Suranah writeup for Poon 2007

This is a review of the paper poon_2007_joint_inference_in_information_extraction by user:Suranah.

This paper tries to apply joint inference on the specific task of matching citations. Instead of first segmenting the citations, and then matching them, they use MLN to do the two steps together. The paper shows some improvement over other techniques.

My suspicion for this specific problem was that a joint inference cannot be a lot different, then passing along a large amount of ambiguity in the segmentation step (as there is a very limited number of combination anyways). It turns out to be somewhat true, as there is not a very substantial improvement as compared to Wellner for CiteSeer dataset.