Usefulness of quality click-through data for training
Craig Macdonald, Iadh Ounis · 2009
Modern Information Retrieval (IR) systems often employ document weighting models with many parameters that require to be appropriately set for effective retrieval performance. To obtain these parameter settings, quality training is usually required, where assessors have manually labelled the relevance of retrieved items for many queries. In this work, we examine the usefulness of high-quality click-through data for training an IR system, on searching the .gov vertical domain of the Web. We find that, compared to training using relevance judgements created using human assessors, the click-through trained settings are as good and occasionally better.