How to generate training data

Step 0) characterize all pairs of papers with a common author name on the 11 variables

Step 1) a pair is chosen according to a selection criterion and submitted to an oracle

Step 2) the oracle compares the pair and returns either 1 (most likely a match) or 0 (most likely a non-match)

Step 3) update the inferred (best fitting) matching function

Repeat Steps 1,2,3 until the relative likelihood of having inferred the underlying matching function is high

Previous slide

Next slide

Back to first slide

View graphic version