Graph Gurus 28: An In-database Machine Learning Solution for Real-Time Recommendations

Machine learning methods are improving the performance of recommendation systems. In a typical machine learning workflow, however, data preprocessing and feature engineering happen in the database that stores the training data, while model training is performed in a separate machine learning framework. 

In-database machine learning has the potential to greatly narrow the gap between model ideation and deployment by eliminating the data exporting time before training as well as having the model already in place for production after training. Since the data and model never leave the database, in-database learning supports continuous training and serving using the latest data.

In this webinar, we will use the recommendation system problem, which can be efficiently solved as a graph problem, to demonstrate the in-database training capability of TigerGraph, a native graph database. A hybrid (memory-based + model-based) recommendation system will be implemented in TigerGraph. Specifically, the latent factor model used for recommendation will be trained within the database. 

In this Graph Gurus episode, we will:

• Review multiple widely-used recommendation methods
• Introduce the concept of in-database machine learning 
• Present an in-database machine learning solution for a real time recommendation system,

Speakers: 

• Mingxi Wu, Vice President of Engineering
• Changran Liu, Solution Architect