Domain-specific Information Retrieval system with a correspondence graph

This paper describes different existing solutions and proposes a new approach for information retrieval with request specified in natural language within a specific domain and in a multilingual context. The experimental platform employed was the SIDIS- Enterprise car-diagnosis System of Siemens AG (Germany). The paper proposes a new methodology to retrieve car failures symptoms and is based on a correspondence graph. This methodology is more based on perception than on similarity computation between request and symptoms. A comparison study between this approach and usual retrieval methods (e.g. term frequency based) provide promising results.