Hala Madrid
Author
- Yoon Ki Jeong (yoongi0428)
Contents
Problem
NAVER KIN - Sentence Similarity
- Given two sentences, predict whether the two is similar or not.
Model
- "Siamese Recurrent Architectures for Learning Sentence Simialrity" J. Mueller, A. Thyagarajan, 2016
- Modify this architecture using Character Level CNN
- Embedding - Character Level CNN - Fully Connected - Manhattan Distance - Prediction
Detail
- Character Level
- Wide CNN
- No dropout or regularization
- 2 FC layers
- Prediction = exp( -sum( Manhattan Distance of the two outputs ) )
- Loss : MSE
Result
- Accuracy : 0.8023 (6th)