Rbosaghz Han writeup of Bunescu and Mooney

This is a review of Bunescu_2007_learning_to_extract_relations_from_the_web_using_minimal_supervision by user:Rbosaghz.

Many of the approaches discussed in class so far have been fully supervised, meaning lots of labeled training data is required. This paper aims at performing the relation extraction task with as little labeled data as they can get away with. However, it is important to note that they need the web as an additional training corpus. It's worth mentioning that the Read-The-Web system does something very similar.

By using only a handful of data to characterize a particular relation (e.g. Google !ACQUIRED Yahoo, Google AQUIRED YouTube, etc), they can mine the web for mentions of the relation and extract more instances of the binary relation.

I liked this paper because it taught me more about how Tom Mitchell's Read-The-Web system can work, and because they can get away with so little supervision.