Researchers have developed a new framework called SIGMA that uses SHAP values to guide the generation of features for Large Language Models (LLMs) without requiring semantic metadata. This approach, known as SHAP-enhanced Implicit-trajectory Generation for Metadata-free AutoFE, addresses two challenges in LLM-based Automated Feature Engineering: the lack of available metadata and the accumulation of trajectories that can lead to instability and duplication of generated features. The proposed framework, SIGMA, leverages SHAP values to provide task-aware signals for guiding feature generation and uses an exposed-feature Implicit Trajectory approach to reduce the risk of exceeding context windows. Empirical results show that SIGMA achieves performance comparable to state-of-the-art LLM baselines while reducing duplicate ratios and improving efficiency in feature utilization. --- Why it matters: This matters because it enables the use of Large Language Models in practical settings where metadata is unavailable, which could expand their applications in various domains such as natural language processing and computer vision. Source: https://arxiv.org/abs/2608.17948