AI-based multilingual credibility assessment

Credibility—the perceived trustworthiness and reliability of information or its source—represents a valuable complementary information to content veracity, which is the primary subject of many existing research works. The process of credibility assessment typically builds upon granular information from individual credibility signals which are at first detected and then aggregated into a single credibility label/score. In this chapter, we address credibility assessment in textual data, with a particular emphasis on the automatic detection of credibility signals using diverse natural language processing (NLP) techniques. Special attention is given to challenges of multilinguality and the practical deployment of such systems. We begin by summarising the current state of the art in credibility assessment with textual credibility signals. Building on our own research activities, we then provide concrete examples illustrating how three categories of credibility signals—framing, genre, and persuasion techniques—can be automatically detected using multilingual language models. In addition, we present two real-world use cases: one in which credibility signals assist media professionals in their daily workflows, and another where these signals are used as features for disinformation detection. Finally, drawing on our own experience in textual credibility signals detection and their deployment, we outline open challenges and opportunities that lie ahead. These reflections aim to support and advance this promising yet currently under-researched area.