Demonstrates ArcadeDB's multi-model capabilities using an academic research knowledge graph that unifies four signal types in a single database:
- Graph traversal — co-authorship network and citation graph multi-hop queries
- Vector similarity — semantic paper search using embeddings
- Full-text search — keyword search across paper abstracts
- Time-series — trending papers by cumulative citation activity
- Docker and Docker Compose
curlandjq- Java 21+ and Maven 3.x (for the Java demo)
docker compose up -d./setup.shThis creates the KnowledgeGraph database, applies the schema, and inserts sample data.
./queries/queries.shcd java
mvn package -q
java -jar target/knowledge-graph.jar| Type | Kind | Key properties |
|---|---|---|
Researcher |
Vertex | id, name, embedding |
Paper |
Vertex | id, title, abstract, year, embedding |
Topic |
Vertex | id, name, embedding |
Institution |
Vertex | id, name |
CO_AUTHORED |
Edge | Researcher → Paper |
CITES |
Edge | Paper → Paper |
COVERS |
Edge | Paper → Topic |
AFFILIATED_WITH |
Edge | Researcher → Institution |
PaperActivity |
Document | paperId, citationCount, ts |
| # | Pattern | Language | Pillar |
|---|---|---|---|
| 1 | Co-authorship Network | Cypher | Graph |
| 2 | Semantic Paper Search | SQL + vectorNeighbors | Vector |
| 3 | Full-Text Abstract Search | SQL + SEARCH_INDEX | Full-text |
| 4 | Trending Papers | SQL | Time-series |
| 5 | GraphRAG Hybrid | SQL + MATCH | Graph + Vector |
- 5 researchers, 4 institutions, 6 topics, 10 papers
- 11 CO_AUTHORED edges, 10 CITES edges, 15 COVERS edges, 5 AFFILIATED_WITH edges
- 10 PaperActivity records (citation events for time-series queries)
- 4-dimensional embeddings throughout
This use case targets ArcadeDB 26.3.1. Vector similarity uses
vectorNeighbors('IndexName[property]', vector, k) with an LSM_VECTOR index.
Full-text search uses SEARCH_INDEX('Paper[abstract]', 'query') against a FULL_TEXT index on
Paper(abstract).