Datasets:
license: mit
size_categories:
- n<1K
task_categories:
- text-classification
pretty_name: liaisons IBM's Claim Stance Dataset Sample
dataset_info:
features:
- name: topic
dtype: string
- name: parent_argument
dtype: string
- name: child_argument
dtype: string
- name: relation
dtype: string
splits:
- name: binary
num_bytes: 21165
num_examples: 110
- name: ternary
num_bytes: 13861
num_examples: 72
download_size: 22681
dataset_size: 35026
configs:
- config_name: default
data_files:
- split: binary
path: data/binary-*
- split: ternary
path: data/ternary-*
tags:
- relation-based argument mining
- argument mining
- sample
⚠️ This repository is a part of an academical project for the Heriot-Watt University, no third-party contributions are accepted.
Dataset Card for Liaison's IBM Claim Stance Dataset Sample
Table of Contents
About the Dataset
Dataset Summary
The present dataset is a result of processing the IBM Debater Claim Stance Dataset to create representative samples. The size has been reduced to roughly 100 entries, enabling the benchmarking of models for relation-based argument mining tasks with limited resources. You can also find here the associated benchmarking framework and results.
The sample also modifies the original dataset to achieve a more balanced plurality of stances and topics, and creates a new "unrelated" class in argument relation (following a simple rule-based data augmentation algorithm). Further details on the preprocessing can be found on GitHub.
Dataset Structure
- parent_argument - The first argument that states a position regarding a topic
- child_argument - Another argument that is compared to the parent argument
- relation - The argumentative relation of the child argument to the parent argument. It can either be support/attack in the binary split or support/attack/unrelated in the ternary split
About Contributions
As mentioned earlier, this work is part of an academic project for the validation of my Master's Degree at Heriot-Watt University, preventing me from accepting any contributions until the final release of my project. Thank you for your understanding.
Associated Works
This work is part of a collection of works whose ultimate goal is to deliver a framework to automatically analyze social media content (e.g., X, Reddit) to extract their argumentative value and predict their relations, leveraging Large Language Models' (LLMs) abilities:
- liaisons (the developed client for social media content analysis)
- liaisons-preprocess (the preprocessing of the original IBM dataset)
- liaisons-experiments (the benchmarking framework that the sample is intended to be used with)
- liaisons-experiments-results (the obtained results with this benchmarking)
- mantis-shrimp (the configuration-as-code used to set up my workstation for this project)
Licensing Information
This work includes data from the following sources:
- Wikipedia content licensed under CC BY-SA 3.0: Wikipedia
- IBM content licensed under CC BY-SA 3.0: (c) Copyright IBM 2014. Released under CC-BY-SA 3.0
Modifications and preprocessing have been made to the original data. This derivative work is licensed under the same CC BY-SA 3.0 license.
Credits
Further information about the original dataset can be found on its original HuggingFace page and its associated research papers: Stance Classification of Context-Dependent Claims and Improving Claim Stance Classification with Lexical Knowledge Expansion and Context Utilization.
Special Thanks
I would like to credits Andrew Ireland, my supervisor for this project.
