This research proposal presents DAG-Negotiate, a novel multi-agent framework that enables context-aware, adaptive requirements negotiation through dynamic agent instantiation and role specialization. The framework addresses limitations of static LLM-based RE systems by introducing a Context Analysis Module, Dynamic Agent Generator, Dialectical Negotiation Protocol, and Consensus Orchestrator to handle emerging conflicts and quality attribute trade-offs. Unlike existing approaches with predefined agent roles, DAG-Negotiate dynamically instantiates specialized agents at runtime based on negotiation context and stakeholder heterogeneity. The work includes a comprehensive experimental validation plan involving comparative analysis against static baselines and ablation studies.
Key findings
Novel architecture for runtime dynamic agent generation enabling adaptation to emergent requirements negotiation scenarios.
Context-aware negotiation protocol that adapts agent strategies based on real-time assessment of stakeholder state and domain characteristics.
Dialectical argumentation framework for systematic resolution of multi-dimensional quality attribute conflicts through structured agent debate.
Consensus orchestration mechanism synthesizing diverse dynamically-generated agent perspectives into coherent requirements specifications.
Limitations & open questions
Framework evaluation relies on simulated negotiation scenarios which may not fully capture real-world stakeholder complexity.
Dynamic agent generation introduces computational overhead potentially limiting scalability for large-scale requirements projects.
Dependence on Large Language Models introduces risks of hallucinations or inconsistent agent behaviors during negotiation.