<p>Entity alignment (EA) plays a crucial role in knowledge graph fusion, as it seeks to identify identical entities across different knowledge graphs. Recent studies have utilized side information, such as entity names, to achieve satisfactory performance. However, these methods often overlook the impact of knowledge facts on EA and suffer significant performance deterioration when side information is unavailable. Additionally, while different modal features enhance alignment performance, they also introduce noise. Existing methods that perform feature-level fusion at an early stage are unable to effectively reduce the noise impact between different features. Thus, we propose a multi-modal feature interaction framework (MFIEA) for EA. Our approach employs a joint knowledge facts embedding strategy to embed features of entities and triples. It utilizes the context information of triples to enhance the alignment effect. Furthermore, we enhance EA with a feature interaction mechanism based on feature similarity, which captures the potential relationships between different features. The proposed approach effectively utilizes various feature information to achieve information complementarity, and it ensures robust performance even in the absence of side information. Extensive experimental results verify that MFIEA outperforms the state-of-the-art baselines on public datasets.</p>