Aiding food security and sustainability efforts through graph neural network-based consumer food ingredient detection and substitution

Data

The dataset used for this work was purchased from the data company Datafiniti, obtained via queries on their data for food and beverages with ingredients from US retailers (Amazon, Walmart and Target)⁴⁶. The data acquired was initially noisy and contained errors. The data was sanitised and processed into a more usable form, resulting in a dataset of \(\sim 3000\) products for node classification, and \(\sim 3200\) for link prediction. There were 7000 ingredients, and 14394 links in the node classification graph; for link prediction there was some variance in link number due to a…