Explanation of matching

[p16i]

Hi,

I'm not familiar with topic modeling, so this question might be naive.

I'm curious whether we can get explanation for each matching pair, e.g.
why is Review A assigned Paper Z?

[titipata]

Hi @heytitle, we don't have the explanation yet for each matching pair. We did create some explanatory model back then but haven't worked on it since (e.g. matching power and variance using some Bayesian model). Happy to chat with you more on this idea tho. I think it would be good to do it.

[p16i]

I've recently found Eberle et al. (2020)'s "Building and Interpreting Deep Similarity Models". It seems something that we can leverage here.

[titipata]

We can even use some of the pre-trained graph convolution model to calculate the distance between abstract here. Not quite sure if it can be explained but hopefully doing better than the baseline. Let me do a but more research to see the possibility 😉