AVerImaTeC-Filled: Precomputed Embeddings for a Multimodal Fact-Checking Knowledge Store

A filled knowledge store with precomputed text and image embeddings for the AVerImaTeC shared task. The dataset supports the VILLAIN system described in the paper 'VILLAIN at AVerImaTeC: Verifying Image-Text Claims via Multi-Agent Collaboration'. It was created by humane-lab and last updated on February 25, 2026.

Use Cases

• Benchmarking multimodal fact-checking models based on the provided knowledge store and embeddings.
• Training or fine-tuning models for image-text claim verification using the precomputed features.
• Researching multi-agent collaboration systems for information verification as referenced in the associated paper.

Strengths

• Precomputed embeddings are provided, which likely reduces computational expense and setup time for users.
• The dataset is associated with a published research paper and a defined shared task (AVerImaTeC), suggesting a research-oriented design.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset scale are unknown, which may limit suitability assessment.

Provenance

Source

humane-lab

Collection Method

Created for the AVerImaTeC shared task; method details are in the associated research paper.

Freshness

Last updated 2026-02-25 07:50:08; freshness should be verified.