Synthetic Image Detection & Attribution Research Map

An interactive research intelligence map connecting papers, authors, and institutions in synthetic image detection & attribution to reveal research trends, gaps, and collaboration opportunities.

Comment

Inspiration

Synthetic image forensics is growing rapidly, but the research landscape is increasingly difficult to navigate. Papers on AI-generated image detection, source attribution, verification, and related forensic tasks are scattered across different venues and use inconsistent terminology, datasets, evaluation protocols, and institutional metadata.

Existing literature databases are useful for finding individual papers, but they do not clearly show how the field is structured: which problems are being studied, which methods and datasets are commonly used, where the research is being conducted, and which areas remain underexplored.

I built Synthetic Image Detection & Attribution Research Map to turn this fragmented literature into an interactive and auditable research intelligence platform.

What it does

The platform organizes synthetic image detection and attribution research into an interactive world map and searchable knowledge interface.

Users can:

  • Explore papers geographically through their authors' research institutions.
  • Search and filter studies by task, method, dataset, publication year, venue, author, institution, country, and other structured metadata.
  • Inspect paper details, affiliations, publication links, and research classifications.
  • Identify active institutions, geographic research patterns, emerging topics, and gaps in the literature.
  • Trace the provenance of curated metadata instead of relying entirely on automatically collected records.

The project currently covers research related to synthetic image detection, source attribution, source verification, and adjacent forensic problems. Its data model is designed to support continued expansion as the field evolves.

How I built it

The project combines automated scholarly metadata collection with a human-in-the-loop curation workflow.

Paper metadata is collected and reconciled from public scholarly sources. An administrative review system is then used to validate paper identities, normalize publication metadata, resolve author–institution relationships, consolidate institution aliases, review geographic information, and manage ambiguous or conflicting records.

The public website exports the reviewed data into an interactive map and research browser. Papers, authors, institutions, countries, links, and classifications are connected through canonical identifiers so that filtering and aggregation remain consistent.

OpenAI tools supported the development process by helping us reason about the data model, design validation workflows, diagnose interface and metadata issues, generate regression tests, and iteratively improve both the public website and the administrative review system.

Challenges

One of the largest challenges was that scholarly metadata is rarely clean enough for direct visualization.

The same institution may appear under several names. Authors may have multiple affiliations, historical affiliations, or incomplete affiliation records. Preprints and published versions can be duplicated. Country and regional information may be missing or inconsistent. Automatically inferred relationships can also conflict with manually verified evidence.

To address this, I developed provenance-aware curation workflows rather than silently overwriting uncertain data. The system separates automatic suggestions from reviewed mappings, preserves canonical institution identities, records evidence, and prevents unresolved high-impact inconsistencies from being published.

Another challenge was presenting a complex research field without overwhelming users. The interface therefore combines geographic exploration with structured filters and compact paper-level details.

Accomplishments

I am especially proud that the project is not only a visualization but also an auditable research-data workflow.

The platform includes:

  • An interactive global research map.
  • Structured paper and institution search.
  • Canonical institution and alias management.
  • Human-reviewed author–institution mappings.
  • Metadata provenance and evidence tracking.
  • Duplicate-paper and publication-link reconciliation.
  • Geographic review and coordinate validation.
  • Validation checks that protect the public dataset from unresolved inconsistencies.

These components make the platform suitable for both casual exploration and systematic research analysis.

What I have learned

I learned that research mapping is fundamentally a data-quality problem. A visually convincing map can still produce misleading conclusions when institutions are duplicated, affiliations are mismatched, or metadata from different sources is merged without a clear precedent. Reliable research intelligence therefore requires explicit provenance, stable identifiers, review states, and regression testing.

I also learned that human review and automation work best together. Automated sources provide scale, while targeted expert review resolves the cases that matter most.

Ultimately, I want the platform to help researchers understand not only what has already been published, but also what should be studied next.

Built With

Share this project:

Updates