Özet:

Today, due to the development of technology, use of social media and the number of social media networks are increasing day by day. This means that the number of hate speech contents on social media increases and, accordingly, these contents are shared. The measures taken by public authorities and social media developers to prevent these posts that negatively affect social media users are insufficient. For this reason, automatic systems that facilitate the detection of hate speech content produced in large quantities are needed. On the other hand, when studies conducted in English and many other languages are examined, sufficient studies have not been achieved yet in detecting hate speech content in Turkish. When existing studies are examined, use of small datasets, scarcity of feature selection methods and similarity of classification algorithms are seen. This study is carried out to eliminate all these deficiencies mentioned in the literature. Within the scope of the study, two tweet datasets of different sizes in Turkish (dataset_1k, dataset_2k) are used. First of all, five different feature selection algorithms are applied on these datasets and the number of features is reduced to 1,000. Then, machine learning algorithms are used on 1,000 features obtained from each feature selection to detect hate speech. These algorithms are K-Nearest Neighbor (KNN), Random Forest (RF), Naive Bayes (NB), Support Vector Machine (SVM), Long Short-Term Memory (LSTM) and Word2Vec+SVM methods, respectively. In experimental results, F-measurement value is taken as a criterion of success. In the experimental results, F-measure is used as the performance metric. NB algorithm with oneR feature selection is achieved the best result for Data set_1k with an F-measure of 88.81%. NB algorithm with InfoGain feature selection is achieved the best result for Data set_2k with an F-measure of 87.71%. It has been observed that LSTM algorithm is more successful than all other algorithms in the experiments with five different feature selections.

Anahtar Kelimeler: