To complement the binary choice format of the task in the original release (see below), we further consider a generative version of the task, as demo-ed in this notebook.
Dataset of personal narratives with Advice-Seeking Questions. Version 1.0.
Distributed together with the paper Asking the Right Question: Inferring Advice-Seeking Intentions from Personal Narratives. Liye Fu, Jonathan P. Chang and Cristian Danescu-Niculescu-Mizil. NAACL 2019.
The dataset has been split into train, test, heldout sets, with 8865, 2500, 10000 test instances each. Each set is saved as an individual json file inside the data directory. We have further reserved 500 instances for human annotations. Check inside the annotations directory for more information.
You can explore our dataset as follows:
import json
# read training set
with open("asq_train.json", "r") as f:
data = json.load(f)
# for each individual test instance, the following fields are provided:
data[0]
# expected output
{"narrative": ..., # the question-stripped narrative
"qn1": ..., # candidate question 1 for the given narrative
"qn2": ..., # candidate question 2 for the given narrative
"label": ..., # 0 if qn1 is the correct answer, 1 otherwise
"id": ...} # Reddit post ID of the post from which the instance is obtained