Note
The original code has been refactored to be more concise and clean. As a result, the product search results could be slightly different from the numbers in our paper.
# First generate dense query/item representations and cache them
python generate_emb.py --plm_name hyp1231/blair-roberta-base --feat_name blair-base
# Then evaluate the product search performance
python eval_search.py --suffix blair-baseCLS --domain
--plm_name
roberta-base
roberta-large
princeton-nlp/sup-simcse-roberta-base
princeton-nlp/sup-simcse-roberta-large
hyp1231/blair-roberta-base
hyp1231/blair-roberta-large
Note
Please update feat_name
and --suffix
accordingly.