README.md

metadata

license: mit
size_categories:
  - n<1K
task_categories:
  - text-classification
pretty_name: liaisons IBM's Claim Stance Dataset Sample
dataset_info:
  features:
    - name: child_argument
      dtype: string
    - name: parent_argument
      dtype: string
    - name: relation
      dtype: string
  splits:
    - name: binary
      num_bytes: 17806
      num_examples: 110
    - name: ternary
      num_bytes: 11486
      num_examples: 72
  download_size: 19883
  dataset_size: 29292
configs:
  - config_name: default
    data_files:
      - split: binary
        path: data/binary-*
      - split: ternary
        path: data/ternary-*
tags:
  - relation-based argument mining
  - argument mining
  - sample

---

Dataset Card for Liaison's IBM Claim Stance Dataset Sample

Table of Contents

• Dataset Summary
• Dataset Structure
• Licensing Information
• Citation Information

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.

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

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.

Citation Information

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.