Automatic Question Identification on Turkish Tweets

In this work, it was tried to identify which are includes question from the tweets written in Turkish shared in Twitter, with a rule-based approach. As a data set, tweets that are shared with a certain hashtag are used instead of randomly sampled tweets. The reason for this, it is aimed to identify the questions asked for a specific focus in this study. For the experiments, 354 tweets were collected, shared with the hashtag that was created in order to contribute of the audience to the program by asking the questions while a historical topic was being spoken in the program broadcast live on a television channel. The Zemberek library has been used to fix typos in these tweets. Then, according to the Turkish question sentence structure, 3 simple rules are defined aiming at keeping the precision value or the sensitivity value high and each one is applied as a separate method. As a result of experiments, 100% precision, 96.48% sensitivity and 0.929 F-score values were recorded as the most successful performances