A siamese recurrent architecture to learn semantic sentence similarity is implemented using Manhattan distance as a distance metric between pairwise representation of sentences.
The model is trained and tested on Quora Question Pair Dataset with test accuracy of 82%.
Original paper: Siamese Recurrent Architectures for Learning Sentence Similarity
Quora Question Pair Dataset can be downloaded from Kaggle.
- Keras