kentoh - Fotolia

Artfinder deploys Neo4j graph database to personalise recommendations

Can a machine understand art? Not yet and maybe never, so the Artfinder art website took a different approach, using a Neo4j graph database instead

Artfinder, the online marketplace for art, has revamped its home page with a recommendation engine powered by the Neo4j graph database to give art lovers personal recommendations on their home page.

The three-year-old site for affordable art, which has 6,000 artists across 97 countries, and half a million users, wanted to present to customers works of art that matched their tastes.

The business’s CTO David Tilleyshort said: “We’ve known we needed a recommendation engine for a long time. We had a basic version on our product pages to enable customers to see similar works of art.”

But the site wanted to captivate customers in a way that presented art that suited their tastes.

It initially looked at a classification system based on image processing to pull out relevant features in the art work, which could then be encoded in machine-learning algorithms to present art works with similar features.

It didn’t work. “Machine learning doesn’t take in the human element,” said Tilleyshort.

The site then looked at how to use a graph database to identify relationships between people with common interests. “If people have similar tastes, it makes recommendations,” Tilleyshort explained.

Collector behaviour analysis

Artfinder’s platform allows customers to follow artists. They can also ‘favourite’ works and add them to collections. Tilleyshort said: “This was the starting point of behavioural analysis. We track customers and artists and artworks, creating interesting relationships between them.”

There are a number of graph database tools on the market but Artfinder selected Neo Technologies for its powerful language. “Every graph database has its own language. Neo4j is very expressive and human-readable while other graph database languages look quite abstract,” he explained.

According to Tilleyshort, there is a significant user community for Neo4j. “It is imperative to be able to get help from a wider community. Neo4j uses a new language and there are some gotchas. You need to profile your queries and optimise them, so having a community really helps.”

Read more about graph databases

In terms of implementation and deployment, Neo Technologies provides a docker container for Neo4j, which Artfinder has used as part of its containerised architecture, which runs on the Amazon Web Services cloud.

The My Artfinder recommendation engine began running in early January 2015. It was officially launched on 23 February. The graph database allows Artfinder to constantly develop its understanding of what customers do and do not like. Tilleyshort said: “We built 15 million relationships over two years.”

He added that Artfinder can now present more art work to people on their personal home pages. “Customers can go down a rabbit hole and follow more artists, which means we get more relevant relationships going forward and also improve our customer engagement.”

Read more on Database management