PyTorch implementation and dataset of Video Highlight Prediction Using Audience Chat Reactions, 2017 EMNLP.

In the Google Drive, there are two new files and one new directory. cutChatRoom.py is the script I used to cut the twitch chats to separate games.

 python cutChatRoom.py (will read cut_video3.txt to generate chat file for each video and store in final_data)

• ffmpeg

• PyTorch (https://github.com/pytorch/pytorch)
• PyTorch/Vision (https://github.com/pytorch/vision)
• MoviePy (https://github.com/Zulko/moviepy)

• https://drive.google.com/drive/folders/0By9LEMeCDdboVDlHTDlqQUNHMnc?usp=sharing
• Before training, it is important to convert the encoding of video from h.264 to mpeg and resize the resolution. Check the compressVideo.py, if you need to change the input resolution of vision models.

cd EMNLP17_Twitch_LOL
python compressVideo.py (This requires around 378 GB)

• Check the run.sh file. This script contains all the configurations of the experiments.
• If you want to evaluate the trained model, simply use the same command used in training and add -e.

Please cite this paper in your publications if it helps your research:

@inproceedings{fu2017highlight,
  title = {Video Highlight Prediction Using Audience Chat Reactions},
  author = {Cheng-Yang Fu, Joon Lee, Mohit Bansal and Alexander C. Berg},
  booktitle = {EMNLP},
  year = {2017}
}