This repository contains the code and other resources for our proposed solution for the COVID-19 Retweet Prediction Challenge at CIKM2020 Analyticup. The proposed solution ranked 4th on the final testing leaderboard among 20 teams (51 teams in the validation phase). The pre-print of our report is available here.
- Download relevant data files from here with password: cgfm, unpack any tar.gz files in the model directory
- model directory contains trained models
- tmp directory contains extracted data for training and prediction
- data directory contains raw data provided by the challenge origranizers
- Check and install relevant packages based on requrirements.txt -
pip install requirements.txt - All experiments are run with a commodity laptop (Intel CoreI5 processor at 2.6 GHz, 8GB of RAM, and with 200GB swap space)
test.pyloads the test data and run different RERFs to get the prediction results in the output directory and ensembles those results. Afterwards, it applied personalized patching to update the final prediction results for users having a sufficient number of tweets in the training set.train.pycontains code for training different RERFs, which are used in the test.py.utils.pycontains necessary utility functions for train.py and test.py.
| No. | RERF Details | Weight |
|---|---|---|
| 1 | LinearRegression(fit_intercept=False) RandomForestRegressor(n_estimators=500, max_depth=20, random_state=7) |
1 |
| 2 | MLPRegressor(batch_size=1024, hidden_layer_sizes=(64,32,16,8,8), random_state=7) RandomForestRegressor(n_estimators=1000, max_depth=18, random_state=7) |
1 |
| 3 | MLPRegressor(batch_size=2048, hidden_layer_sizes=(128,64,32,16,8,8), random_state=77) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=77) |
1 |
| 4 | MLPRegressor(batch_size=2048, hidden_layer_sizes=(128,64,32,16,8), random_state=18) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=18) |
1 |
| 5 | MLPRegressor(batch_size=2048, hidden_layer_sizes=(64,16,8), random_state=19) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=19) |
1 |
| 6 | MLPRegressor(batch_size=2048, hidden_layer_sizes=(128,64,16,8), random_state=20) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=20) |
1 |
| 7 | MLPRegressor(batch_size=4096, hidden_layer_sizes=(128,64,16,8,4), random_state=201) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=201) |
1 |
| 8 | MLPRegressor(batch_size=4096, hidden_layer_sizes=(128,64,32,8), random_state=211) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=211) |
2 |
| 9 | MLPRegressor(batch_size=4096, hidden_layer_sizes=(128,64,32,8,8), random_state=22) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=22) |
1 |
| 10 | MLPRegressor(batch_size=4096, hidden_layer_sizes=(128,64,32,16,8), random_state=27) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=27) |
1 |
| 11 | MLPRegressor(batch_size=4096, hidden_layer_sizes=(128,64,32,32,16), random_state=211) RandomForestRegressor(n_estimators=500, max_depth=18, random_state=28) |
1 |
| 12 | xDeepFM() RandomForestRegressor(n_estimators=500, max_depth=16, random_state=28) |
1 |
| 13 | DeepFM() RandomForestRegressor(n_estimators=500, max_depth=16, random_state=29) |
2 |
| 14 | DeepFM() RandomForestRegressor(n_estimators=500, max_depth=16, random_state=28) |
1 |
Guangyuan Piao and Weipeng Huang, "Regression-enhanced Random Forests with Personalized Patching for COVID-19 Retweet Prediction", CIKM Analyticup Workshop at CIKM'20, Galway, Ireland, 2020. [PDF] [bibtex] (CKIM Analyticup Proceedings-http://ceur-ws.org/Vol-2881/)