Abstract
Social media and the internet have connected the world like never before. This connectivity, while hugely beneficial, has some stark negatives. One such negative is the advent of cyberbullying. Cyberbullying takes place on many social media platforms, such as X, formerly known as Twitter. X and many other social media platforms do their best to moderate their platforms but there is room for improvement. This project proposes a novel form of real-time bad “behavior” detector that ideally could be utilized to help moderate and create a friendlier environment on X. As opposed to most conventional batch-based models utilized in content moderation this model is streaming-based, which should allow it to better adapt to changes in cyberbullying techniques. This model is an ensemble-based voting classifier that utilizes an X post’s characteristics as well as natural language encoding of the raw text of the post. The project utilizes X posts collected and labeled from prior research.
The proposed ensemble-based voting classifier was able to perform competitively with traditional batched-based classification models and other streaming-based classification models. The proposed model’s accuracy, f1, recall, and precision scores were in line with or outperformed the scores of other approaches. These results present some validity to the application of the model in real-world social media platforms.