Similarity function

[andresespinosapc]

Could you explain how the similarity function training work?

If I'm understanding well, what you do is that instead of just using the candidates with one word in common, you train a similarity function and then pick the candidates that are most similar according to that function. Am I right? It is something they doesn't do in the original paper, right? What are the differences in results using the method of the original paper and yours?