---
configs:
- config_name: default
  data_files:
  - split: train
    path: "metadata.jsonl"
dataset_info:
  features:
    - name: file_name
      dtype: string
    - name: question_id
      dtype: string
    - name: fraud_type
      dtype: string
    - name: fraud_sub_types
      sequence: string
    - name: ground_truth
      sequence: string
    - name: mask_indices
      sequence: int32
    - name: mask_caption
      dtype: string
    - name: mask_path
      dtype: string 
    - name: figure_caption
      dtype: string
    - name: figure_related_sentences
      dtype: string
    - name: inconsistent_part
      dtype: string
    - name: tampered_sentences
      sequence: string
    - name: golden_sentences
      sequence: string
    - name: metadata
      struct:
        - name: all_images_paths
          sequence: string
        - name: tampered
          dtype: bool
        - name: paper_ids
          sequence: string
        - name: panel_ids
          sequence: string
        - name: figure_ids
          sequence: string
        - name: scenario
          dtype: string
tags:
- scientific-paper-fraud-detection
- vision-fraud-reasoning
- fraud-reasoning
- mllm-benchmark
pretty_name: THEMIS
---

This repository contains the data of the paper [THEMIS: Towards Holistic Evaluation of MLLMs for Scientific Paper Fraud Forensics](https://openreview.net/forum?id=y3UkklvoW9)