Embedded Emotion-based Classification of Stack Overflow Questions Towards the Question Quality Prediction

Software developers often ask questions in Stack Overflow Q & A site, and their posted questions sometimes do not meet the standard guidelines. As a consequence, some of the questions are edited by expert users, some of them are down-voted, or some are even deleted permanently. Besides, the users (i.e., developers) might not get the expected solutions for their problems. In this paper, we study up-voted and down-voted questions from Stack Overflow, and analyze the relationship of embedded emotions with question quality. We use Sentiment140 API for identifying embedded emotions in the question texts, and then apply Feed-Forward Multilayer Perceptron (MLP) and Support Vector Machine (SVM) on the emotion data for developing a quality prediction model. Experiments using 38,920 Stack Overflow questions suggest about 70% precision and about 74% recall for our model with 10-fold cross-validation, and these findings clearly reveal the impact of human emotions upon the quality of a question.