Models for finding quality of affirmation and points of intervention in an academic discussion forum

Abstract

In academic discussion forums, people exchange thoughts, discuss, argue, and learn from each other. However, sometimes discussions can be inconclusive, messy, and vague. In order to make academic discussions meaningful, relevant interventions may need to be made whenever discussions lose clarity. Generally, academic discussion forum data is in the form of a sequence of questions, answers, and comments on the answers. All this data relating to a question (i.e. question, answers, and comments), considered together, is called an affirmation. If we can assess the clarity of an affirmation, it will be possible to find points where we can intervene in an academic discussion in order to steer it towards agreement or conclusion. We have trained and tested a model for assessing the clarity of an affirmation which is considered to be based on scores of agreements, disagreements, and partial answers. We have also trained a model for finding points where we can intervene whenever a discussion is leading to disagreement or inconclusiveness. We extend our previous work on assessing the quality of a discussion thread using a Random Forest classifier, by proposing an integrated affirmation assessment system here with a new sentence-level model. The sentence-level model gave us points where we could intervene in order to steer the discussion towards agreement or conclusion. The F1 score for the sentence-level model is 66.6%, and for the combined model it is 67.11%. In the paper, we have explained the models with experimentation and discussed the example by running the simulation of the model to show how the intervention works.

Keywords

• Social learning
• Discussion forum analysis
• Quality of affirmation
• Intervention
• Mediation