Accurately and Reliably Extracting Data from the Web: A Machine Learning Approach