Generative Conversational Networks
This addresses the challenge of data scarcity for conversational AI systems, offering a novel method to enhance performance in tasks like intent detection and slot tagging, though it builds incrementally on existing meta-learning and generative teaching concepts.
The authors tackled the problem of training conversational agents with limited data by proposing Generative Conversational Networks, a framework where agents generate their own labeled training data using reinforcement learning and achieve an average improvement of 35% in intent detection and 21% in slot tagging over baseline models across multiple datasets.
Inspired by recent work in meta-learning and generative teaching networks, we propose a framework called Generative Conversational Networks, in which conversational agents learn to generate their own labelled training data (given some seed data) and then train themselves from that data to perform a given task. We use reinforcement learning to optimize the data generation process where the reward signal is the agent's performance on the task. The task can be any language-related task, from intent detection to full task-oriented conversations. In this work, we show that our approach is able to generalise from seed data and performs well in limited data and limited computation settings, with significant gains for intent detection and slot tagging across multiple datasets: ATIS, TOD, SNIPS, and Restaurants8k. We show an average improvement of 35% in intent detection and 21% in slot tagging over a baseline model trained from the seed data. We also conduct an analysis of the novelty of the generated data and provide generated examples for intent detection, slot tagging, and non-goal oriented conversations.