University of Maryland University College
Interactive Data Exploration (IDE) applications typically involve users that aim to discover interesting objects by iteratively executing numerous ad-hoc exploration queries. Therefore, IDE can easily become an extremely labor and resource intensive process. To support these applications, the authors introduce a framework that assists users by automatically navigating them through the data set and allows them to identify relevant objects without formulating data retrieval queries. Their approach relies on user relevance feedback on data samples to model user interests and strategically collects more samples to refine the model while minimizing the user effort. The system leverages decision tree classifiers to generate an effective user model that balances the trade-off between identifying all relevant objects and reducing the size of final returned (relevant and irrelevant) objects.