#sentiment-analysis

Question 1: Single choice

mediumAI-checkedRecurrent Neural Networks

A sentiment LSTM reads reviews padded or truncated to 200 tokens and classifies each review from the LSTM output at the last time step only, lstm_out[:, -1, :]. Why are short reviews padded with zeros on the left, e.g. [0, 0, …, 0, 117, 18, 128]?

So that the last time step comes right after the real words1
Because the embedding layer cannot look up index 0 at the end of a sequence2
To make the LSTM read the review backwards3
Because right padding would change the length of the feature rows4

Timed test