Project 01
We introduce a comprehensive task-oriented conversational agent in low resource settings utilizing a novel pipeline ensemble technique to enhance natural language understanding tasks. The experiments conducted shows that our pipeline ensembling approach outperforms individual pipelines in precision, recall, f1-score and accuracy in both intent classification and entity extraction tasks. Furthermore, we implemented a Reinforcement Learning based dialogue policy learner addressing the overfitting issue by proposing a novel approach for synthetic agenda generation by acknowledging the underlying probability distribution of the user agendas with a reward-based sampling method that prioritizes failed dialogue acts
Project 02
Team members - Tharindu Madusanka, Thisara Welmilla