Grunnleggende konsepter
Zero-data approaches can effectively train dialog systems, with synthetic data matching human data performance.
Statistikk
"Improving the original approach, and show that agents trained on synthetic data can achieve comparable dialog success to models trained on human data."
"We further demonstrate the scalability of our approach by collecting and testing on two new datasets: ONBOARD, a new domain helping foreign residents moving to a new city, and the medical domain DIAGNOSE, a subset of Wikipedia articles related to scalp and head symptoms."
"Our main contributions are: 1) Creating two new datasets, ONBOARD and DIAGNOSE. 2) Improving the training procedure for the CTS agent, increasing absolute dialog success by more than 18%."
Sitater
"The goal of CTS, as outlined by Väth et al. (2023), is to train an RL agent to traverse a dialog tree, guiding a user to the answer for a given question."
"Our changes to the CTS agent improve the combined success rate by over 10% compared to the original agent on the German REIMBURSE dataset and 18% for the English REIMBURSE-En."