Semantic Similarity

Objective

In this paper, we tackle the task of creating different models to classify semantic equivalence on question pairs. We utilize Quora’s 400,000 question pairs as released as part of a Kaggle competition. We utilize the Manhatten LSTM model that achieved state-of-the-art performance in this task, but also compare it against a universal sentence encoder that was recently released by Google. Our paper finds that Google’s sentence encoder was outperformed by a Siamese LSTM with Word2Vec embeddings.