Emotion Extraction from Turkish Text
Mansur Alp Toçoğlu, Adil Alpkoçak · 2014
In this study we present an emotion extraction system from Turkish text. The system is able to recognizes even emotional states from a given text for happy, shame, guiltiness, disgust, sadness, angry and fear categories. We consider Emotion Extraction as a Text Classification problem, which requires a training set. Thus, we first obtained a survey which is done with 500 university students to develop a training set where they are asked to describe their most intense moments they remember for seven emotions categories. Then, the text describing emotional moments are pre processed and modeled in Vector Space Model where tf × idf weighting scheme is used. Then we applied Naive Bayes classifier and tested with 10-fold cross validation, in WEKA tool. We evaluated the system in terms of accuracy, precision, Measureand recall measures. The results we obtained from the first experimentation are very promising where it achieved around 86% accuracy for seven emotional classes in average.