In natural language processing, sentiment analysis analyses the sentiment a given piece of text conveys. This can be broadly applied to different bodies of text: reviews, emails, tweets.
A primitive, naïve sentiment analysis model would look up the embedding for each word, add up the word embeddings for a full-body embedding, and input this into a fully connected network. But this ignores that certain sentences could have an order or combination of words that imply something else numerically.
Instead, one strategy we could try is using a RNN.