<![CDATA[The dissemination of fake news (hoaxes) on digital news portals represents a significant challenge in the digital era, as it may mislead the public and reduce trust in circulating information. The rapid and open nature of digital media enables unverified information to spread widely within a short period of time, while manual verification processes require substantial time and effort. This study proposes a semantic similarity-based approach to support real-time news verification using the Multilingual MPNet model. The proposed approach utilizes content text as input, followed by keyword extraction using KeyBERT to represent the core information of the news. The extracted keywords are employed in a news scraping process to obtain comparative news articles from digital news portals. A dataset consisting of 200 Indonesian news articles, including 100 factual news articles and 100 hoax news articles, was used for evaluation. Subsequently, semantic similarity measurement is conducted to evaluate the degree of semantic relevance between the test news and the scraped news. Evaluation metrics were applied to assess the effectiveness of the proposed approach. The findings demonstrate that semantic text representation using Multilingual MPNet effectively supports hoax detection and provides relevant supporting evidence in the form of semantically related news articles, enabling users to access comparative news sources that support the verification process. Experimental results show that the proposed approach achieved an accuracy of 83.5%, precision of 97.18%, recall of 69.0%, F1-score of 80.70%, and an AUC of 0.695, indicating that Multilingual MPNet can effectively support news verification through semantic similarity analysis.]]>
Copyrights © 2026
