The paper on automatic analysis of neutrality in Wikipedia content, authored by our researchers, was presented at IJCAI 2025 (the 34th International Joint Conference on Artificial Intelligence). The event took place on 16-22 August 2025 in Montreal, Canada.
Wikipedia, as one of the most widely used sources of knowledge on the Internet, plays a key role in shaping opinions and providing information to millions of users worldwide. Ensuring its quality is therefore of fundamental importance. One of the most crucial quality criteria is the Neutral Point of View (NPOV), which requires that content be presented objectively, without personal bias or subjective opinions, and with fair and proportional representation of all significant perspectives. However, maintaining this principle is a considerable challenge due to Wikipedia’s open-editing model, where editors may (often unintentionally) introduce their own bias into articles. For this reason, it is essential to develop methods that allow for automated monitoring of NPOV compliance.
Sentiment analysis, commonly used to assess opinions in social media, reviews, or comments, enables the classification of content as positive, neutral, or negative. State-of-the-art models achieve high accuracy but are typically designed for short, informal texts. In contrast, Wikipedia articles are longer and more structured, which limits the direct applicability of such tools. This creates the need for advanced methodologies that can effectively apply sentiment analysis to longer and more complex types of content.
In the presented study, “Cross-Topic Sentiment Analysis of Wikipedia Articles: A Comparative Study of AI Models”, our researchers analyzed nearly 7 million articles from English Wikipedia, applying four sentiment analysis models—two lexicon-based (TextBlob, VADER) and two transformer-based (RoBERTa, DistilBERT). The results revealed that the sentiment of Wikipedia articles varies depending on the topic and that the choice of sentiment model significantly affects the outcomes. Additionally, the study provided a dataset, published on the Hugging Face platform, containing sentiment assessments of Wikipedia articles obtained using the applied models.
This work proposes a new approach to the automatic assessment of Wikipedia’s neutrality—one of the most important and widely used sources of knowledge online. Unlike previous research, which focused on short texts or limited sets of articles, the authors analyze the entirety of English Wikipedia, employing advanced sentiment analysis models. As a result, the study delivers a practical framework for systematically monitoring compliance with the Neutral Point of View (NPOV) principle. Moreover, the proposed methodology can be applied to evaluating the quality and reliability of other online knowledge resources. Authors of the study: Dr. Włodzimierz Lewoniewski, Dr. Milena Stróżyna, Izabela Czumałowska, Aleksandra Wojewoda, Prof. Krzysztof Węcel.