KeisukeKamataki writeup of Brin 1999

This is a review of Brin_1999_extracting_patterns_and_relations_from_the_world_wide_web by user:KeisukeKamataki.

Summary: They propose a technique to detect relations of entities from web called DIPRE. The key point of the approach is generating/expanding patterns of relation occurrences according to the observation of data. Since the initial set of patterns is based on the matching with manually defined samples, it is very small. But the patterns can be easily extended by trying to make the patterns as general as possible (but not to be too general) with simple rule.

I like: Though the approach itself looks very simple and naive, it could be a kind of powerful for their problem because web text is often messy and thus hard to be handled with too specific approach. Rather than that, keeping the framework simple/flexible and making use of data could be more robust.

Not clear: It would be better if we can know more about experimental result and error analysis deeply. The analysis doesn't seem to be enough.