Coding Small Group Communication with AI: RNNs and Transformers with Context

Abstract

This study compares the performance of recurrent neural networks (RNNs) and transformer-based models (DistilBERT) in classifying utterances as dialogue acts. The results show that transformers consistently outperform RNNs, highlighting their usefulness in coding small group interaction. Furthermore, the study explores the impact of incorporating context, in the form of preceding and following utterances. The findings reveal that adding context leads to modest improvements in model performance. Moreover, in some cases, adding context can lead to a slight decrease in performance. The study discusses the implications of these findings for small group researchers employing AI models for text classification tasks.