The University of Amsterdam at WebCLEF 2006

Our aim for our participation in WebCLEF 2006 was to investigate the robustness of information retrieval techniques to crosslingual retrieval, such as compact document representations, and query reformulation techniques. Our focus was on the mixed monolingual task. Apart from the proper preprocessing and transformation of various encodings, we did not apply any language-specific techniques. Instead, the target domain meta field was used in some of our runs. A standard combSUM combination using Min-Max normalization was used to combine runs, based on a separate content and title indexes of documents. We found that the combination is effective only for the human generated topics. Query reformulation techniques can be used to improve retrieval performance, as witnessed by our best scoring configuration, however these techniques are not yet beneficial to all different kinds of topics.