<p>Abstractive summarization aims to generate concise and informative summaries from source documents. However, existing neural models often suffer from factual inconsistency, where the generated summaries contain information that is incorrect or contradicts the source content. To address this challenge, we propose a fact-aware enhancement and re-ranking approach for abstractive summarization, designed to improve factual consistency while preserving informativeness. Specifically, our method first extracts factual information from the source document and constructs a corresponding fact relation graph. A graph attention network-based fact encoder is then employed to update node representations within the graph. To further enhance factual accuracy, we introduce a fact-aware enhancement mechanism that evaluates the reliability of factual content by aligning source document facts with those in the candidate summaries. Our fact-attentive decoder incorporates an attention mechanism to guide the generation process with source factual information. Finally, we design a composite metrics re-ranker to select the most consistent summary from multiple candidates. Extensive experiments on the CNN/DailyMail and XSum datasets demonstrate that our approach significantly improves factual consistency and overall summarization quality compared to state-of-the-art models.</p>