ACOSUM: Ant Colony Optimized Multi-level Semantic Graph Summarization

Arabic text summarization faces unique challenges due to the language's complex morphology, diverse dialects, and intricate syntax, making it difficult for conventional methods to produce high-quality summaries. To address these challenges, this paper introduces ACOSUM, a novel abstractive summarization framework that combines Multi-level Semantic Graphs (MSG) with Ant Colony Optimization (ACO). ACOSUM constructs hierarchical semantic graphs to capture nuanced textual relationships and employs LSTM networks with attention mechanisms to generate coherent summaries. Additionally, ACO optimizes hyperparameters, enhancing both accuracy and efficiency. Experiments on a dataset of 25,000 Arabic articles demonstrate ACOSUM's superior performance, achieving ROUGE-1, ROUGE-2, and ROUGE-L scores of 41.4%, 22.8%, and 37.3%, respectively. These results outperform baseline methods like TF-IDF and Transformer-based models, highlighting ACOSUM's ability to produce concise, contextually accurate summaries. The framework's success underscores its potential to advance Arabic text summarization and its applicability to broader natural language processing (NLP) tasks, such as news aggregation and document analysis.