InforadarML: A multi-lingual information discovery tool exploiting automatic document categorization

In this thesis we present the design of Inforadar ML a multilingual extension for Inforadar, the first search engine supporting automatically generated visual query hierarchies. The central hypothesis of this work is that retrieval effectiveness of multilingual documents can be improved by simultaneously providing the search engine human-translated multilingual queries identified with their source languages. Inforadar ML enhances Inforadar by adding support for multilingual queries and document collections. We have developed a test collection of multilingual web documents, queries and human-generated relevance judgments freely available to the scientific community. We have conducted precision/recall experiments to assess the effectiveness of three document ranking algorithms. Our experiments suggest that automatic ranking of multilingual results sets even using naive ranking algorithms yields results comparable to independent manual sifting of separate results from equivalent queries in different languages. We feel that more efficient multilingual ranking algorithms can provide more valuable response to specific multilingual information needs.