Keras LSTM

Create Word Matrix

1. Using Glove algorithm to create word matrix for the titles of each paper
2. Then use PCA make every word matrix have the same size (3,50)

Create Train/Validation Dataset

1. X and Y sets are created by citation relationship. if paper A is cited by paper B, then the X here is the word matrix of paper A and the Y is the word matrix of paper B.
2. Then we get X set with size (5967, 3, 50) and Y set with size (5967, 3, 50).
3. Finaly, we shuffle the dataset and split the dataset into train set and validation set.

Create Sequence Model

1. Use Keras LSTM to create a seq2seq model with input size (batchsize, 3, 50) and the output size (batchsize, 3, 50)

Get Predict Matrix

1. We use our obtained train set and validation set to train the model.
2. Use the trained model to get a corresponding predict matrix for the word matrix of each paper, which have the size (3,50).

Obtain Distance Matrix

1. Use the newly obtained predict matrix to create distance matrix as we do last week.

Performace of Word Embedding

• Select the 10 articles which has the most similar titles to paper 2554
• Translate these titles of the 10 articles and compare their real meaning with the title of 2554
• Find the articles which cited 2554 and articles which are cited by 2554
• Compare these articles with the 10 articles