AIKoGAM: An AI-driven Knowledge Graph of the Antiquities Market: Toward Automatised Methods to Identify Illicit Trafficking Networks

The longstanding illicit trafficking of archaeological artefacts has persistently presented a global issue, posing a substantial threat to cultural heritage. This paper introduces an innovative automated system that utilises Natural Language Processing (NLP), Machine Learning (ML), and Social Network Analysis (SNA) to construct a Knowledge Graph for antiquities. The objective is to offer insights into the provenance of artefacts and identify potential instances of illicit trafficking. The paper delineates a comprehensive methodology, from the ontology to the Knowledge Graph. The methodology comprises four distinct phases: the initial phase involves tailoring existing ontologies to match project-specific needs; the second phase centres on selecting appropriate technologies, and scraping and text-mining tools are designed to extract pertinent data from textual sources; the third phase centres in the creation of a robust and accurate Knowledge Graph that captures artefact provenance. The paper suggests employing NLP models, specifically utilising Named Entity Recognition (NER) techniques. These models automatically extract relevant information from the unstructured provenance texts, organising them as events to which both objects and actors participated with their locations and dates. The final phase is concerned with defining and building the Knowledge Graph. The authors explore a property graph model that distinctively represents nodes and relationships, each augmented by associated properties. Employing an SNA approach, the model is projected in multiple network levels of ownership histories (actor-object network) or actor relationships (actor-actor network). This approach reveals patterns within the antiquities market. When integrated with the authors’ recommended strategies such as crowdsourced ontology definition, collaboration with reputable organisations for quality sources, and the application of transfer learning techniques, the suggested approach holds promising implications for the protection of cultural heritage.