Detecting Structural Changes in the Stock Market and Explaining Them through Cluster Structure Embeddings

The stock market is a complex system influenced by diverse factors, such as monetary policy, economic cycles, geopolitical risks, and innovation. These dynamics significantly affect investment performance. Detecting such structural changes early and understanding the underlying factors is crucial for effective investment decision-making. This paper proposes an explainable intelligent decision support framework for the detection and interpretation of stock market structural changes. The proposed method monitors the dynamics of correlation-based cluster structures computed from high-frequency return data and detects market-structure changes using an entropy-based indicator. To interpret detected changes, we introduce a structure-aware representation, termed Cluster Structure Embedding (CSE), which maps mathematically grounded market structure signals into a language-compatible space. This representation enables the retrieval of relevant financial news associated with detected structural shifts, thereby providing interpretable explanations for market regime changes. Applying the proposed method to S&P 500 and S&P Mid-Cap 400 constituents demonstrates its ability to detect structural changes not captured by conventional market indicators while providing interpretable explanations to support investment decisions. Furthermore, comparisons with multiple baseline approaches, including LLM-based re-ranking, neural rerankers, and time-series–based methods, demonstrate the utility of the proposed method. These results highlight the effectiveness of the proposed method as an explainable and practically deployable investment decision support system.