Deep models predict missing product attributes from text and images.
problem Incomplete or missing product attributes in e-commerce catalogs.
method Combining textual and visual data with a novel modality-merging method.
result Our approach improves attribute prediction on Rakuten-Ichiba and other datasets.