 -
French document.
Get it : 682 Ko /Kb
Contact : par mail / e-mail
Thème/Team: Apprentissage et Acquisition de Connaissances
Titre français : Construction et Exploitation de Réseaux Sémantiques Flous pour l'Extraction d'Information Pertinente : Le système RELIEFS
Titre anglais : Building and Using Fuzzy Semantic Networks for Extraction of Relevant Information : The RELIEFS System
Abstract : Relevance is a central notion for Artificial Intelligence. From a cognitive point of view, it is also a key concept and researchers use it to explain the way we interact with our environment and the way we communicate. At present, the proliferationion of digital data and in particular, the great expansion of Internet reinforce the need for a concrete definition of this notion. Information is present, but users have to extract it from gigantic databases and the recurrent question is: which is the relevant information ?
In this document, we adopt a pluridisciplinarity approach to this notion which is oriented toward cognitive sciences. The noticeable diversity of problems and solutions suggested in different disciplines is explained and we put forward a solution which allows us to define " relevance " more globally. This solution is based on a construction and exploitation of an associative network. In this network, the nodes represent properties or conjunctions of properties, which are possibly fuzzy (in the fuzzy set theory sense) and connexions represent implications beetween these properties. We define a method for building a "fuzzy semantic network" from a set of observations. We also suggest a method for using this network in order to select information relative to a user's request.
The system which implements these methods (called RELIEFS), has been used for analysing different kinds of data (epidemics, sensorial evaluations). Lastly, a study shows how the system can be used in a Man-Machine communication context.
Key-words : Relevance, semantic networks, fuzzy sets, data-mining
Publications internes LIP6 2000 / LIP6 research reports 2000