Making chatbots better by entropy-based data filtering (and future directions)
Richárd Csáky
Sept. 28, 2018, 14:00
SZTAKI
SZTAKI
We present our research on making general chatbots less boring. Our methods include semantic clustering of sentences, and filtering conversational datasets based on entropy. We also present future research directions and ideas, involving variational neural network models, self-play, and the Alexa prize (https://developer.amazon.com/alexaprize).