Throughout the development of sentiment analysis systems, there have been tremendous interest in accurately determining valence in sentences and tweets, as indicated by various surveys. In these natural language systems, textual instances are often represented as vectors in a feature space, and is converted into binary, which can later be trained and observe whether the word is a positive or negative term. More recently, there have been significant improvements classification accuracy that have been obtained through low-dimensional continuous representations of instances and words. This method relies on the lotion that these vectors have only a few hundred dimension, and continuous refers to the real-valued nature of the dimensions, and that is the dimension does not have just 0 or 1 values, but can have any real number values, which can make the result more accurate.