An Architecture of Media Analysis Platform for Analyzing Posts on Social Networking Services from Multiple Perspectives
In this paper, we propose an architecture of “Media Analysis Platform (MAP)” that enables multi-perspective analysis of posts on social networking services that may contain potentially misleading or suspicious information. Typically, a post consists of multiple elements such as text and images. Existing methods against false information focus on analyzing each element, for example, detecting AI-generated images or examining claims in a text. An analyst, who analyzes the post, has their own intent and may combine multiple methods for each element. However, a platform that integrates these heterogeneous analysis methods has not yet been established. To address this gap, we propose MAP, a framework for applying and combining diverse analysis methods to each element. MAP stores the original post, decomposed elements, and each analysis result. The architecture allows analysis methods to be connected as modular components. To demonstrate the feasibility of MAP, we implement the prototype system that can analyze posts on X.com and dummy posts. To consistently represent each analysis result in MAP, we also design the SPOTL (Subject, Predicate, Object, Time, Location) data model for the prototype system. With the prototype system, we demonstrate analysis of a hypothetical dummy post and an actual post on X.com and discuss future research directions. We anticipate that MAP will serve as a basis for integrating various analysis methods for posts.