Implementing MaLSTM on Kaggle's Quora Question Pairs competition


Few months ago I came across a very nice article called Siamese Recurrent Architectures for Learning Sentence Similarity which offers a pretty straightforward approach at the common problem of sentence similarity. Siamese network seem to perform good on similarity tasks and have been used for tasks like sentence semantic similarity, recognizing forged signatures and many more. Word embedding is a modern way to represent words in deep learning models, more about it can be found in this nice blog post. Inputs to the network are zero padded sequences of word indices, these inputs are vectors of fixed length, where the first zeros are being ignored and the non zeros are indices that uniquely identify words.