I’m trying out sentiment analysis where I convert the sentence into a Graph with nodes being word embedding and edges being dependency between the two words. I’m still confused how exactly should I pre-process the sentence. For example, removal of stop-words may result in wrong parsing of the sentence.