Social media has a very important contribution to human lives today. Through social media platforms people can share their information, ideas, knowledge, and activities with connecting people in the form of videos, images, texts, and audios. In the context of sharing information, incorrect information is also shared along with the correct information. In this way, unauthentic (fake news), misleading (rumors), abusing, toxic, extremist contents are also shared through social media platforms. This paper reviews the influences of social media content. In this context, vector representation of the social media sentences, word embedding models has been best applied for better accurate results. Natural language processing (NLP) and text analysis techniques are being used to extract useful information from social media content. The NLP techniques are widely used for correcting the sentences and identifying their meaning also. Currently, machine learning (Decision Tree, Random Forest, SVM, Na ̈ıve Bayes) and deep learning (LSTMs, BLSTMs, GRUs, CNNs) models are successfully being implemented to classify social media contents. In the comparative study of different works of literature and results from LSTM and CNN-LSTM deep learning model have been proved that deep learning and the word embedding model provide better accurate results for social media contents categorization.