Deep Interest-Shifting Network with Meta-Embeddings for Fresh Item Recommendation

Algorithm 1

The training and testing procedure.

	Input: : auxiliary dataset
	Input: : warm-start dataset
	Input: : cold-start dataset
	Input: : a testing example
	Output: : the predicted label of
1	repeat
2	the first stage, pretrain the model using auxiliary data
3	Randomly sample a batch of data from
4	Calculate the predicted label by equation (1)
5	Update , , by gradient descent
6	until Converge
7	Fix , , to , ,
8	repeat
9	the second stage, train DisNet using warm-start data
10	Randomly sample a batch of data from
11	Calculate , using ,
12	Compute the shifted interest vector by equation (6)
13	Calculate the predicted label using by equation (7)
14	Update , by gradient descent
15	until Converge
16	Fix all the trainable parameters except the item id embeddings
17	repeat
18	the third stage, train RM-IdEG using cold-start data
19	Randomly sample an item and get its support/query sets from
20	Aggregate embeddings of relational items of by equation (11)
21	Generate an id embedding for using RM-IdEG
22	Compute the cold-start loss on by equation (2)
23	Update the id embedding of to by equation (3)
24	Compute the warmed loss on by equation (4)
25	Update RM-IdEG by gradient descent
26	until Converge
27	if is a cold-start item then
28	Generate an id embedding for using RM-IdEG
29	else
30:	Get the id embedding of from the embedding layer
31:	end if
32:	Return a label for by equation (7) using DisNet