---
title: "情感增强深度强化学习在主动交易中的应用：基于 Alpha 奖励的 CLaC@FinMMEval 2026 系统"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.16028v1"
permalink: "https://feed.astralor.com/items/itm_7548a6a8ba9448db090789ca"
published: "2026-07-17T15:01:35.000Z"
score: 77
channel: "finance"
tags: ["CLEF 2026","DDPG","DQL","情感分析","深度强化学习","量化交易"]
---

# 情感增强深度强化学习在主动交易中的应用：基于 Alpha 奖励的 CLaC@FinMMEval 2026 系统

## AI 点评

> 这篇论文对量化交易从业者和金融AI研究者都很有价值。它不仅展示了如何在真实市场数据中结合LLM情感得分与深度RL，还诚实揭示了策略从牛市迁移到熊市的泛化问题，这对实际部署中的风控和回测框架设计有直接启发。建议关注Alpha奖励的设计思路以及DDPG与DQL在不同资产上的性能差异。

## 速览

本文系统介绍了参加 CLEF 2026 FinMMEval Lab Task 3 的方案，利用 LLaMA 3.2 1B 生成的每日新闻情感得分，结合技术指标与日历编码，为比特币和特斯拉构建深度强化学习交易代理。论文比较了 PG、PPO、DQL 和 DDPG 四种算法，并引入基于超额市场回报的 Alpha 奖励以激励跑赢买入持有策略。实验显示，DDPG 在测试集上综合表现最优，DQL 因验证夏普比率最高而被选为线上端点。特斯拉标的下 DDPG 和 DQL 分别获得 54.96% 和 52.62% 的累计回报，远优于买入持有的 16.45%；比特币标的下 DDPG 实现 1.58% 正回报，而买入持有亏损 34.27%。研究同时揭示了牛市训练策略向熊市迁移时的泛化缺口。

## 关键点

- 使用 LLaMA 3.2 1B 模型对每日新闻进行情感评分，并嵌入深度强化学习状态空间。
- 提出 Alpha 奖励函数，以超额市场回报作为优化目标，引导代理跑赢买入持有基准。
- 在特斯拉和比特币标的测试中，DDPG 和 DQL 均大幅超越买入持有策略，尤其在熊市中展现出正回报能力。
- 发现了显著的验证-测试泛化缺口，表明牛市训练的模型在熊市环境中性能下降，对实际部署具有重要警示意义。

## 全文

This paper presents our system for Task 3 of the CLEF 2026 FinMMEval Lab, which requires daily long, flat, or short trading decisions for Bitcoin (BTC) and Tesla (TSLA) using news and historical market data. We formulate the problem as a discrete-action Markov Decision Process and compare four deep reinforcement learning algorithms: Policy Gradient (PG), Proximal Policy Optimization (PPO), Deep Q-Learning (DQL), and Deep Deterministic Policy Gradient (DDPG). The agents use technical indicators, cyclical calendar encodings, and daily news sentiment scores produced by LLaMA 3.2 1B. To reduce overfitting and align training with the objective of outperforming buy-and-hold, we introduce an alpha reward based on excess market return and randomize episode start dates. Hyperparameters are optimized with Ray Tune over 180 trials per algorithm-asset pair, with early stopping and model selection based on validation Sharpe ratio. On the CLEF Task 3 test set, DDPG achieves the strongest overall performance. DQL was selected a priori for the live endpoint because it obtained the highest validation Sharpe ratio, with selection performed without access to the test period. For TSLA, DDPG and DQL achieve cumulative returns of 54.96% and 52.62%, respectively, compared with 16.45% for buy-and-hold. For BTC, DDPG achieves a positive return of 1.58% while buy-and-hold declines by -34.27%. The results also reveal a substantial validation-to-test generalization gap, highlighting the difficulty of transferring policies selected in bull-market conditions to a bear-market regime.

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_7548a6a8ba9448db090789ca