Okay, it might sound a little nerdy. I've been thinking about modeling Vialogue conversations with a Hidden Markov Model (HMM) over the Halloween weekend. I briefly outlined the procedure. Any comments or suggestions?
Hidden Markov Chain Approach
- A good conversation leads to learning
- Pre-test — determine the participants' initial cognitive patterns (i.e. mastery or non-mastery of different cognitive attributes).
- Divide participants into two groups. They both watch the same material. The first group is then engaged in a structured conversation with a moderator involved. (The moderator will guide participants through a conversation directly related to each cognitive attribute). The second group is then engaged in an unstructured conversation without the involvement of a moderator.
- Post-test — determine the participants' learning outcomes. (i.e. Change in cognitive attribute patterns; acquired attributes)
- Identify conversations leading to learning or the opposite. Human code the conversation. A sequence of observed variables is then generated.
- Model conversations with a Hidden Markov Model (HMM). Cognitive (learning) process (each hidden state models different cognitive states) is modeled by hidden states. Observed variables are what happened in a conversation. Fit the model.
- Identify the hidden state sequence that leads to learning. Look into what happened in the conversation by identifying the corresponding observed sequence. (The most probable observed sequence)
- We have our good conversation!
- Build a model for each of the groups. Compare them.