ME-GCN: Multi-dimensional Edge-Embedded Graph Convolutional Networks for Semi-supervised Text Classification
Official Implementation for ICLR 2022 on DLG4NLP (https://arxiv.org/abs/2204.04618)
Wang, K.*, Han, C.*, Long, S., & Poon, J. (2022).
ME-GCN: Multi-dimensional Edge-Embedded Graph Convolutional Networks for Semi-supervised Text Classification
In proceeding of ICLR 2022 on DLG4NLP
You can simply run the code with your data using final.ipynb, remember to fill in your dataset into a list of documents/labels
# ALL_INPUT = a list of input sentences
# ALL_OUTPUT = a list of output labels
# example
# original_train_sentences = ['this is sample 1','this is sample 2']
# original_labels_train = ['postive','negative']Also, some other parameters can be modified
WORD_EMBEDDING = 0 # 0=word2vec, 1=fasttext, 2=glove, this is for word node embedding
DIM = 25 # number of streams
D2D_THRESHOLD = 15 # two documents sharing more than 15 words will have edges between them
POOLING = "max" # "max","min","avg", the final pooling method
ALL_USED = False # If True, only use partial of the dataset
USED_SIZE = 3000 # The number of samples used
TRAIN_PORTION = 0.01 # The proportion of labelled data
HIDDEN_DIM = 25
DROP_OUT = 0.5
LR = 0.002
WEIGHT_DECAY = 0
EPOCH = 2000
EARLY_STOPPING = 100
VAL_PORTION = 0.1
REMOVE_LESS_FREQUENT = 5
NUM_TEST = 5