Active learning for hierarchical wrapper induction
Ion Muslea, Steve Minton, Craig A. Knoblock · 1999
Information mediators that allow users to integrate data from several Web sources rely on wrappers that extract the relevant data from the Web documents. Wrappers turn col-lections of Web pages into database-like tables by applying a set of extraction rules to each individual document. Even though the extraction rules can be written by humans, this is undesirable because the process is tedious, time consuming, and requires a high level of expertise. As an alternative to manually writing extraction rules, we created STALKER (Muslea, Minton, & Knoblock 1999), which is a wrapper induction algorithm that learns high-accuracy extraction rules. The major novelty introduced by STALKER is the concept of hierarchical wrapper induction: the extraction of the relevant data is performed in a hierar-chical manner based on the embedded catalog tree (ECT), which is a user-provided description of the information to be extracted. Consider the sample document