CHORUS framework generates realistic online deliberation data with LLM agents
Researchers A. Koursaris, G. Domalis, and A. Apostolopoulou propose CHORUS, an agentic framework that uses LLM-powered actors with consistent personas and a Poisson process-based timing model to synthesize realistic online deliberation discussions at scale.
Score breakdown
Teams building or studying agentic discussion systems can use CHORUS as a blueprint for generating realistic, large-scale synthetic deliberation datasets without relying on restricted or ethically fraught platform data.
- 01CHORUS is an agentic framework that orchestrates LLM-powered actors with behaviorally consistent personas to generate synthetic deliberation discussions.
- 02Each actor is governed by an autonomous agent with memory of the evolving discussion.
- 03Participation timing is controlled by a Poisson process-based temporal model to approximate real user engagement patterns.
CHORUS is an agentic framework proposed by A. Koursaris, G. Domalis, and A. Apostolopoulou to tackle a persistent bottleneck in online discourse research: the scarcity of large-scale deliberation data. Restrictive accessibility policies, ethical concerns, and inconsistent data quality on interactive web platforms make it difficult to obtain the kind of rich, realistic discussion data needed for analysis. CHORUS addresses this by orchestrating multiple LLM-powered actors, each assigned a behaviorally consistent persona and equipped with an autonomous agent that maintains memory of the evolving discussion thread.
The framework also supports structured tool usage, allowing actors to access external resources and enabling integration with interactive web platforms.
A key technical contribution is the framework's use of a principled Poisson process-based temporal model to govern when each actor participates, approximating the heterogeneous engagement patterns observed among real users. The framework also supports structured tool usage, allowing actors to access external resources and enabling integration with interactive web platforms. CHORUS was deployed on the Deliberate platform and evaluated by 30 expert participants who assessed it across three dimensions: content realism, discussion coherence, and analytical utility. The expert evaluation confirmed CHORUS as a practical tool for generating high-quality deliberation data suitable for online discourse analysis.
Key facts
- 01CHORUS is an agentic framework that orchestrates LLM-powered actors with behaviorally consistent personas to generate synthetic deliberation discussions.
- 02Each actor is governed by an autonomous agent with memory of the evolving discussion.
- 03Participation timing is controlled by a Poisson process-based temporal model to approximate real user engagement patterns.
- 04The framework supports structured tool usage, enabling actors to access external resources and integrate with web platforms.
- 05CHORUS was deployed on the Deliberate platform for evaluation.
- 0630 expert participants evaluated the framework across content realism, discussion coherence, and analytical utility.
- 07The framework targets the scarcity of large-scale deliberation data caused by restrictive platform policies, ethical concerns, and inconsistent data quality.