Schedule

Years 1 and 2:

• Expand and develop a training dataset from the existing corpus of Instagram material, which currently consists of ca. 15 000 photographs, posts, and metadata collected for Huffer and Graham 2017
  • Identify salient markers upon which to create a supervised learning model
  • Identify appropriate deep-learning unsupervised approaches to create an unsupervised model
  • Use Google’s Inception v3 trained model to identify clusters based on shared image features and compare these results with our own models
  • Evaluate these models against a subset of the original data held in reserve
  • Move forward with the appropriate model(s) (objective 1)

Continue to collect posts and metadata over the duration of the project from the targeted social media platforms Instagram and Facebook to keep the research timely (objective 1)

Situate our work in broader context of research into the antiquities trade (objectives 3 & 5)

Years 3 - 5

• Develop and train NN targeting different facets of the material: licit versus illicit, sentiments, provenance, demographics to develop a holistic picture (objectives 1, 2, & 3)
• Develop and release a body of code (under version control) in e.g., Tensorflow and the R statistical computing language, for reproducibility and replicability to other domains of archaeological or cultural materials. Tensorflow is currently the state-of-the-art in exploring NN; the R language and its associated ecosystem of publication workflows has become a standard for digital humanities work. (objectives 1 & 2)
• Social media platforms evolve quickly. We must continually explore the implications of the changing social media ecosystem, its ‘terms of service’, and evolving thought on the ethics of such research (objective 5)

Years 4 - 5

• Push the data and the models further: can we identify likely descendent communities o Associated archival and historical research to support this task (objective 4)
  • Identify other lines of evidence that support the case for intervention from a legal standpoint (objectives 3, 4, & 5)
• Identify the migration of materials across social media platforms by tracking visually similar images (objectives 3, 4, & 5)
• Develop public-facing tutorials that ethically communicate the results of this research to raise awareness with possible descendent communities, law enforcement, academic and professional audiences, and to advocate for policy changes (objectives 3, 4, & 5)