With the increasing shift to web-based platforms, particularly in e-commerce, physical verification of products has become limited. As a result, virtual verification through customer reviews plays a crucial role in influencing purchasing decisions. However, the presence of fake reviews compromises the trustworthiness of these decisions. This paper introduces a novel approach for detecting fake reviews by fusing embeddings to generate high-dimensional input for the identical arms of a Siamese network. Specifically, MiniLM BERT embeddings capture contextual relationships, while Word2Vec embeddings focus on semantic relationships. This fusion of embeddings into a unified high dimensional vector enhances the training of the Siamese network with LSTM layers, improving its ability to learn meaningful patterns. The output is further processed through a fuzzy logic system to ensure a robust decision-making process. Experimental results demonstrate significant improvements over state-of-the-art methods in fake review detection.