BUPT-Reasoning-Lab/THEMIS
收藏资源简介:
--- 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)
--- configs: - 配置名称(config_name): default data_files: - 数据集拆分(split): train path: "metadata.jsonl" dataset_info: features: - name: 文件名(file_name) dtype: 字符串(string) - name: 问题ID(question_id) dtype: 字符串(string) - name: 欺诈类型(fraud_type) dtype: 字符串(string) - name: 欺诈子类型(fraud_sub_types) sequence: 字符串(string)序列 - name: 真实标注(ground_truth) sequence: 字符串(string)序列 - name: 掩码索引(mask_indices) sequence: int32序列 - name: 掩码说明(mask_caption) dtype: 字符串(string) - name: 掩码路径(mask_path) dtype: 字符串(string) - name: 图片说明(figure_caption) dtype: 字符串(string) - name: 图片相关语句(figure_related_sentences) dtype: 字符串(string) - name: 不一致部分(inconsistent_part) dtype: 字符串(string) - name: 篡改语句(tampered_sentences) sequence: 字符串(string)序列 - name: 标准语句(golden_sentences) sequence: 字符串(string)序列 - name: 元数据(metadata) struct: - name: 所有图片路径(all_images_paths) sequence: 字符串(string)序列 - name: 已篡改(tampered) dtype: 布尔值(bool) - name: 论文ID(paper_ids) sequence: 字符串(string)序列 - name: 面板ID(panel_ids) sequence: 字符串(string)序列 - name: 图片ID(figure_ids) sequence: 字符串(string)序列 - name: 应用场景(scenario) dtype: 字符串(string) tags: - 学术论文欺诈检测(scientific-paper-fraud-detection) - 视觉欺诈推理(vision-fraud-reasoning) - 欺诈推理(fraud-reasoning) - 多模态大语言模型基准(mllm-benchmark) pretty_name: THEMIS 本仓库收录了论文《THEMIS:面向学术论文欺诈取证的多模态大语言模型(Multimodal Large Language Model,MLLM)全面评估》的相关数据集,论文链接为:https://openreview.net/forum?id=y3UkklvoW9



