The task in the VideoLectures case study is to develop a software component that will aid the VideoLectures editors in categorizing recorded lectures (i.e. ontology population). This functionality is required due to the rapid growth of the number of hosted lectures as well as due to the fact that the categorization taxonomy is rather fine-grained (200 categories and growing). In addition to aiding the categorization of new lectures, the software will also be used for re-categorization and additional categorization of lectures already categorized.

We will show that we were successful in our task as the categorizer is highly accurate – it achieves accuracies that stretch 12–20% above the baseline – and highly robust in terms of missing data. The latter means that a lecture might be missing textual annotations (such as the description and slide titles) but is still categorized correctly. Furthermore, the categorizer has been successfully integrated into the VideoLectures Web site. Categorization suggestions (termed "quick links") are provided to the author in the categorization panel.