Iterative Ensemble Classification for Relational Data: A Case Study of Semantic Web Services

For the classification of relational data, iterative algorithms that feed back predicted labels of associated objects have been used. In this paper we show two extensions to existing approaches. First, we propose to use two separate classifiers for the intrinsic and the relational (extrinsic) attributes and vote their predictions. Second, we introduce a new way of exploiting the relational structure. When the extrinsic attributes alone are not sufficient to make a prediction, we train specialised classifiers on the intrinsic features and use the extrinsic features as a selector. We apply these techniques to the task of semi-automated Web Service annotation, a task with a rich relational structure.