0
$\begingroup$

Recently, I read a paper about recurrent recommender system, I am very curious about how it training its network.

Assume I have the Netflix dataset as

UserID    itemID    timestamp
  2         10          1244
  2         13           895
  2          6          1256
  2          7          1865
  2         11           256
  1          9          1284
  1         13           653
  1          8          1200
  1          4          1321

The paper says it split the data as training and test dataset based on time

assume we split the training set as

UserID(training)    itemID(training)    timestamp(training)
  2                   10                    1244
  2                   13                     895
  2                    6                    1256
  2                   11                     256
  1                    9                    1284
  1                   13                     653
  1                    8                    1200

And test dataset as

UserID(test)    itemID(test)    timestamp(test)
  2                7               1865
  1                4               1821

Now we can start to train the network:

enter image description here

my question is (1) how they feed above instance (user, item, timestamp) into above network?

(2) what if different has different number of item and how to represent $y_{i,t-2}, such as user1 has 10 items in the training data; user2 has only 2 items in the training data, then the length of LSTM will be different. does it matter.

(3) if want to using mini-batch gradient, How to split the training data? just randomly split or we need to split based on timestamp?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.