We have added a new classifier that can determine the discourse of a text. It can for example distinguish questions from answers, if the answer is an agreement or disagreement. It even tries to see if there is humor in the text. The classes are listed below.
- Agreement
- Announcement
- Answer
- Appreciation
- Disagreement
- Elaboration
- Humor
- Negative_reaction
- Other
- Question
Since long texts often has mixed discourse, containing questions, answers, elaborations, humor an so on – it may make sense to pass single sentences or phrases for classification (split the text).
It’s based on the dataset from the paper “Characterizing Online Discussion Using Coarse Discourse Sequences (ICWSM ’17)” The dataset is built from annotated reddit comments.