Exploiting Keyword Structure for Domain-Specific Retrieval

Structured elements, such as manually assigned keywords or key-phrases in scientific collections, are pervasive in digital libraries. Special dictionaries or thesauri for the meta-information are not always available. Our strategy is to compute the similarity of keywords based on their occurrence in the collection. The resulting keyword space is brought to bear on a variety of tasks. Combined with an information retrieval system, we can recover keywords for queries, and thus provide a technique can be used for automatic classification. Moreover, it can be used to rerank retrieved documents, leading to a significant improvement of retrieval effectiveness in domain-specific collections. Experimental evaluation is done on the German GIRT and French Amaryllis collections, using the test-suite of the Cross-Language Evaluation Forum (CLEF).