Recent advancements in computational power and information management have made it easier to anticipate stock market behavior. Data can be examined and intricate patterns found using sophisticated deep learning methods. Two distinct types of inputs are used by the most recent forecast models: both textual information such as news headlines or news content, as well as numerical information like historical prices and technical signs. But developing text models is required when using written data. Due to issues with word sparsity in datasets, conventional techniques like word embedding might not be appropriate for capturing the substance of financial news. In this study, we apply a deep learning approach to textual and numerical data, including financial URLs for finance news stories, to enhance stock market predictions and sentiment analysis. As input, we take into account market history data, event embedding vectors drawn from news headlines, and a number of technical indicators. Among other metrics, they total the net sentiment for each day and demonstrate that it has a strong ability to forecast future stock market movement. We use LSTM-BERT networks in our prediction technique. Loss MAE and RMSE, which are annualised returns based on trading simulation performance metrics, were derived using the GRU model. When losing, RMSE is 0.04735 as well as Losing MAE is 0.0012.