It is National Science Week, which means for two days we will have tweens on the 4th floor. Dominic and I worked with Matt Purver and Stu Battersby. Stu and Matt have been working on this Twitter Classifier where they are trying to train a system to determine the emotional intention of tweets. The system is mainly trained using emoticons to determine the emotion of the tweeter. As expected, the system is not very good.
Dominic and I worked on a gaming experience to engage the students with the classifier. We played off that the system does not work well and that the tweens are smarter and better able to determine emotional meaning then the system. We set the experience around they were smarter and we need them to help teach the system, putting them in position of power.
There were three rounds:
Round 1 – we give them 6 sentences crafted to map to the 6 emotions. It was easy for both students and the system to map the emotion.
Round 2 – we give them 4 sentences that are vague. It is harder for both the students and the system to map.
At the end of the round, we discussed why these were hard to categorize by text alone. We asked the students to offer up ideas of how they would change the sentence to make it represent one emotion. They could do things like add punctuation, emoticons, and acronyms. It was typically hard to get categories like disgust or anger.
Then we have a free for all round where the students could submit sentences that they think represent one of the emotions. If the system was confused they would work together to try to make it more explicit. When they feel they have reached a point where the sentence is right for the emotion. It was submitted to the system to teach it that this sentence equals X emotion.
We were pleased and surprised by the level of engagement. Often conversations brought up the issue of sarcasm.
One of the interesting rounds was when the students submitted “l love you” for happy. When we asked them to make it sad. They changed it to “I loved you.”
One group had their teach bring them came back, after playing with avatars, to ask if the system would work with foreigns languages.
It was a good experience for me to look and think about communication outside of context and limited communication tools.