I have been exploring the R library qdap recently and it has many functions for doing exploratory analysis for dialogues.
Here is something I just found by using the function "question_type" in this library. This function counts the occurrences of different types of questions, such as "where", "why", and "how", in the sentences. I grouped the sentences according to the ratings of the vialogue conversations they belong to. See the plot below for the result. We can make a few interesting conclusions:
Vialogue 16171 talks about the Advanced Placement (AP) courses offered in different high schools which reflects the problem of educational inequality and some possible intervention methods. We applied the hidden topic Markov models on this vialogue. More specifically, we assumed that the comments from an individual post share the same topic and we treated the comments within each timestamps as individual documents.
Here is what we found. We used four different colors to represent four different topics: topic 1 is red; topic 2 is gre...
The following is the first part of my review on conversation modeling. Please feel free to highlight any ideas, feedback or recommendations. The hidden Markov models are widely used in the field of speech recognition, bioinformatics and so on. In the paper “Unsupervised Modeling of Twitter Conversations” by Ritter et al. (2010), it was applied to model Twitter conversations. Here is the plate diagram of this model.
Here are my understandings of what this plate diagram means: