[Context and motivation.] Extracting features from mobile app reviews is important for multiple requirements engineering (RE) tasks, but noisy and ambiguous feedback makes it difficult to obtain interpretable insights. [Question/problem.] Syntactic approaches often miss semantic context, while LLM-based methods may miss fine-grained features and typically output flat, weakly organized feature lists, limiting interpretation and comparability. [Principal ideas/results.] We propose FeClustRE, a framework that combines hybrid feature extraction with auto-tuned hierarchical clustering and LLM-based semantic labeling to produce structured feature taxonomies. FeClustRE is evaluated through benchmarking on two annotated app review datasets, one expert-annotated and one crowdsourced, and through a sample study on reviews of generative AI assistant apps. The evaluation indicates that the hybrid extraction approach reduces missed features and achieves a stronger overall balance between precision and recall than single-method baselines. For feature structuring, the auto-tuned hierarchical clustering produces coherent multi-level taxonomies that remain stable under varying data volumes, and the LLM-generated cluster labels are generally semantically consistent and align with official app descriptions. [Contribution.] This paper contributes (1) FeClustRE, an open-source framework for feature extraction and taxonomy generation, (2) an auto-tuning clustering and labeling pipeline with a reproducible evaluation methodology, and (3) empirical evidence that combining syntactic and LLM-based extraction generates more complete and interpretable feature representations from app reviews.