As the volume of textual data grows exponentially, automatic text summarization has become essential for efficiently processing and understanding large bodies of text. This paper reviews key advancements in abstractive and extractive summarization techniques, emphasizing the integration of deep learning models such as BERT, transformers, and adversarial learning frameworks. Abstractive methods like Neural Attention Models, DeepTextSummGAN, and T2SAM focus on generating coherent, human-like summaries by paraphrasing content, whereas extractive models like FuzzyTP-BERT and MFMMR-BERT aim to select the most relevant sentences, minimizing redundancy. Graph-based approaches such as RankSum further improve extractive summarization by capturing thematic relevance and semantic relationships. Despite significant advancements, challenges remain in producing highly unique and coherent summaries, particularly in abstractive methods. Additionally, handling subjective data in opinion summarization—which involves diverse user sentiments and informal language—remains difficult. This paper also highlights future directions, including improving content uniqueness, advancing multilingual summarization, developing domain-specific models, and creating hybrid approaches that combine extractive and abstractive methods. Addressing these challenges will further enhance the accuracy and relevance of automatic summarization systems, making them invaluable tools for processing diverse textual content across industries.