CompEvo：竞争诱导进化提升多智能体新闻驱动时序预测
原标题：CompEvo: Competition-Induced Evolution for Multi-Agent in News-Driven Time Series Forecasting
AI 摘要
研究者提出 CompEvo，一个面向新闻驱动时间序列预测（NTSF）的多智能体竞争诱导进化框架，旨在解决智能体思维同质化（DoT）和策略更新缺乏理论依据（ITG）两大问题。该方法将多智能体交互建模为不完全信息进化博弈，证明混合策略贝叶斯纳什均衡的存在性，并给出局部收敛、长期遗憾和适应度权重单调性的分析。框架整合策略执行、基于适应度的可微选择和竞争诱导策略进化三个模块。在四个真实数据集上，相比最强多智能体基线，RMSE 平均降低 29.8%，MAPE 平均降低 30.3%，代码与数据已开源。
正文节选
CompEvo: Competition-Induced Evolution for Multi-Agent in News-Driven Time Series Forecasting Abstract News-driven time series forecasting (NTSF) uses evolving textual events together with historical observations to predict future values, supporting applications such as market risk monitoring and resource scheduling. In multi-agent settings, two challenges still remain. The first is degeneration of thought, where agents converge to similar evidence-seeking behaviors. The second is insufficient t