Identification of Sport News in Turkish Tweets Using Deep Learning Architectures

In social media and microblogging platforms, it is very popular to share news about sport activities from all around the world. This makes it important to extract information about sports out of text crawled from such platforms. In the scope of this paper, a binary classification is presented to identify tweets containing any information of sports in Turkish. To do so, firstly a dataset, composed of two categories (namely, sport and non-sport), is collected from eleven different Twitter news accounts. Afterwards, a preprocess phase takes place to remove the punctuation marks, the extra spaces, and the numeric characters. In the classification phase, accuracy values of four deep-learning architectures (namely, convolutional neural network, recurrent neural network, gated recurrent unit, and long short-term memory) are calculated to show the classification performances of each architecture. At last, the deep learning classification accuracy values are compared to the most commonly used supervised learning algorithms (namely, Naïve Bayes algorithm, Support Vector Machines, Random Forest, Dense Artificial Neural Network and Decision Tree).