The first knowledge graph to combine a culturally rich domain with comprehensive coverage of
text, images, video, and audio. Engineered on schema.org
for semantic interoperability, discoverability, and structural quality.
IMDB4M overcomes the bimodal bottleneck of existing knowledge graphs by integrating
text, images, video, and audio as first-class semantic objects.
📝
Text
100%
Plots, reviews, keywords, genres, cast bios
48.6 avg/movie
🖼️
Images
100%
Stills, posters with captions & entity links
7.9 avg/movie
🎬
Video
99.2%
Trailers with thumbnails, duration, dates
0.99 avg/movie
🎵
Audio
94.1%
Soundtracks with performers & composers
11.2 avg/movie
Knowledge Graph
By the Numbers
A comprehensive resource comprising over 1.8 million RDF triples describing movies,
artists, and their multimodal content.
Metric
Value
RDF Triples
1,815,922
Unique Nodes
660,039
Unique Predicates
58
Seed Movies (fully annotated)
376
Total Movies (after expansion)
50,756
Artists (actors, directors, composers)
5,484
PerformanceRole Instances
232,492
ImageObjects
36,844
Wikidata Alignments (Artists)
4,284 (78.1%)
Entity Types
17
Comparison with Related Work
Dataset
Text
Image
Video
Audio
#Entity
#Relation
MKG-W
14,123
14,463
–
–
15,000
169
MKG-Y
12,305
14,244
–
–
15,000
28
TIVA-KG
11,858
11,636
10,269
2,441
11,858
16
KVC16K
14,822
14,822
14,822
14,822
16,015
4
IMDB4M
385,595
37,220
3,983
4,211
660,039
58
Ontology
Schema.org Foundation
Built on widely adopted vocabulary standards for semantic interoperability
and Web-scale discoverability.
PerformanceRole Pattern
Actor participation captures actor, movie, and character name together,
preserving identity separately from fictional roles.
N-ary Structures
Typed blank nodes with xsd:date, xsd:dateTime, xsd:duration, xsd:integer,
and xsd:decimal for machine-interpretable values.
Two-level Audio
schema:MusicRecording for performed audio artifacts and
schema:MusicComposition for the underlying musical work.
Quality Assurance
Validated & Verified
Systematic validation through SPARQL-based question answering and
human-in-the-loop link verification.
94.4%
F1 Score
99.3%
Precision
90.0%
Recall
99.3%
Query Success
87.2%
YouTube Link Accuracy
0.993
Levenshtein Similarity
Evaluated against 18 competency questions formalized as SPARQL queries,
covering directors, writers, actors, ratings, plots, trailers, soundtracks, images, and more.
Use Cases
Research Applications
IMDB4M enables research across multiple domains in the Semantic Web
and Multimedia communities.
🎥
Movie Recommendation
Content-based recommendation using visual style of posters and acoustic
features of soundtracks.
Audio embeddings from linked YouTube videos
Visual style analysis of movie posters
Temporal features from trailers
🔍
Multimodal QA
Knowledge Graph Question Answering (KGQA) with perceptual grounding
and RAG systems.
Joint reasoning over symbolic and perceptual modalities
Complex queries with reified relations
Multimodal RAG benchmarking
🧩
KG Completion
Multimodal Knowledge Graph Completion including link prediction
and entity alignment.
Infer genre from poster and plot
Cross-platform entity alignment
Multimodal KG embeddings (TransE+)
Reference
Cite This Work
If you use IMDB4M in your research, please cite our paper.
@inproceedings{imdb4m2026,
title = {{IMDB4M}: A Large-Scale Multi-Modal Knowledge Graph of Movies},
author = {Reklos, Ioannis and de Berardinis, Jacopo and Simperl, Elena and Mero{\~n}o-Pe{\~n}uela, Albert},
year = {2026},
note = {Under review}
}
