Harassment by cyberbullies is a significant phenomenon on the social media.\nExisting works for cyberbullying detection have at least one of the following\nthree bottlenecks. First, they target only one particular social media platform\n(SMP). Second, they address just one topic of cyberbullying. Third, they rely\non carefully handcrafted features of the data. We show that deep learning based\nmodels can overcome all three bottlenecks. Knowledge learned by these models on\none dataset can be transferred to other datasets. We performed extensive\nexperiments using three real-world datasets: Formspring (12k posts), Twitter\n(16k posts), and Wikipedia(100k posts). Our experiments provide several useful\ninsights about cyberbullying detection. To the best of our knowledge, this is\nthe first work that systematically analyzes cyberbullying detection on various\ntopics across multiple SMPs using deep learning based models and transfer\nlearning.\n