A Random Walk Approach To Sampling Hidden Databases

A large part of the data on the World Wide Web is hidden behind form-like interfaces. These interfaces interact with a hidden back-end database to provide answers to user queries. Generating a uniform random sample of this hidden database by using only the publicly available interface gives us access to the underlying data distribution. In this thesis, we propose a random walk scheme over the query space provided by the interface to sample such databases. We discuss variants where the query space is visualized as a fixed and random ordering of attributes. We also propose techniques to further improve the sample quality by using a probabilistic rejection based approach and conduct extensive experiments to illustrate the accuracy and efficiency of our techniques.