Sentiment scores of arguments and chunks don't align to each other

[jiqicn]

It's clear that this is caused by way of normalization of sentiment scores of arguments. Should find a way to have the scores normalized to reflect the actual sentiment.

[jiqicn]

To fix this I changed the way of normalizing the argument sentiment score as this: norm(x) = (x + 1) / 2.

Also, to have coherence scores that make more sense, I choose to use the Gaussian activation function with a deviation equals to 0.2.