Turkish Text Classification with Machine Learning and Transfer Learning

The problem of text classification is one of the most fundamental topics of study in the field of natural language processing, but when reviewing the literature, it is seen that there is an inadequate number of studies for the issue of Turkish text classification. Two different Turkish datasets were created for this aim. Word vectors were created on the first dataset of unlabeled texts. These word vectors were transferred to the second dataset created with data collected from various news sites by transfer learning. Text classification was applied with the machine learning algorithms on this dataset. The effects of transfer learning and transferring of word vectors on the accuracy rate and the performance of machine learning methods were analyzed in detail. When studying the experimental results, it was determined that Support Vector Machine model was performed more successful and It was seen that the accuracy rate was improved with transfer learning.