Sub Node Extraction with Tree Based Wrappers

String based as well as tree based methods have been used to learn wrappers for extraction from semi-structured documents (e.g., HTML documents). Previous work has shown that tree based approaches perform better while needing less examples than string based approaches. A disadvantage is that they can only extract complete text nodes, whereas string based approaches can extract within text nodes. This paper proposes a hybrid approach that combines the advantages of both systems and compares it experimentally with a string based approach on some sub node extraction tasks.