In this chapter, we built a very deep neural network with the help of TensorFlow in order to detect duplicated questions from the Quora dataset. The project allowed us to discuss, revise, and practice plenty of different topics previously seen in other chapters: TF-IDF, SVD, classic machine learning algorithms, Word2vec and GloVe embeddings, and LSTM models.
In the end, we obtained a model whose achieved accuracy is about 82.5%, a figure that is higher than traditional machine learning approaches and is also near other state-of-the-art deep learning solutions, as reported by the Quora blog.
It should also be noted that the models and approaches discussed in this chapter can easily be applied to any semantic matching problem.
