I actually trying to implement this paper: https://dl.acm.org/doi/10.1145/3323873.3326592
and it’s required me to input all the words seq in each review by list like:
[ [This, restaurant , is, great] ,[This ,hotel, has, a, great, service] , …]
btw, all the word is actually word index and the problem is I can’t apply the word embedding layers to each word in each seq and even if I can apply word embedding layers to each list of word ids, I can’t think of a way to feed each seq of word embedding independently to one GRU layers. To illustrate, by feeding the above word embedding seqs I wanted a resulted of m*n matrix in which "m" is emb size and n is number of reviews.
Source: Python Questions