Reproduce “MoMoE: A Mixture of Expert Agent Model for Financial Sentiment Analysis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Peng Shu, Junhao Chen, Zhengliang Liu, Hanqi Jiang, Yi Pan, Khanh Nhu Nguyen, Zihao Wu, Huaqin Zhao and 3 more
A novel framework, MoMoE, integrates a Mixture-of-Experts architecture into a collaborative multi-agent system, enhancing financial sentiment analysis. This approach, built on LLaMA 3.1 8B, achieved an F1-score of 76.6% and 75.5% accuracy on financial benchmarks, outperforming leading general-purpose and specialized models.
Key insights
- Dual-level specialization, combining expert routing within a single model (LLaMoE) and collaborative decision-making among multiple expert agents (MoMoE), effectively captures financial sentiment nuances.
- Dynamically allocated expert modules within LLaMoE, along with an improved load-balancing loss, enhance computational efficiency and ensure balanced expert utilization.
- The multi-agent framework enables iterative collaborative optimization, allowing the final agent to synthesize diverse perspectives and correct errors from individual models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.