Wikidata’s Worldview: Inspecting an AI Knowledge Pipeline with Semantic Network Analysis

Abstract

As AI systems increasingly depend on structured data to provide meaningful context, understanding the role of knowledge graphs like Wikidata becomes important. A collaborative, multilingual, and free database, Wikidata is at the heart of many AI applications that influence the results of search engines, digital assistants, and automated decision-making systems. It is incumbent on media and communication researchers to understand that machine-readable data is interpretable data and that we must analyze data structure, categorization, and interpretation in the systems that feed the AI knowledge pipeline. This paper provides such an analysis by examining the ontological structure, terminology, and sociocultural biases of Wikidata using semantic network analysis. We expose several problems relating to ambiguous terminology, the classification of concepts, and the social constructions of data entities. We claim that knowledge graphs do not represent objective facts waiting to be transformed into AI communications but instead provide deep cultural assumptions that influence machine communication’s decision-making process. This research calls for radical transparency and criticism of proprietary AI knowledge systems to show their impact on society by allowing researchers to examine the classification architecture of databases used in consumer products.