Concept drift and the importance of examples

For many learning tasks where data is collected over an extended period of time, its underlying distribution is likely to change. A typical example is information filtering, i.e. the adaptive classification of documents with respect to a particular user interest. Both the interest of the user and the document content change over time. A filtering system should be able to adapt to such concept changes. Examples may be important for different reasons. In case of a drifting concept, the importance of an example obviously depends on its age. If a user is interested in several topics, these may be of different importance to her/him. Hence the importance of an example is influenced by the topic it belongs to. Of course these two effects may cumulate. In this paper we model the importance of an example by weighting its importance for the final decision function. This paper investigates how to handle these two effects with support vector machines extending the approach of [12], which showed that drifting concepts can be learned effectively and efficiently with little parameterization. Several approaches addressing the different effects are compared in experiments on real-world text data.