Hello everyone, I was wondering, has anyone had any experience with record linkage (entity deduplication) using fastai?
In my case I am working on detecting duplicate products, a common way of doing this is via text similarity using siamese networks, but im sure if I managed to combine the product image + the text it would yield better results.
How would you approach such problem?
This paper, Merging Datasets through Deep Learning, could be a good starting point. It sounds like they pursued the implementation (not using fastai) that you’re describing
Thanks for sharing!
I have read that paper, there are a some implementations using siamese networks for text.
I was more interested if anyone has used fastai for this problem, particularly combining image and text.