Document Similarity with Bipartite Graph Matching for Cheapfake and Fake News Detection

Detecting cheapfake and fake news has become increasingly crucial in today’s digital age, given the rapid spread of misinformation across various online platforms. In this paper, we introduce a useful solution that utilizes representation sets of documents to efficiently cluster similar text documents and bipartite graph matching to link pairs of sentences between documents. Our method aids users in identifying and exploring documents with identical or nearly similar content at both the paragraph and sentence levels, thereby helping to detect potential deceptive alterations. We present our system and validate its effectiveness through experiments. Our system successfully retrieves highly similar documents from a repository of 1000 articles, achieving high average similarity scores in user evaluations. Furthermore, experiments show that our system reduces the time required to identify document modifications by at least 40%, even when sentences are shuffled. Our system is expected to enhance information integrity and support more informed decision-making by providing a practical tool for detecting fake and altered news.