reference: http://www.aclweb.org/anthology/P16-2034
please download GoogleNews-vectors-negative300.bin by yourself when you want to run train.py or train_add_tensorboard.py
new_dataset: a dataset which contains same split with Jiaxin.（contain pairs）
Jiaxin: original dataset(contain connect between pairs, mentions and mention_ids.)
att_bilstm.py: Model
data_helpers.py: Data preprocessing
data_preprocess.py: Functions might be used when we load data (open folder and file/ load data)
dataset_function.py: Functions might be used when we load data (generate a dataset:[[id1,mention1,relation1],[id2,mention2,relation2]...])
train.py: train
train_add_tensorboard.py: train + tensorboard
train_test.py: generate training and testinng dataset
train_test_by_product.py: generate training and testinng dataset by different products . 4|1
new_label: correct mentions' label in pos pair
train_test_update: generate training and testinng dataset after correct label
data_update: some function will be used in correct label