Where is the part inter_feat, user_feat and item_feat are all merged together?

[Jeriousman]

I am checking your code now to see how you are dealing with data. But I am kinda lost now as code is quite a bit lengthy. Which part of your code is the part where inter_feat, user_feat and item_feat are all merged together? I wish you can pinpoint it for me.

Thank you in advance.

[Sherry-XLL]

@Jeriousman Hello, the data processing part of inter_feat, user_feat and item_feat is in dataset.py, but I don't quite understand what you mean by merge together? inter_feat stores the interaction features, while user_feat and item_feat store the user and item features. They are three independent data structure and are not merged together.

[Jeriousman]

In the end, in dataloader, all the features from .inter, .user, .item will all be in dataloader which means they are merged into one data variable somewhere. Can you show me the code snippet where it happens?

Thank you.

[hyp1231]

Hi,

Certainly!

• For general / context-aware / knowledge-based recommendation, the features are merged in our Dataset.
  

  RecBole/recbole/data/dataset/dataset.py

  Lines 1256 to 1269 in 4f76169

  	def join(self, df):
  	"""Given interaction feature, join user/item feature into it.
  	Args:
  	df (Interaction): Interaction feature to be joint.
  	Returns:
  	Interaction: Interaction feature after joining operation.
  	"""
  	if self.user_feat is not None and self.uid_field in df:
  	df.update(self.user_feat[df[self.uid_field]])
  	if self.item_feat is not None and self.iid_field in df:
  	df.update(self.item_feat[df[self.iid_field]])
  	return df

• For sequential recommendation, it's a bit hard to join item features to the item sequences (for the efficiency concern). So we prefer to not merge them, but to fetch the item features via item ID in sequences in models. An example model that can be referred to is DIN.

[Jeriousman]

Thank you for your clarification! I will check those. The library is awesome and vase and because it is vast, I am lost quite often. Your help clears the way. Thank you again.