Nlao writeup of Bunescu 2007

This is a review of Bunescu_2007_learning_to_extract_relations_from_the_web_using_minimal_supervision by user:Nlao.

I enjoy reading this paper. It clearly present problems related to weakly supervised extraction: noise training data, type I & II bias, etc. The method is simple and effective.

[minor points]

- Not like other methods, the SSK-SIL method was trained on only one bag. This makes the comparison of their performances much less meaningful.

- It is interesting to see the result of sentence parsing + full sequence kernel. The authors just assume that parsers cannot be applied to web data, which may or may not be true.

- The evaluation only have a handful of queries. Larger data set is desirable.