Leveraging language model-based conversation summarization to enable effective and efficient retrieval of similar dialogues for few-shot dialogue state tracking.
A novel approach that synergizes dynamic demonstration retrieval and cognitive-aspect situation understanding to enhance the quality of emotional support provided in conversational systems.
Hexa is a self-improving method that can enhance the performance of modular dialogue systems by leveraging a novel bootstrapping scheme with guided prompts and a modified loss function to generate diverse and appropriate intermediate steps and final responses without access to ground truth data.