Project - Winning Recommendation Engine
Building a winning recommendation engine prototype in 24 hours using Neo4j that resulted in an 8% increase in click-through rates.
- Client
- Beslist.nl (Hackathon)
- Year
- Service
- Innovation, Prototyping
The Challenge
In 2015, Beslist.nl organized a 24-hour Hackathon. The challenge was open: build something cool and valuable.
Our team (2 developers, 1 data scientist, 1 DevOps) identified a gap: Beslist.nl lacked a smart "Recommended for you" feature based on user behavior.
The Solution
We chose Neo4j (Graph Database) to model the relationships between users and the products they viewed. By analyzing overlapping interests, we could recommend products that similar users had clicked on.
- Analyzed visitor clicks to find overlapping interests.
- Used Neo4j graph database to recommend products based on user behavior.
- I developed a high-performance batch importer in Java.
My Role
- Neo4j Implementation: As the only member with Graph DB experience, I designed the schema and queries.
- Performance: I wrote a custom Java Batch Importer to ingest 750,000 daily search queries into the graph in seconds.
- Optimization: Achieved query response times of < 100ms on a dataset covering a full year of traffic.
The Outcome
We secured first place against significantly larger teams—a victory I remain particularly proud of.
- Analyzed visitor clicks to find overlapping interests.
- Used Neo4j graph database to recommend products based on user behavior.
- I developed a high-performance batch importer in Java.
- Result: A significant improvement of +/- 8% clicks compared to previous manual recommendations.
- Neo4j
- Java
- Graph Algorithms
- Rapid Prototyping
- Place
- 1st
- Timeframe
- 24h
- CTR
- +8%
- Latency
- <100ms