Pass basket into FPMC instead of individual items?

[gcbrown]

Hello, I'm trying to generate some next-basket predictions. I have a dataset with columns (user, timestamp, basket), where basket is a set of ~10 items. I'd like to use FPMC to look at baskets as discrete sets like in the original paper. But in RecBole I can only figure out how to pass in individual items, which means I'm having to split the orders up into separate interactions/rows and I'm losing the discrete boundaries between orders. I can see that item_id_list is a potential input but I can't figure out how to pass it in, because whenever I put just item_id_list, run_recbole fails looking for the item_id column.

Is there something I need to pass into config_dict to tell the model to operate on token_seqs instead of individual tokens? Is it possible in RecBole?

Thanks

[Sherry-XLL]

@gcbrown Hello, item_id_list and item_id are generated automatically based on users' interactions in data_augmentation as follows. item_id_list is the historical item set in chronological order and item_id is the user's next interactive ground truth item.

RecBole/recbole/data/dataset/sequential_dataset.py

Lines 73 to 89 in ddff621

	def data_augmentation(self):
	"""Augmentation processing for sequential dataset.
	E.g., ``u1`` has purchase sequence ``<i1, i2, i3, i4>``,
	then after augmentation, we will generate three cases.
	``u1, <i1> | i2``
	(Which means given user_id ``u1`` and item_seq ``<i1>``,
	we need to predict the next item ``i2``.)
	The other cases are below:
	``u1, <i1, i2> | i3``
	``u1, <i1, i2, i3> | i4``
	"""

In RecBole, the existing code can only recommend individual items instead of basket items for the time being. If you want to operate on token_seqs instead of individual tokens, you should modify SequentialDataset in data preprocessing and refactor data loading according to your needs. Only changing the configuration file cannot realize the next-basket recommendation.

Thanks for your attention to RecBole.