Muş Alparslan Üniversitesi Mühendislik-Mimarlık Fakültesi Dergisi 2(2), 91-100,2021 Complexity Matrices in Twitter Sentiment Analysis of Thoughts on Mobile Games Using Machine Learning Algorithms Erol KINA 1* , Recep ÖZDAĞ 2 Van Yüzüncü Yıl University, Özalp Vocational School, Van, Turkey, ORCID: 0000-0002-7785-646X Van Yüzüncü Yıl University, Department of Computer Engineering, Van, Turkey, ORCID: 0000-0001-5247-5591 Corresponding Author: erolkina@yyu.edu.tr Geliş tarihi:13.12.2021 Kabul tarihi:29.12.2021 Abstract In modern times, people have started sharing their opinions, thoughts, and feelings with other people through social media. The increasing number of social media users and their shares in social media platforms has naturally drawn the attention of researchers to this field. Twitter is one of the leading data sources in this field. Since Twitter has millions of users from different cultures and classes, it is possible to collect comments in different languages and content. Tweets that people write and share in 280 characters are used for research and analysis. Because not all tweets can be read by people, in this study, sentiment analysis was performed using Naive Bayes (NB) classification algorithm and multilayer artificial neural networks (ML-ANN) based on the content of comments on mobile games. As a result of the analysis, it was found that multilayer artificial neural networks gave better results than the other methods on both training and test data. Keywords: Mobile Games, Sentiment analysis, Twitter, Naive Bayes, Artificial Neural Networks I. INTRODUCTION Today people spend a lot of their time on the internet and social media. They can reach a large mass by sharing their feelings and ideas through social media. Personal likes and interests, many comments and shares on certain topics allow us to get a picture of the general mentality and attitude of the society we live in [1]. Sentiment analysis plays an important role in providing richer and more accurate results by allowing shy and introverted people, who are better at expressing themselves on social media platforms, to share their ideas with others. Sentiment analysis on Twitter consists of the steps of organizing, parsing, analyzing, and reporting the tweets sent by users [2]. Certain meanings are extracted and presented to companies or individuals through a special user interface from these results. Sentiment analysis studies determine whether the data have positive, negative or neutral content. In this study, to process data with machine learning algorithms, training sets are created by splitting the data into positive-negative. The aim of Sentiment analysis is being a part of data science that automatically handles the process of understanding and analyzing textual data as wanted. Thus, certain results are obtained. These results are used to determine people's ideas and thoughts about the subject of the study. Sentiment analysis studies can even be used to determine the general opinion not only of individuals but also of a particular audience. Thus, sentiment analysis can be used as a guide by showing the reaction to a decision to be made for a particular audience and the viewpoint of mobile game players on a particular game based on previous comments. Machine learning is necessary because it is not possible to analyze a large number of data, positive or negative opinions, large data sets one by one and come to a conclusion. Fig. 1 shows the basic machine learning process. First, the data set is obtained. Afterward, the data is passed through the preprocessing stages. The model is divided into a training dataset and a test dataset. The model is developed according to the results obtained. The accuracy rate for the training dataset obtained during the model evaluation process should be high.