Mkg Analogy: Multimodal Knowledge Graph Reasoning for ICLR 2023

Created by zjunlp for ICLR 2023, this dataset supports multimodal analogical reasoning over knowledge graphs. It provides a structured environment for the MARS framework, linking visual and textual data to relational graph structures for reasoning tasks.

Use Cases

• Training models to predict relations between multimodal knowledge graph entities
• Evaluating the zero-shot reasoning capabilities of pre-trained language models using analogical prompts
• Fine-tuning vision-language models on structured relational data from knowledge graphs

Strengths

• Peer-reviewed foundation from ICLR 2023
• Includes both visual and textual modalities for knowledge graph entities
• Released under the permissive MIT license

Limitations

• Lack of explicit record counts in the raw metadata
• Potential domain bias depending on the source knowledge graphs used for entity selection

Provenance

Source

zjunlp (Zhejiang University), ICLR 2023

Collection Method

Curated from knowledge graphs and augmented with multimodal entity data for the MARS framework.

Freshness

Last updated in July 2024; based on ICLR 2023 research.