Chorus: LLM Framework Generates Realistic Online Deliberation Data
A team of researchers has introduced Chorus, a framework designed to utilize LLM-driven agents that exhibit behaviorally consistent personas for the creation of authentic deliberation data. Each agent operates autonomously, retaining memory of the ongoing conversation, while engagement timing adheres to a Poisson process to reflect diverse participation patterns. This framework accommodates the use of structured tools for accessing external resources and integrating web platforms. It was implemented on the Deliberate platform and assessed by 30 experts. This initiative aims to tackle the lack of extensive deliberation data, which is hindered by accessibility regulations, ethical issues, and varying quality.
Key facts
- Chorus is an agentic framework for generating deliberation data.
- It uses LLM-powered actors with behaviorally consistent personas.
- Each actor has memory of the evolving discussion.
- Participation timing is governed by a Poisson process-based temporal model.
- The framework supports structured tool usage for external resource access.
- It was deployed on the Deliberate platform.
- Evaluated by 30 experts.
- Addresses scarcity of large-scale deliberation data.
Entities
Institutions
- arXiv
- Deliberate platform