Spam fi̇lteri̇ng on turki̇sh youtube comments

In parallel with the increasing spread of Internet usage, social media usage rates are also increasing rapidly. One of the most preferred platforms by social media users is YouTube. Increased use of YouTube has brought some problems. Repeated spam comments, which are unrelated with shared video content and used for advertising purposes, cause usage of resources unnecessarily in general. This study aims to detect spam comments automatically on YouTube comments. Research results of previous studies performed for this purpose showed that although systems were developed for solving text classification problems in other languages, these kinds of studies for Turkish language were quite limited. In this thesis, datasets consisting of Turkish Youtube comments were created and the performance of automatic text classification algorithms were evaluated on the datasets. An important contribution of this thesis is the creation of Turkish datasets that will be available for use in future academic studies. In the study, the performances of classification algorithms that yield good results in terms of accuracy and speed were compared using the Weka natural language processing tool. In terms of accuracy values, the SMO and Random Forest machine learning algorithms appear to be more successful on the Turkish Youtube comments classification problem than others.