Sequential recommendation is an important research direction within the field of recommender systems, which predicts the items that users may be interested in based on the chronological order of user-item interactions. It receives increasing attention from various researchers in recent years. Although many previous models have achieved remarkable results, data sparsity remains a significant issue that has hindered the performance improvement of sequential recommendation models. Additionally, most current data augmentation models ignore time intervals. To address this, we propose Dynamic Adaptive Refinement of Interaction Sequence (DARIS), employing a novel dynamic adaptive strategy to augment sparse interaction sequences for enhanced recommendation performance. DARIS contains two core modules: the Item Reshaper replaces low-quality items with high-quality alternatives, and the Sequence Refiner subsequently identifies optimal subsequences from the reshaped sequences for final recommendation tasks. Comprehensive experiments on three datasets demonstrate that DARIS effectively augments original sequences and improves model performance, outperforming baselines across multiple evaluation metrics.