Author, genre and gender identification using deep learning algorithms

Nowadays, the increasing amount of data has brought the need to classify these data. Classification is the process of categorizing similar data. In this study, it is aimed to make a modeling study in which Turkish news texts are selected as data and that these data can be classified according to author, genre and gender, machine learning and deep learning algorithms are used as classifiers. For this purpose, firstly, a total of 14 new data sets with large-scale and multiple classes, which can be used in author identification, genre identification and gender identification processes, containing columnists of a newspaper, were created. These data sets, which are 7 for author identification, 6 for genre identification and 1 for gender identification, have been made ready for the modeling phase, where the classifiers for identification are applied and the highest accuracy successes are investigated by passing through natural language processing steps specific to Turkish language. In the modeling phase, Multinominal Naive Bayes (MNB) and Random Forest (RF) algorithms, which are machine learning algorithms for the solution of author identification, genre identification and gender identification problems in Turkish texts, and Convolutional Neural Networks (CNN) and Long Short Term Memory (LSTM) from deep learning algorithms have been applied to data sets as classifiers. In addition, hyperparameter values with the highest performance from these classifiers have been tried to be found as a result of long experimental studies. As a result of modeling, using the accuracy, precision and recall values of the best models for each data set, the performance of each model was found. As a result of the modeling stage for author identification, it was seen that the CNN algorithm achieved the highest 95.81% accuracy in the AI-TNKU-7 data set compared to other algorithms used. As a result of the modeling for genre identification, an accuracy of 96.73% was achieved with the LSTM algorithm in the GI-TNKU-6 data set. It has been observed that the success of deep learning algorithms is higher than machine learning algorithms in other data sets used in genre identification. As a result of the modeling phase for gender identification, the LSTM algorithm performed better than other classifiers and an accuracy success of 88.68% was achieved.