• Remove repeated words.
• Remove repeated emoji.
• Remove links.
• Remove punctuations.
• Remove stop words.
• Replace every emoji with its text.
• Change back abbreviated words like (I'm --> I am).
• Lemmatization: transform words to its root.
• Replace each word by its corresponding vector in [Glove & Word2Vec] word embeddings.
• Using an encoding strategy to encode the two sentences into one vector by LSTM.
• concatenate and feed the two vectors to one node.
• calculate cosine distance and manhattan distance between the two vectors.
• concatenate the 3 nodes and feed them in the final layer with sigmoid activation.
• loss: 0.3663 - acc: 0.8568 - val_loss: 0.5125 - val_acc: 0.7950