license: cc-by-sa-4.0
language:
- en
base_model:
- stanfordnlp/coqa
tags:
- knowledge-graphs
- ambiguity
- qa
- abg-coqa
- coqa
KG-Driven Ambiguity Generation (Abg-CoQA Extension)
This project focuses on generating single-turn ambiguous and unambiguous questions from Knowledge Graphs (KGs) extracted from 1000 distinct stories in the CoQA dataset.
The core objective is to programmatically create ambiguous questions (where a relationship edge has multiple valid parents) and unambiguous questions (where a relationship edge is unique).
Project Overview
- Source Data: 1k distinct stories from CoQA dataset (https://stanfordnlp.github.io/coqa/).
- Method: Knowledge Graph (KG) traversal and relationship mapping.
- Output: A balanced dataset of single-turn ambiguous questions and unambiguous questions.
Algorithms
We utilize graph topology to deterministically generate or avoid ambiguity.
Dataset Structure
The final dataset consists of two primary categories derived from 1k CoQA stories:
- ambiguous: Questions generated via our multi-parent algorithm.
- non-ambiguous: Questions generated via our single-parent algorithm.
References
This work builds upon the concepts of ambiguity in conversational QA presented in Abg-CoQA dataset.
Guo, M., Zhang, M., Reddy, S., & Alikhani, M. (2021). Abg-CoQA: Clarifying Ambiguity in Conversational Question Answering. 3rd Conference on Automated Knowledge Base Construction. https://openreview.net/forum?id=SlDZ1o8FsJU
Original Repo: https://github.com/MeiqiGuo/AKBC2021-Abg-CoQA
Authors: Abhishek Khadanga, Reevu Maity, Pankaj Awasthi, Vikash Choudhary, Aarnov Adhikari, Dhruv Jyoti Das, Raunak Jalan