Assessing Neutrality of Edits on Wikipedia

This research article presents a study on assessing the neutrality of edits on Wikipedia, utilizing a custom machine learning classifier. The classifier evaluates whether an edit increases, decreases, or does not affect the Neutral Point of View (NPOV) of Wikipedia articles, with a focus on the ongoing conflict in the Gaza Strip as a case study. The authors argue that maintaining article neutrality is crucial in the face of rising disinformation. The article outlines the methodology for building the classifier, which involved selecting 62 articles and extracting 21,530 edits for analysis. The authors manually labeled a subset of these edits to train the classifier and compared its performance to a large language model, GPT-4. Preliminary results indicate that the neural network slightly outperforms the GPT-4 baseline. The article concludes with future steps for enhancing the classifier's accuracy and discusses its potential applications in flagging NPOV-decreasing edits and analyzing editing behavior over time.