This repository contains an annotation UI to label self-reported medication timelines in social media posts. The tool was created for crowdsourcing on AWS Mechanical Turk, but can be adapted for use on other platforms.
You can try it out on this demo site. Demo site is under construction.
To run locally: $ python app.py
This task defines entity-timeline annotation as a chronological series of temporal relation labels (e.g. Before, On, After..) for (Entity, Date) tuples. This tool is specialized for the medication-timelines, but the general framework can be applied to other domains.
It is typically challenging to obtain dense and complete relation graphs due to the steep learning curve and worker-fatigue induced from the task complexity. This UI was designed with these challenges in mind, and utilizes an intuitive timeline visualization that increases annotation efficiency while reducing cognitive load.
-
Dynamic temporal reasoning to ensure the timeline is valid
i.e. I can't stop a medication before I start it.
-
Automatically label propogation after every update to the timeline
i.e. if I stopped the medication in 2015, then it can be deduced that 2020 was definitely after my medication intake.
- Text
- Set of entities and their corresponding mentions in the text.
- Set of time-expressions in the text (following the TimeML specification).
