An NLP-based Malicious Destruction Recognition on Web-based Online Articles using Machine Learning

In today's digital era, online articles have become the primary source of knowledge acquisition, replacing traditional library research. However, the ease of online editing by web users, especially on platforms like Wikipedia, poses a significant challenge of detecting and mitigating malicious destruction in web-based articles. Existing studies have employed web-mining techniques and cross-language learning with the Gradient Tree machine learning algorithm to identify vandalism on specific web contents, achieving up to 80% accuracy. To address this pressing issue, our research proposes a text-based online web articles vandalism detector using a Decision Tree algorithm. Building upon previous studies, we aim to enhance the accuracy performance of the proposed Vandalism Detector System on Wikipedia pages by implementing both basic and advanced machine learning pre-processing techniques with a Logistic Regression learning model. To create a reliable dataset, we cleaned the Wiki Data and merged it with the current Wikidata Vandalism Corpus, comprising four CSV files (edits.csv, annotators.csv, annotations.csv, and and gold-annotators.csv). The resulting cleaned dataset, termed “wikivand.csv,” served as the primary source dictionary, prepared using Python libraries and the Label Binarizer technique. Our proposed model was then trained to predict Yes or No labels, enabling it to identify whether a web article contains vandalism content.