Analyzing data by inferring from unstructured data about customers is one of the main purposes of the tourism and many other industries as well. However, performing unstructured data analysis using traditional methods is quite inconvenient and costly. This can be overcome by using sentiment analysis, an area of application of text mining. Since there is no proven methodology for sentiment analysis, researchers often perform their studies by trial and error. Many studies on sentiment analysis have focused on comparing the preprocessing or the performance of various machine learning algorithms. Both for these reasons and since research on sentiment analysis with Turkish content is limited, this study aimed to determine the effects of labeling, stemming, and negation on the success of sentiment analysis using Turkish touristic site analysis. From the data set prepared for this study, 12 different variations were created according to labeling, number of classes, stemming, and negation. These data sets were classified using the algorithms Naive Bayes (NB), Multinominal Naive Bayes (MNB), k-Nearest Neighbor, and Support Vector Machines (SVM), often used in sentiment analyses, and the findings were compared.