tianke0711 / cikm2020

CIKM2020 - 1 st place solution.

CIKM2020

This is the 1st Place solution of CIKM2020 Analyticup COVID19 Retweet Prediction Challenge - https://competitions.codalab.org/competitions/25276.

MODEL ARCHITECTURE:

HOW TO RUN

python main.py [data_path] [external_data_indicator(1/0)] [path_to_external_data] [path_to_the_pretrained_embedding_file] [pretrained_embedding name] [embedding_size] [path_to_save_model] [path_to_save_test_result] [use_CNN(1/0)]

Here is an example

python main.py /projects/CIKM/ 1 /projects/CIKM/ /projects/CIKM/embeddings/glove.6B.300d.txt glove.840B.300d.txt 300 ./lstm_glove_model.ckpt ./test.predict test.data 0

Train data, train labels and test data should be placed in "data_path" to run the code. 0 in "external_data_indicator" represents that there is no data present in the external data path directory. In that case, the code just runs with train_data,train labels and test data. If use_CNN=0, LSTM is used extrat features from word vectors.

External dataset can be downloaded from the TweetsKB dataset provided here : https://data.gesis.org/tweetskb/. For this exercise ,all tweets from the months of April(month_2020-04.n3.gz) and March 2020(month_2020-03.n3.gz) were used.

Pretrained word vectors tried in this challenge are fasttext, glove840, glovetwitter(200d), fasttextwiki and LexVec. You can use any of these pretrained vectors to run the code.

DEPENDENCIES

Python 3.6
Tensorflow 2.2.0