Stack Overflow Query Outcome Prediction

Stack Overflow’s core mission is to create an online encyclopedia for all programming knowledge. In order to ensure quality content in the face of rapid growth, community moderators frequently close low quality questions, often asked by newcomers. In order to alleviate moderator burden and ease newcomers’ transition, we devise two classifiers to predict 1) whether a question will be closed and if close 2) its reason for closure. We train our models using logistic regression, SVMs, and boosting before selecting the optimal classifier. We found that the adaptive boosting algorithm best classified whether a question would be closed, whereas lasso-regulated logistic regression best classified the reason for closure. Our next steps to improve our classifiers include using word vectors, splicing the data by time period, and extracting more features from code segments.