The paper describes a hybrid reinforcement learning framework for trading that combines intraday market data with sentiment signals derived from a transformer model. The approach is designed to integrate both quantitative price action and qualitative financial news sentiment into the decision-making process.

The system extracts sentiment dynamics and semantic event clusters from text data, which are then fed into the reinforcement learning agent alongside market data. This suggests an attempt to capture both market microstructure and narrative-driven influences on asset prices.

The abstract does not include experimental results or performance comparisons, so the practical effectiveness of the framework remains unverified. The proposal is notable for its explicit combination of modern NLP techniques with RL in a trading context.