LSTM in QA over KGs

[dinani65]

Hi
I am not sure if it is a right place to ask my question or not.
I am new in this field and I want to use a baseline dataset such as WebQuestionsSP and create representation of questions using LSTM via a pre-trained word embedding such as GloVe. My goal is to use the last hidden state of LSTM as representation of the questions.
When I examine the source code of existing works and compare it with what I have understood from LSTM, I become confused.
Could u please explain, e.g. how it has been done in the EmbedKGQA code?
thanks in advance,