Deep learning and hybrid metaheuristic algorithm based hate speech detection system in online social networks

With the rise of social media in recent years, online networks became a vibrant environment where billions of people all over the world interact with each other, share various content and perform countless daily activities. This intensive use of social networks brought along many problems. One of these problems, Hate Speech; It is all of the posts targeting a person, event, organization, society or order and exceeding the freedom of expression. It is important to prevent the sharing of these contents, primarily because they violate human rights. Unfortunately, social media data differs significantly from conventional data that is familiar. Considering the speed of sharing along with its enormous size, it makes it necessary for the Hate Speech detection system to have a structure that has the ability to make quick decisions with artificial intelligence-supported methods. Considering the speed of sharing along with its enormous size, it makes compulsory for the Hate Speech detection system to have a structure that has the ability to take quick decisions with artificial intelligence-supported methods. Hate Speech problem, which has a complex structure, brings demands for different new techniques in addition to the existing artificial intelligence methods applied for the solution of many interdisciplinary problems. With this study, metaheuristic optimization algorithms and hybrid version of these algorithms have been proposed, which will bring a new breath to the literature for the Hate Speech detection problem. In this completed thesis, the Hate Speech detection problem is modeled as an optimization problem. The proposed hybrid metaheuristic optimization algorithm is used for the first time to solve the Hate Speech problem. With the proposed method, the success of the system was evaluated by using various state-of-the-art optimization algorithms and artificial intelligence-based algorithms. Whether the proposed method works effectively and reliably in order to detect and prevent the spread of Hate Speech content in online social networks and its success against other methods was examined using various evaluation criteria. In the light of the promising results obtained, it is predicted that the methods proposed can be applied in solving many different social network problems.