Datasets:
Best approach to train product categorization?
Hello, first of all thx for the dataset. Any tips for better training approach? I tried distilbert/distilbert-base-uncased with AutoModelForSequenceClassification but it has very low accuracy.
I'm pretty newbie in model training and any additional info would be very helpful.
Multimodal models should help capture signals that are coming from both text and image.
Having gone through the dataset myself, I want to first say I appreciate that Shopify has made this public. That said, I was surprised by how inaccurate the taxonomy is. I'd expect a company of Shopify's size to provide a dataset where the taxonomy is reliable. Achieving high accuracy here isn't realistic without a painstaking cleanup first. So, Meliborn, I'd push back on the idea that your low accuracy is purely a matter of model choice, the underlying data quality is very likely playing a role too.