We introduce Iterative Perceptual Learning (IPL), a novel approach for learning computational models for social behavior synthesis from corpora of human-human interactions. The IPL approach combines perceptual evaluation with iterative model refinement. Human observers rate the appropriateness of synthesized individual behaviors in the context of a conversation. These ratings are in turn used to refine the machine learning models. As the ratings correspond to those moments in the conversation where the production of a specific social behavior is inappropriate, we can regard features extracted at these moments as negative samples for the training of a machine learning classifier. This is an advantage over traditional corpusbased approaches, in which negative samples at extracted at random from moments in the conversation where the specific social behavior does not occur. We perform a comparison between the IPL approach and the traditional corpus-based approach on the timing of backchannels for a listener in speaker-listener dialogs. While both models perform similarly in terms of precision and recall scores, the results of the IPL model are rated as more appropriate in the perceptual evaluation.We additionally investigate the effect of the amount of available training data and the variation of training data on the outcome of the models.
|Name||CTIT Technical Report Series|
|Publisher||University of Twente, Centre for Telematics and Information Technology (CTIT)|
- Machine Learning
- Social behavior synthesis
- HMI-MI: MULTIMODAL INTERACTIONS
- HMI-CI: Computational Intelligence
- Perceptual evaluation