This website uses cookies and similar technologies to understand visitors' experiences. By continuing to use this website, you accept our use of cookies and similar technologies,Terms of Use, and Privacy Policy.

Nov 03 2014 - 10:49 AM
Model Vialogue Conversations with A Hidden Markov Chain Model

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

  1. A good conversation leads to learning
  2. Pre-test — determine the participants' initial cognitive patterns (i.e. mastery or non-mastery of different cognitive attributes).
  3. 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.
  4. Post-test — determine the participants' learning outcomes. (i.e. Change in cognitive attribute patterns; acquired attributes)
  5. Identify conversations leading to learning or the opposite. Human code the conversation. A sequence of observed variables is then generated.
  6. 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.
  7. 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)
  8. We have our good conversation!
  9. Build a model for each of the groups. Compare them.
|By: Xiang Liu|1742 Reads