Extractive text summarization is a technique in natural language processing (NLP) that aims to pick out key sentences from a text to create a concise summary, preserving the original content. This paper presents a framework for extractive text summarization that uses linguistic features such as part-of-speech (POS) tags, word embeddings, and sentence embeddings to create a feature matrix highlighting the most important aspects of the document, from which the sentence matrix is generated. The framework employs Restricted Boltzmann Machines (RBM) to enhance this matrix, allowing for the selection of the top N sentences that best summarize the document. The quality of the generated summaries is assessed using the ROUGE metric, which compares them against reference summaries to quantify overlap and similarity. Further, the efficiency of the proposed methodology is compared with other baseline models using T-Test. The experimental setup utilizes the BBC News Summary dataset, consisting of 2225 documents across five categories—business, entertainment, politics, sports, and technology, making it a benchmark for extractive summarization. This methodology addresses the challenges of existing techniques like TextRank, LSTM, and BigBird, such as computational cost, scalability, and complexity. The application of RBM to enhance feature representation helps in better capturing the underlying structure of the data, making the feature matrix more robust and representative of the document’s content. The proposed RBM model shows an average improvement of 19.6% in ROUGE 1 F1-score, 32.7% in ROUGE 2 F1-score, and 21.3% in ROUGE L F1-score compared to the average scores of baseline models. This research provides a robust framework for extractive summarization and evaluation, with potential applications in information retrieval, natural language processing, and content generation.