Knowledge Editing (KE) aims to modify specific knowledge within large models and address issues of knowledge misinformation or inaccuracies. Most previous studies were mainly based on structured facts, whereas real-world knowledge updates commonly emerge in unstructured texts. In this work, we focus on the CCKS-IJCKG 2024 Competition of Knowledge Editing, which features a Chinese dataset containing both structured and unstructured knowledge. We propose a two-stage approach for KE, using LoRA based method for model editing and incorporating the In-Context Learning Knowledge Editing (IKE) method to refine the editing process. In the first stage, we demonstrate that standard fine-tuning based on LoRA can yield superior performance with some key modifications. We align the inference process with the training process and implement template-based training, specifically without masking special tokens. To enhance portability, we also incorporate reverse masked loss during training. In the second stage, the IKE method leverages in-context learning to improve the model’s understanding and application of the edited knowledge in various contexts. Our method takes the first place in the CCKS-IJCKG 2024 Knowledge Editing Evaluation Task Competition, achieving a score of 52.6388%.