Social media and other online platforms like websites, blogs, and articles are filled with images, videos, and text. This has created a rich and diverse online environment. Individuals post short, funny content that uses symbols and humor to express ideas and spark debate. But the shareability also gives a way to spread easily harmful content, such as hate speech and stereotypes. Because of this reality, there needs to be proper mechanisms for detecting such content. Natural Language Processing (NLP) helps in analyzing text to find inappropriate language. This study looks at how advanced machine learning algorithms can be used to find offense or hate in images, videos, and text. Video transcripts added to the better understanding of the content. Identifying harmful content goes way beyond just acknowledging images or faces. Many posts use irony, sarcasm or cultural references, which makes detection more challenging. So, it’s important to use strategies that understand context and can find patterns and emotions through sentiment analysis. In this study, different word embedding was implemented, which include Bag of Words, Word2Vec, GloVe, FastText, and TF-IDF. After training the model with each method, their results were compared to see which one performed the best.