Polı̇tı̇kada sayılar: Dış polı̇tı̇ka yönelı̇mı̇ ve seçı̇m tahmı̇nı̇nde karşılaştırmalı sayısal analı̇z ve modelleme

To advance social science in the direction of accurate and reliable quantitative models, especially in the fields of International Relations and Political Science, new novel methodologies borrowed from the Computer Science and Statistics should be employed. In International Relations, quantitative analysis can be carried out to understand foreign policy topic relations in public discourse of decision makers. In domestic politics, Election Forecasting is a suitable area, because of its offering of already quantified vote results and its importance in the decision-making process in domestic politics and foreign policy. This work embarks upon a computational- statistical model built on social media (Twitter) data and texts' meaning extracted, analyzed and modeled with the state-of-the-art methodologies from Computer Science (Machine Learning and Natural Language Processing) and statistics to forecast election results and foreign policy orientation. To verify the model, Turkish General Election 2015, US Presidential Election 2016 and campaign period of Donald Trump are analyzed. This work shows that, sentiment of political tweets can be captured with high predictive accuracy (92% in Turkish, 96% in English) and using opinion poll results for a given period of time, vote percentage fluctuations can be predicted. Furthermore, it is possible to capture the foreign policy orientation of a candidate by his and his team's tweets in the campaign period. Keywords: Election Forecasting, Foreign Policy Analysis, Machine Learning,Natural Language Processing, Sentiment Analysis