Classification Turkish documents using deep learning techniques

In our study, it is aimed to distinguish and classify author profiles and text subjects with vectors which were created from authors posts. The vectors of the columnists of a newspaper were formed and analyzed for how much they could be separated from each other. Hence, author of any post, can be determined by this model. It also can group similar styles together. The DeepLearning4J Java library and the Doc2Vec class included are used during development. 5,10,15, 20 author vector models and their subject models were created according to their posts. The similarity vectors obtained in this way were compared with a certain threshold value, and the model performances based on the selection of different threshold values were measured. According to the results, some authors differed significantly from other authors. Articles can be successfully labeled according to their topics. This structure can be used especially in areas such as author profile extraction, article detection or subject grouping.