GROCLM: Grocery Category Recommendation in E-Commerce with Large Language Models

GROCLM is a fine-tuned large language model for grocery category recommendation in e-commerce, using LoRA training and constrained decoding.

It encodes cyclical purchasing patterns directly into model parameters, enhancing rebuying signal utilization compared to prompt-based methods.

Yes, e-commerce companies in the region can adopt the model to improve grocery category recommendations and increase cart-adds.

It achieves a 7.5% relative improvement in cart-adds per impression in a live production setting, outperforming baselines.