Abstract: The classification methods based on regular expressions and traditional machine learning had the problems of manual extraction of patterns and performance bottleneck, we proposed deep learning-based classification methods for question post, the deep text mining model TextCNN and integrating attention mechanism—TextRNN were used to construct a classification model. The experimental results show that the classification performance of deep learning-based methods is better than the existing benchmark methods on most of the question purpose categories, and the Adam optimizer is better than the SGD optimizer, and the Glove pre-trained word vector is better than randomly generated word vectors. The method classifies posts for the purpose of asking question, which can add a new dimension to the analysis of post discussion topics on Stack Overflow (SO).

Key words: post question purpose, deep learning, text mining, word vector