As part of my learning path with ML/AI methods, I recently developed code implementations for deduplication — one on using learned representations with a classic neural network, and another on combining meta-blocking with a clustering method - which I then documented in two recent publications.
Tabular data turned out to be very tricky for the network as relationships between columns aren’t always obvious or structured. Meta-blocking gave me a great opportunity to experiment with and learn graph-based techniques - very powerful stuff !
Entity Resolution: Meta-Blocking and KLSH: https://app.readytensor.ai/publications/entity-resolution-learned-representations-of-tabular-data-with-classic-neural-networks-MtUrsAPP6Mdt
Entity Resolution: Learned Representations of Tabular Data with a Classic Neural Network: https://www.researchgate.net/publication/392752969
Github Repos: https://github.com/sergiosolorzano/entity_resolution