你好。我想分享我对Kaggle Quora比赛（2018-2019）的看法。这种方法在公共场合得分为0.69645，私人的0.70338得分为0.70338，排名第134位。主要的架构思想是使用带有胶囊，注意力和循环学习率的模型和K折。一些代码和想法是从其他Kaggler的帖子中获取的，因此我鼓励您查看它们。
- https://www.*k*ag*gle.com/gmhost/gru-capsule
- https://www.kagg***le.com/christofhenkel/how-to-preprocessing-when-using-embeddings
- https://www.*kagg**le.com/theoviel/improve-your-score-with-some-text-preprecorsing
- https://g*i*thub.*com/mttk/rnn-classifier/blob/master/model.py
- https://www.kag***gle.com/ziliwang/baseline-pytorch-bilstm
- https://www.k*a*g*gle.com/hengzheng/pytorch-starter
