In order to produce succinct and contextually appropriate summaries, this work examines Bidirectional Encoder Representations from Transformers (BERT) for extractive semantic summarization, highlighting its capacity to recognize and utilize keywords and important phrases. BERT outperforms conventional frequency-based keyword extraction techniques by identifying semantical properties from text sentences and visual pictures thanks to its contextual embeddings and self-attention processes. In order to evaluate BERT’s performance in extractive summarization tasks, the article describes a thorough experimental setup that includes dataset selection, preprocessing, model construction, and evaluation measures. To maximize model performance, important hyperparameters are discussed, including batch size, learning rate, and optimization techniques. Findings show that BERT has a strong knowledge of semantics, context-aware extraction, and cross-domain adaptation, which makes it an effective tool for producing high-quality summaries in a variety of applications, including news aggregation, scientific research.