Human-Oriented Fuzzy-Based Assessments of Knowledge Graph Embeddings for Fake News Detection

In the era of information overload, distinguishing between real and fake news is a critical challenge, particularly in public social networking domains. For this purpose, an approach based on the synergy accomplished by a Neurosymbolic AI system, powered with Fuzzy Logic techniques, is introduced to achieve understandable fact-checking classification results. The work proposes a fact-checking approach based on Knowledge Graph Embedding (KGE) techniques. It extracts the involved entities from textual data in the form of triples that, projected in a vector space, form graphs that effectively highlight contextual information. The classification results are interpreted by exploiting fuzzy set modelling, which aims to improve the presentation of the final results. Specifically, we use the Hits@N metric to design fuzzy variables whose linguistic terms reflect news distribution. Then, by exploiting fuzzy rule design, human-like classification performance evaluation is provided. Through experimental evaluation of the benchmark dataset, our approach shows its effectiveness in discriminating between real and fake news, enhanced by straightforward explanations driven by the fuzzy rule design.