Recommendation graph

Latest stable Based on Graph release 0.5.1

Problem and backend

This example turns user-item interactions into candidates and a deterministic ranking that tests can explain. It uses TinkerGraph to isolate modeling from container and network variance. Read core model and TinkerPop first; use the selection guide before production.

Model

Nodes: User/Product
Edges: PURCHASED/FOLLOWS
Key properties: userId, productId, category, quantity, purchasedAt

Prerequisites and release boundary

Use JDK 21, commit 3e0fa7cb9e3bc70c2743aeebda2487f3e45e4907, and the checked-in wrapper. Examples are not published; run this release fixture as a Gradle project from the release source checkout. In a consumer application, select only bluetape4k-dependencies:<ecosystem-version> and add the required graph module without an individual version.

Run and observe

./gradlew :recommendation-examples:test --tests "io.bluetape4k.graph.examples.recommendation.TinkerGraphRecommendationTest"

The tests assert three separate results: product recommendations include p-tripod, follow recommendations include u-carol, and the ranked product result includes p-camera. A failure usually reflects changed interaction weights, candidate filtering, or ranking order.

Reading order

Continue from knowledge-graph, then read linkedin-graph. Also see paired APIs, testing, and operations.

Exercises and production gaps

Add one result-changing edge and assertion; repeat through the suspend API; then run a persistent-backend concrete test serially. Add disconnected and malformed inputs as diagnostics. This fixture does not prove throughput, clustering, authorization, tenant isolation, migration, backup, remote-driver timeout, or index quality.

Release diagrams

These diagrams are loaded directly from README assets published with the 0.5.1 release and pinned to its immutable commit. They describe this manual’s released structure and runtime flows, not later Snapshot changes. Select a preview to open the SVG at the same release commit.