YandongL writeup of Poon 2007

This is a review of the paper Poon_2007_joint_inference_in_information_extraction by user:Yandongl.

Joint inference on Segmentation and Entity Resolution. Markov Logic Network based approached with a set of predicates MC-SAT as the satisfiability solver Voted perceptron algorithm for weights learning. Use different learning rate for each weight to speed up computation.

Segmenting similar ones can help each other. E.g. One with explicit boundaries between each entities can help one without boundaries. Edit distance based title similarity helps.

MLN learning: 30 iterations of gradient descent.

Performance improved, but only to a limited extent.