Extract two sets of your own conversations, one with someone you like and one with someone you dislike.
You can define "liking" in any way that works for you, but try to have the two conditions be opposites along the same dimension (someone you admire vs. someone you despise, someone you are close to vs. someone you don't know well). Obviously, this will be constrained by the set of people you have a significant amount of conversational data with (at least 20 turns on each side). Please let me know if you are having difficulty getting this kind of data.
For some feature, find the similarity between the two sides of the conversation for each condition.
You can choose any feature that we have seen to be useful for some task.
You can define similarity in any way you want. (Difference in average usage per turn, difference in average usage overall, averaged difference in pairwise turns, min/max difference in pairwise turns, probability of feature after other person used feature...)
Compare the similarity between the likee and dislikee conversations.
Fill out liking_results.md.
Wednesday 5/10 11:59pm
Email me ([email protected])
- The two categories of data (like/dislike, close/stranger, etc.)
- Feature(s) examined
- Similarity measure(s)
- Featurized data (as csv)
- IPython notebook or python script for loading features and calculating similarity measures and differences