From the last few years, the use of social media has increased resulting into the rise of fake news and their spreading on a large scale. Recent political events have increased the spread of fake news. As seen by the widespread impact of the huge beginning of fake news, people are inconsistent in the absence of effective fake news detectors. This work has made an attempt to automate the fake news detection process by employing the logistic regression (LR) and latest and modified word embedding technique. In this paper, we worked on the fake news recognition mechanism for 2 different datasets, viz. dataset comprising online traditional news articles and news collected from a wide range of sources. The results are compared with long short-term memory (LSTM) and traditional machine and deep learning methods for both the datasets. It reveals that the traditional mechanism for attention does not function as expected. With the help of word2vec embedding, we modified the original attention mechanism, which is more effective in dealing with this issue. The proposed method is compared with several outstanding approaches and the results are presented. Our work outperforms these methods in many parameters. This approach has created a framework that captures various fake news indicators and classifies the news as genuine or fake and makes decisions.