Relationship Prediction in a Knowledge Graph Embedding Model of the Illicit Antiquities Trade

Abstract

The transnational networks of the illicit and illegal antiquities trade are hard to perceive. We suggest representing the trade as a knowledge graph with multiple kinds of relationships that can be transformed by a neural architecture into a “knowledge graph embedding model.” The result is that the vectorization of the knowledge represented in the graph can be queried for missing “knowledge” of the trade by virtue of the various entities’ proximity in the multidimensional embedding space. In this article, we build a knowledge graph about the antiquities trade using a semantic annotation tool, drawing on the series of articles in the Trafficking Culture Project's online encyclopedia. We then use the AmpliGraph package, a series of tools for supervised machine learning (Costabello et al. 2019) to turn the graph into a knowledge graph embedding model. We query the model to predict new hypotheses and to cluster actors in the trade. The model suggests connections between actors and institutions hitherto unsuspected and not otherwise present in the original knowledge graph. This approach could hold enormous potential for illuminating the hidden corners of the illicit antiquities trade. The same method could be applied to other kinds of archaeological knowledge.

Investigating antiquities trafficking with generative pre-trained transformer (GPT)-3 enabled knowledge graphs: A case study

“…There were initially 946 unique ‘verbs’ or predicates in the statements. We created a knowledge graph embedding model by loading the knowledge graph into Ampligraph 1.4.0 and creating a model with the same parameters as in Graham et al (2023) . 15…”

“…Those suggested connections can then be followed up on via traditional investigative research. This was the technique we employed in our first exploration ( Graham et al, 2023 ).…”

“…In our case, we wanted to achieve something similar in the domain of antiquities trafficking for an experiment presented in Graham et al (2023) and discussed below, but we had nowhere near the required amount of training data to take this approach. Indeed, many potential archaeological applications with these special-purpose models will founder on the rocks of not enough training data.…”

“…The so-called ‘zero-shot learning’ capability caught our attention because it suggests that we do not need to develop a training corpus (see below). We had just conducted an experiment which required manually annotating over 100 articles with subject, object, and predicate relationships taking several months of work, but only identifying a few hundred entities (see Graham et al, 2023 ). With GPT, we saw that there was great potential to automate the annotation of useful texts in this domain saving months of work.…”

“…We transform the knowledge statements into a knowledge graph embedding model which uses a neural network to turn statements into mathematical vectors which can be measured for similarities. We compare the descriptive statistics for the knowledge graph embedding model derived from this automated knowledge graph with the model and its predictions from our earlier experiment that we built manually ( Graham et al, 2023 ) for speed, accuracy, and usability. We believe that this has important implications for the use of our previously-proposed methods for researchers and investigators.…”