基于注意力循环神经网络的Stack Overflow问题质量评估方法

A Stack Overflow Problem quality assessment method based on attention recurrent neural networks

文章对200万个Stack Overflow问题内容及评分进行统计分析,提出一种基于循环神经网络和注意力机制的问题质量评估模型SOG(stack overflow grader).该模型利用循环神经网络挖掘问题文本的深层语义特征,利用注意力机制学习多模特征的重要程度.实验结果显示,SOG模型的准确率高达93.98%,优于现有模型的问题质量评估性能.

The article conducts a statistical analysis on the content and ratings of 2 million Stack Overflow problems,and proposes a problem quality assessment model SOG based on recurrent neural networks and attention mechanisms.This model utilizes recurrent neural networks to mine deep semantic features of problem texts,and utilizes attention mechanisms to learn the importance of multimodal features.The experimental results show that our proposed SOG model has an accuracy of up to 93.98%,which is superior to the existing models in problem quality assessment performance.