yinfeiy / PICO-data

PICO-data for Participants

This repo contains raw annotaions for PICO dataset used in paper:

Aggregating and Predicting Sequence Labels from Crowd Annotations An Thanh Nguyen, Byron C. Wallace, Junyi Jessy Li, Ani Nenkova and Matthew Lease Association for Computational Linguistics (ACL), 2017.

A SDK and sample codes are provided for retrieving the annotations.

Description

The dataset is in annotations/, it is splited into 4 parts:

1. train/ contains random selected 3549 abstracts.
2. dev/ contains random selected 500 abstracts.
3. test/ contains random selected 500 abstracts.
4. acl17-test contains 191 abstarcts with annotations by a medical student.

In each folder:

1. PICO-annos-crowdsourcing.json contains annotations from crowd sourced workers.
2. PICO-annos-crowdsourcing-agg.json contains aggregated results from crowd sourced annotations. The aggregation methods are described in Aggregating and Predicting Sequence Labels from Crowd Annotations.:
3. PICO-annos-professional.json for acl17-test only, contains annotations from a medical student.

Environment and dependencies:

• Python 2
• spaCy for basic tokenization etc
• Sample code in src/examples/ folder

cd src
python -m examples.load_annotation