Keyword extraction is a Natural Language Processing task that aims to automatically identify significant terms representing the core themes and subjects covered in a document. This process entails providing a text to a computer, which then returns a set of keywords suggesting significant words related to the document’s content. Keyword extraction methods find applications in various fields, particularly in information retrieval, where they help provide an overview of a document. This is particularly crucial as people often search for essential information based on keywords. While several techniques exist for English text, there is a notable scarcity of efforts in the domain of Bengali keyword extraction and context comprehension. In this article, we present an unsupervised approach ‘SobdoKrom’ ( ), which utilizes POS tagger, Bengali pre-trained word embeddings, and cosine similarity to extract keywords from the “Prothom Alo" news portal dataset in Bengali. We utilized various pre-processing steps to derive candidate keyphrases from the documents. Our model achieves a precision of 81% for extracting 5 keywords and 66.67% for extracting 10 keywords. Compared to existing approaches, specifically the neural-based approach, our approach demonstrates a notable improvement of 22% and 19% for extracting 5 and 10 keywords, respectively.