relive-qa / README.md
monsoon-nlp's picture
demo scraping recent WikiNews
e2a3756
|
raw
history blame
1.92 kB
metadata
license: mit
task_categories:
  - question-answering
language:
  - en
tags:
  - realtime
  - news

relive-qa

Re-hosting questions from RealtimeQA while we discuss ways to continue posting articles + Q&A using a semi-automated format.

Suggestion: en.WikiNews.org

Scraper version

Prerequisites: pip install openai lxml cssselector requests xmltodict and OpenAI API key

I've added two scripts:

  • scrape.py : base script to load plain text from the latest WikiNews articles
  • scrape_with_openai.py : pass scraped text to OpenAI's GPT-4o to generate questions and answers for each article
  • scrape_morerecent_with_openai.py : scrape recent articles which WikiNews hasn't yet published from Category:May 2024, then use OpenAI's GPT-4o for Q&A

An LLM evaluated on this Q&A could read the WikiNews summary, the links collected from the story, or do its own web searches.

Issues

  • Prompt needs review to avoid making answer choices which are obvious or more specific than alternative choices.
  • Questions and answers should be based on new information in the article, and not general knowledge.
  • Links and the title may give away the answer as the subject of the article, rather than using reading comprehension.
  • WikiNews articles may be niche or local stories, where facts are not known to an LLM unless it reads the specific article

Original paper and dataset

https://github.com/realtimeqa/realtimeqa_public

Citation:

@inproceedings{
kasai2023realtime,
title={RealTime {QA}: What's the Answer Right Now?},
author={Jungo Kasai and Keisuke Sakaguchi and yoichi takahashi and Ronan Le Bras and Akari Asai and Xinyan Velocity Yu and Dragomir Radev and Noah A. Smith and Yejin Choi and Kentaro Inui},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
year={2023},
eprint={2207.13332},
url={https://openreview.net/forum?id=HfKOIPCvsv}
}