Predicting Tags of Stack Overflow Questions: A Deep Learning Approach

Abstract

Software developers and programmers will have questions about code syntax, coding best practices, and solutions based on specific error types, learn through discussions in various forums to improve their knowledge. These forums contain an abundant number of questions and answer discussions on a broad range of topics. To get the exact match based on the specific search on the internet, these posts must contain tags related to that specific post. These tags allow the post to gain more exposure and easier discovery of the post for both solutions as well as answers on the forums. The objective of this paper is to predict tags of Stack Overflow posts using Long Short Term Memory, a special type of Recurrent Neural Network algorithm. The proposed system is examined with contrast to Multi-Layer Perceptron and Gated Recurrent Unit (GRU). The tag prediction system is assessed on test accuracy, hamming loss, subset accuracy, jaccard score, precision, recall and f1-score.