Experiment code for the paper by Lukas Brückner, Ioannis Arapakis, and Luis A. Leiva
- Python 3
- TensorFlow 2
- and more as listed in requirements.txt
Install all required packages in a virtual/conda/... environment with pip install -r requirements.txt
Entry point is always run.py
.
Parameters of the experiments and their execution are listed in experiments.sh
.
Examples:
python run.py --max_events 50 --min_events 2 --all_bad
(simple minority voting)python run.py --max_events 50 --min_events 2 --standardize --patience 5 --stop_val val_f1_score --folds 10 --lr 1e-4 --layers 2 --units 100 --dropout 0.3 --batch_size 4 --use_time
(final parameters for custom augmentation with RNN model)
The data used is from Predicting User Engagement with Direct Displays Using Mouse Cursor Information.
The original data is proprietary, but we were allowed to share the cleaned subset of the data that was used in this study (see data/
folder). That way, the experiments can be run as is, and minor modifications and adjustments should be possible.
Please use the following Bibtex when citing this work.
@InProceedings{Bruckner20_abandonment,
author = {Lukas Brückner and Ioannis Arapakis and Luis A. Leiva},
title = {Query Abandonment Prediction with Deep Learning Models of Mouse Cursor Movements},
booktitle = {Proc. CIKM},
year = {2020},
doi = {10.1145/3340531.3412126},
}