Recommendation engines are driving user engagement across streaming services, e-commerce, social networks, and more. We look at how AI-based approaches like reinforcement learning, "contextual bandits," and "graph AI" are pushing the tech forward.
Retail and media companies are boosting their revenue by deploying recommendation systems that predict what product or content will resonate with a user.
For example, streaming giant Netflix said that “a very significant majority” of its content consumption — previous reports have indicated that this number is around 75% — is driven by its recommendation systems during its Q3’20 results call.
The tech is catching on. Conversations around recommendation systems in earnings calls peaked during the pandemic, when consumers’ online activity skyrocketed. During its Q2’21 call, Nvidia called recommendation engines one of the three “gigantic breakthroughs” in AI in the last several years. The other two — conversational AI and natural language understanding — can also enable and complement powerful recommendation systems.
Below we look at the AI tech powering the next generation of recommendation systems.
GRAPH AI TECHNOLOGY
Most machine learning techniques are designed to work on tabular data or relational databases. But the rise of graph databases such as Amazon Neptune, neo4j, and TigerGraph has created a need for machine learning techniques tailor-made for graphs.
Graph databases consist of nodes (individuals/entities) and edges (the relations between them). In online food ordering, for example, users and the dishes they order can each be nodes, and the connections between them are the edges.
This has given rise to interest in graph neural networks (GNNs) — the application of neural nets to graphs — for recommendation engines across e-commerce, on-demand delivery, and social networks.