Leveraging Large Language Models and Deep Learning for Wikipedia Quality Assessment

As of 2024, Wikipedia hosts over 60 million articles across 326 language editions, growing to approximately 1.8 million new articles annually. While traditional encyclopedias rely on domain experts to ensure high content quality, Wikipedia's open-source and collaborative model often leads to issues such as vandalism, link spamming, broken or unavailable links, and non-copyrighted images and graphs. These issues compromise the reliability and quality of Wikipedia articles, causing academic professionals to be wary of using it as an academic reference. However, Wikipedia hosts many high-quality articles. The primary objective of this research is to develop a robust model for analyzing the Information Quality (IQ) of Wikipedia articles, focusing on assessing their factual reliability whereas previous research has mainly focused on statistical techniques. To address the Wikipedia article content quality issue and enhance Wikipedia's credibility as a reliable academic resource, this research proposes an advanced content-based approach to create an embedding layer using the GPT2, Large Language Model(LLM), and convolutional neural network(CNN) to perform the binary classification(high/low IQ). The CNN classification model demonstrates superior accuracy and performance compared to existing methods. Specifically, our approach shows an 87.5% in classification accuracy, effectively distinguishing between high-quality and low-quality articles. Integrating GPT2 and CNN models, our research significantly advances the automated quality assessment of Wikipedia articles. This contributes to the broader goal of making Wikipedia a more credible and dependable resource for academic and research purposes.