Graph-structured data is essential in numerous domains. The quality of node representations plays a crucial role in graph mining tasks. Random-walk-based techniques, such as DeepWalk (DW) [1], are effective for this task. However, the embeddings they generate often exhibit noise and redundancy. As a result, clustering efforts may face potential obstacles. To tackle this issue, we introduce the DWACE (DeepWalk Autoencoder Contrastive Enhancement) framework. This proposal enhances these embeddings, thus increasing clustering efficacy. DWACE employs a nonlinear autoencoder to denoise and compress the raw embeddings while preserving their structural integrity. It additionally integrates a contrastive learning objective to preserve the consistency of embeddings and enhance intra-community coherence. Our investigations on five benchmark datasets demonstrate that DWACE surpasses baseline approaches, attaining state-of-the-art performance in clustering.