SocraPedia: Enhancing Wikipedia’s Content Quality through Collaborative Large Language Models

Anonymous

TL;DR: This paper employs a dialogue-based approach between Large Language Models (LLMs), utilizing conditional statistics and contentiousness adjustments, to address and correct content inaccuracies and biases.

Abstract: This paper commences by examining the quality challenges present in Wikipedia through a statistical lens. Our analysis reveals that a predominant share, exceeding 90%, of its pages are deemed low in quality by Wikipedia's editorial community, with merely 1.2% achieving the highest quality designation. We scrutinize the rating criteria and pinpoint approximately 5\% of pages that are significant yet underdeveloped---prime candidates for augmentation through LLMs. Our proposed implementation strategy involves crafting algorithmic techniques to assess various quality dimensions, subsequently utilizing SocraSynth to facilitate enhancements in mitigating biases and highlighting nonfactual claims. The efficacy of this approach is validated through trials on a select group of pages characterized by their high importance yet currently low quality.

Paper Type: long

Research Area: Ethics, Bias, and Fairness

Contribution Types: NLP engineering experiment

Languages Studied: English