Adaptive User-Context Based Search Agents

A major reason of the much noise still encountered in the answers of search engine, is due to the fact that human queries, per se, often are ambiguous. In human communication such ambiguities are usually removed using knowledge of the relevant context; in contrast, every time the user communicates with the engine, it behaves, as was it the first time. The project aims at improving this situation through extracting and utilizing context knowledge derived through user-behavior mining in the usage log, and though other relevant techniques, all in the framework of fuzzy sets, such as fuzzy term association mining, intelligent filtering using adaptive user profiles, improved document indexing through utilizing metadata such as visual emphasizing applied in the text, and efficient, flexible aggregation operators as applied in the matching. The project is supported by the Danish Natural Science Research Council and done in collaboration with University of Granada, Spain, and the Machine Intelligence Institute, USA. (Henrik Legind Larsen)