Sentiment analysis

[Dragoncall]

[see Tweepy API] [Also check nltk library]

This should take a list of Statuses and return a prediction.

[Dragoncall]

Pipeline has been created. Implementation needs to be done. The input data should be treated as abstract as possible to allow other combinations in a pipeline.

[annlytram]

First exploration of sentiment analysis prediction has been done in a separate notebook :

• sentiment analysis libraries such as TextBlob or VADER give poor results, accuracy ~ 40%
• implemented a first draft of an LSTM model (deep learning) trained on a Twitter sentiment analysis dataset (2016 GOP presidential debate) : accuracy ~ 65%
• used this model to predict on tweets extracted from a "donald trump" search query

I will now integrate this in the sentiment analysis pipeline. The model will be improved later on.