Enriching text analytics with graph databases

Scalable third-generation graph database

TigerGraph has brought a new level of scalability to graph databases with the first natively parallel graph database designed to handle billions of vertices and edges at high speed. Introduced in 2017, TigerGraph is aimed at big data applications in fraud detection, customer profiling, recommendation engines, and healthcare. “The amount of data that people had when graph databases were introduced was only about 2% of what it is now,” said Gaurav Deshponde, VP of marketing at TigerGraph. “Our software was built to analyze very high volumes of data and can traverse through many more levels of connections or ‘hops’ than is typically feasible.”

One healthcare application at Amgen has 5 billion vertices, representing claims, patients, doctors, and products, and 50 billion edges constituting relationships among them. A biotechnology company with an active R&D program, Amgen uses the connectivity of TigerGraph to determine which patients are getting referred to doctors so that they can get access to new therapies.

In the case of customer profiling for ecommerce, many different systems are mined to create a 360-degree view and determine what products should be recommended. “The customer might have filled out a form on the website,” Deshponde continued, “and we use a natural language processing engine to extract keywords and context to build a real-time recommendation engine.” Additional data is incorporated to build out the profile. “We may find information in emails, or input notes from the CRM system, purchase history, and other sources.”

As product attributes from purchases are incorporated into TigerGraph, the recommendation engine can become much more granular, including details such as preferred colors and features. “Some recommendation engines do only a few hops; for example, to books by the same author or products generally like those purchased by a demographically similar segment,” Deshponde pointed out. “TigerGraph can provide true hyper-personalization by storing great detail about product attributes and customer preferences.”