Back to datasets
Dataset assetOpen Source CommunityMultimodal AnalysisNeural Relation Extraction
MNRE
MNRE is a challenging multimodal dataset for neural relation extraction using visual evidence in social‑media posts. The dataset requires understanding both visual and textual modalities and aims to push multimodal alignment toward higher semantic levels.
Source
github
Created
Apr 4, 2021
Updated
Nov 23, 2021
Signals
228 views
Availability
Linked source ready
Overview
Dataset description and usage context
MNRE Dataset Overview
Dataset Versions
- MNRE‑2: A trimmed version released on 2021‑06‑22, consolidating ambiguous categories and adding more supporting samples. The original version has been moved to Version‑1.
Objectives
- Introduce a new task: multimodal neural relation extraction.
- Provide the MNRE dataset for model evaluation.
Statistics
Comparison with Prior NRE Datasets
| Dataset | # Images | # Words | # Sentences | # Entities | # Relations | # Instances |
|---|---|---|---|---|---|---|
| SemEval‑2010 Task 8 | - | 205k | 10,717 | 21,434 | 9 | 8,853 |
| ACE 2003‑2004 | - | 297k | 12,783 | 46,108 | 24 | 16,771 |
| TACRED | - | 1,823k | 53,791 | 152,527 | 41 | 21,773 |
| FewRel | - | 1,397k | 56,109 | 72,124 | 100 | 70,000 |
| MNRE | 9,201 | 258k | 9,201 | 30,970 | 23 | 15,485 |
Category Distribution
- Relations are annotated according to entity types, e.g., person‑person relations such as "alumni", "spouse", "relative", etc.
Data Collection
- Sources: Twitter15, Twitter17, and a custom Twitter crawl.
- Entities and types were extracted using the pretrained NER tool elmo.
Usage
- Textual relation files are located in
./mnre_txt/. - Image data can be downloaded here.
- Each line contains: text, head entity and position, tail entity and position, image ID, relation and entity categories.
Case Studies
- Demonstrates how visual information benefits relation extraction, covering object and attribute recognition as well as person‑person and person‑object interactions.
Need downstream help?
Pair the dataset with AI analysis and content workflows.
Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.