Research on Data Assets Pricing Based on LLM Agent

Haibo Han

Lanzhou University of Finance and Economics

Abstract

Conventional supply-side pricing models fail to capture the dynamic value of data assets in complex markets. This dissertation proposes a multi-agent AI framework where specialized autonomous agents collaborate to determine data asset prices through tool-augmented reasoning and collective deliberation.

The system comprises Value Assessment, Market Analysis, and Risk Evaluation agents coordinated by an Orchestrator. Each agent autonomously retrieves information through external tools and APIs, while a multi-agent debate protocol resolves valuation conflicts. A three-tier memory system enables continuous learning from historical cases.

Empirical results demonstrate superior performance: R² of 0.8473 (vs. 0.8154 for single-agent and 0.6423 for traditional ML), with multi-agent collaboration contributing 18.7% and tool augmentation 24.3% to overall accuracy. The debate mechanism reduces valuation variance by 34.2%, while human-in-the-loop verification achieves 96.8% expert concordance.

This research establishes a novel paradigm for autonomous economic decision-making, integrating multi-agent collaboration and human-AI symbiosis for robust data asset pricing.

Keywords: Data Assets; Pricing Model; Deep Learning; Large Language Model; Demand-side Pricing