The ensemble of multiple Transformer language models has emerged as a promising strategy to address the inherent variability, uncertainty, and complexity of natural language processing tasks. However, traditional ensemble methods often lack the flexibility to adaptively weigh the diverse and context-dependent outputs of individual models. To overcome these limitations, this work introduces a novel ensemble approach based on fuzzy aggregation, applying the Weighted Ordered Weighted Averaging (WOWA) operator to combine the predictions of several pre-trained Transformer models. This fuzzy-based ensemble technique allows integrating the relevance and reliability of each model, providing a more adaptable and interpretable prediction result. The effectiveness of the proposal is demonstrated in the challenging problem of aspect extraction from user-generated content, as part of aspect-based sentiment analysis (ABSA). Experimental results show that the WOWA-based ensemble enhances aspect extraction performance respect to the reported solutions. This research offers new insights into advanced ensemble techniques at the intersection of fuzzy logic and deep learning, with broader implications for complex natural language processing tasks.