How to compute similarity of di and q?

[mug2mag]

hi @rodrigonogueira4 The paper, "Multi-Stage Document Ranking with BERT", mentioned that we use the [CLS] vector as input to a single layer neural network to obtain a probability of the candidate di being relevant to q. I can not find the corresponding codes for computing the probability of the candidate di being relevant to q. Can you help to give a tip? Thanks very much.

[rodrigonogueira4]

This probability is computed by monoBERT, whose implementation is here: https://github.com/nyu-dl/dl4marco-bert

If you are looking for a pytorch implementation, I suggest using this code: https://github.com/castorini/pygaggle/blob/master/docs/experiments-msmarco-passage-subset.md