This paper introduces a novel deep learning model for stock return prediction that leverages structured event representation (SER) and attention mechanisms. The authors use a pre-trained large language model (LLM) to extract event features from news articles, which are then structured into a format suitable for the SER model. The model employs attention mechanisms to weigh the importance of different events and their features, enabling effective cross-sectional stock return prediction. The key contribution is the demonstration that structured event representations, derived from LLMs, significantly enhance predictive performance and interpretability compared to existing text-driven models. The empirical results show that the SER-based model achieves superior out-of-sample performance, highlighting the value of structured inputs in financial forecasting. Additionally, the interpretable feature structures allow for deeper analysis of the economic mechanisms driving stock return predictability, offering insights into how news events impact stock prices. The findings underscore the potential of combining LLMs with structured modeling approaches for quantitative finance applications.