Mkgformer: Multimodal Knowledge Graph Data for SIGIR 2022 Benchmarks

Multimodal knowledge graph completion data featuring text and image modalities for link prediction and relation extraction. Released by zjunlp for the SIGIR 2022 conference, it supports the training of hybrid transformer models for knowledge graph enrichment.

Use Cases

• Link prediction to identify missing objects in (subject, relation, ?) triples using multimodal embeddings
• Relation extraction to classify the connection between entities based on text and image inputs
• Multimodal Named Entity Recognition (MNRE) to identify entities within paired visual and textual data

Strengths

• Peer-reviewed at SIGIR 2022
• Provides aligned textual and visual features for knowledge graph entities
• MIT licensed for open research

Limitations

• Specific record counts and schema details are not explicitly listed in the repository metadata
• Highly specialized for the Mkgformer architecture

Provenance

Source

zjunlp (Zhejiang University NLP Group)

Freshness

Last updated April 2025; based on 2022 research.