Persona-Guided LLM Agents for Task-Oriented Dialogue
Not provided in the content
Abstract
This paper investigates the effectiveness of large language models in expressing personality traits during goal-directed dialogues and the impact of adapting to user personality on interaction quality.
Reality Card
Adapting to the user's personality improves task performance metrics such as constraint satisfaction and user satisfaction, despite a trade-off with truthfulness.
The cue-based adaptation method (Try) provides a more reliable route to personality-aware task-oriented dialogue without requiring fine-tuning.
The study's findings may vary based on the specific personality traits being expressed, as some traits are realized less reliably than others.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.