ted-talks-annotation

This is a code for EMNLP-IJCNLP 2019 AnnoNLP Workshop Paper: "Computer Assisted Annotation of Tension Development in TED Talks through Crowdsourcing"

Annotation Tool

An annotation tool used in the paper to annotate the tension development.

Pre-requisites

1. Install and run Mongodb.

Setup

1. To connect the Mongodb, make your own config.py: cp config.sample.py config.py

   • If the default setting of the Mongodb has not been changed, you don't need to modify the config.py

2. Install python requirements: pip install -r requirements.txt

3. Download TED talks videos on the data/video_list.csv: python downloader.py

Usage

1. Run the web-based annotation tool: export PYTHONPATH=.; python annotation/app.py

2. Annotate! 😵

   • Click one of the given options on each video clips. The selected value will be saved to DB automatically with the sentential information.
   • We provided the annotators with this guideline document when using Amazon Mechanical Turk.

3. Export the annotation data: export PYTHONPATH=.; python annotation/dbscript.py --run=export_data

   • The exported data path: data/output/tension.json
   • Example:

   {
       "doc_id": "5db7ac1c88e6da63a07a9c2e",
       "doc_title": "The power of vulnerability",
       "source": "https://www.youtube.com/watch?v=iCvmsMzlF7o",
       "sent_id": "5db7ac3388e6da63a07a9c6a",
       "sent_index": 60,
       "text": "And it turned out to be shame.",
       "labels": [1, 1, 1],
       "start_ts": 279364,
       "end_ts": 281543
   }
   ...