Logo des Repositoriums

Auflistung nach Autor:in "Gardarin, Georges"

1 - 2 von 2
  • Konferenzbeitrag
    Approaches to feature selection for document categorization
    (Natural language processing and information systems, 2003) Kou, Huaizhong; Gardarin, Georges; Zeitouni, Karina
    One of the problems faced by document categorization is that terms present in the collection of example documents are numerous. From the point of view of coherence between the models used in document categorization, we analyses the frameworks of both k-NN and NB categorization models and feature selection problem. Two algorithms CBA and IBA to feature selection are proposed. The empirical results done with k-NN and NB classifiers show that the coherence between models in the categorization system can bring benefits for performance.
  • Konferenzbeitrag
    SEWISE : An ontology-based web information search engine
    (Natural language processing and information systems, 2003) Gardarin, Georges; Kou, Huaizhong; Zetourni, Karina; Meng, Xiaofeng; Wang, Haiyan
    Since the begin of the 90's, the World Wide Web (WWW) rapidly guides the world into a newly amazing electronic village, where everybody can publish everything in electronic form and find almost all required information. The volume of available information is increasing exponentially in different formats, 80% being text. It remains hard to find interesting information directly from Web sources. SEWISE is an ontology-based Web information system to support Web information description and retrieval. According to domain ontology, SEWISE can map text information from various Web sources into one uniform XML structure and make hidden semantic in text accessible to program. The textual information of interest is automatically extracted by Web Wrappers from various Web sources and then text mining techniques such as categorization and summarization are used to process retrieved text information. Finally, text descriptions are built in XML format that can be directly queried. SEWISE provides support for topic-centric Web information search. The SEWISE prototype is implemented and has been experimented using French financial Web news from several popular sites.