This repository shows the step by step way to model the mutual influence in emotion from scratch. The project is still in progress

Extract the utterance name, transcript, valence, activation, gender and dialog-id from the raw transcripts of the dataset.

Create the turns from the utterance. The turn can be defined as utterances of a same speaker where almost no interruption occurred. We make the averge of valence and activation value of such utterances to produce the valence and activation of the turns.

To define and label emotion, we use two dimensional approach from valence and activation. The figure explains it:

We use a Markov chain to build transition matrix from one person's emotional state to the other.

We will create a hidden-markov-model to model the emotional dependency.