Neo4j today announced the upcoming release of Neo4j 3.5, a native graph platform solution designed for AI-powered, real-time business applications.
Neo4j, a pioneer in data modeling technology, today announced the upcoming release of Neo4j 3.5, a native graph platform solution designed for AI-powered, real-time business applications.
Informed by AI customer deployments for companies like eBay and Caterpillar, Neo4j 3.5 delivers foundational upgrades for AI-powered systems including knowledge graphs, fraud detection, recommendation systems and conversation engines.
“Our customers are pushing the envelope on what can be achieved with graph-powered AI, which we think of as intelligent systems of connection,” said Neo4j CEO and co-founder Emil Eifrem.
Most models and techniques that underpin AI systems are not made to detect connections or relationships within datasets. But property graphs link attributes and complex relationships across the graph, thereby making them the ideal data structure to power machine learning models.
“This relationships-first approach adds context to data, which is key to accurate, well-informed predictions,” Eifrem said. “With Neo4j 3.5 we have worked extensively with our customers to deliver the robustness and scale they need for tomorrow’s graph-powered artificial intelligence systems.”
According to a release issued by the company, the Neo4j 3.5 will include the following upgrades:
- Full-Text Indexing: The 3.5 release brings the power of full-text search into the graph, enabling text-intensive graph applications such as knowledge graphs, metadata management and bill of materials (BoM). It also opens up AI possibilities with natural language processing (NLP).
- Official Language Driver for Go: The Go programming language is gaining in popularity across a variety of applications, including AI, thanks to its ability to support CPU-level parallel processing while remaining simple to read, maintain and deploy. Neo4j 3.5 will support an official Golang driver.
- Graph Algorithms for Enhanced AI: Additions to the Neo4j graph algorithm library in the 3.5 release include unsupervised learning methods such as: Random Walks, Personalized PageRank, Similarities (e.g., Jaccard index), DeepGL, and DeepWalk.
- Framework for Building Driver Apps: The 3.5 release also includes a framework to create Bolt drivers for new languages based on C. Nicknamed “Seabolt,” this framework handles the complexities of interacting with a Neo4j cluster, which allows driver authors to focus on idiomatic language support and higher-level abstractions.
Performance & Scale
- Expanded Native Indexing: Faster data insertion by up to 5x with expanded use of native indexes for all data types (including spatial, temporal and Boolean values), with added support for composite indexing.
- Faster Sorting: Native indexes can now be used for sorting operations, greatly speeding up Cypher queries.
- Improved Handling of Large Write Transactions: A new off-heap transaction memory subsystem – along with optimizations in Neo4j’s Raft-based clustering – greatly improves handling of large writes.
- Subject Alternative Name / Hostname Verification: In private and public cloud environments where IP addresses and hostnames may change, an option to provide multiple IP addresses and servers to an X.509 standard Common Name certificate enables better verification.
- Encryption of Intra-Cluster Discovery Traffic: 100% of intra-cluster traffic is now encrypted, adding more security for multi-data center cloud applications.
Neo4j 3.5 will be available in the fourth quarter of 2018.