Perspectrum

A dataset of claims, perspectives and evidences

Introduction

With an increasing "information pollution" presence in our daily lives, there is a significant demand for better organization and access to information.
In order to facilitate research and progress on this problem, we built Perspectrum, a dataset of claims, perspectives and evidences. We hope that this dataset would bring more attention to this important problem and would speed of progress in this important direction.

Releases and Downloads

The entire dataset contains 1k claims, accompanied with pools of 10k and 8k perspective sentences and evidence paragraphs, respectively.

To find some details on the structure of the dataset conventions, check out this link.

Example

Given a natural language claim, a hypothetical system is expected to discover various perspectives that are substantiated with evidence and their stance with respect to the claim. The supporting and opposing perspectives are indicated with green and red colors, respectively.

...

Explore the dataset

Complete list of claims, perspectives and evidence

Visualize a random subset of the dataset by their topic.

Graph visualization of some example claims (takes couple seconds to load :)

Paper

Work is published under the Creative Commons (BY-SA) license.
If you find this data helpful in your work, please cite this paper: