Figure 1 – AllegroGraph's hybrid distributed/federated deployment
The basic idea behind AllegroGraph is that, it will a) transform your existing enterprise data into triples, which can then be thought of as either entities or events; and b) store it all inside, effectively, one enormous table comprised of all of your entities, events and sub-events, possibly including entire taxonomies. This method of storage, makes it much easier and simpler to write many queries. For example, picking out a single user or other entity and obtaining a comprehensive, 360-degree view of their relationships within the system is almost trivial, requiring only a single line of SPARQL and essentially treating the graph as a key-value store.
Franz offers multiple methods for deploying AllegroGraph across the enterprise, including federated and distributed deployment models. It also provides a hybrid distributed/federated approach, as shown in Figure 1. This methodology increases performance and scalability by storing replicas of your unshardable data (datasets that must be stored as a single piece, such as knowledge bases, terminology systems, and statistical systems) on each of your machines, then federating that data with the repositories located on that machine. This means that you only need to load in your unshardable data once for each machine during any given query.
Figure 2 – Sending a notification using AI in AllegroGraph
AllegroGraph utilises “multi-mode” artificial intelligence, consisting of both machine learning and a CEP (Complex Event Processing) system developed in Prolog. The idea is to use multiple kinds of AI to estimate the likelihood of future events occurring based on currently observed events stored in a graph. For example, if you are a police force, you might want to monitor the outgoing and incoming calls used by a criminal cell, estimate the probability they are about to meet face to face, and send a notification when that probability rises above a certain threshold. You could accomplish this with the rule displayed in Figure 2.
Although the structure of this rule is set in stone, the parameters (highlighted in yellow) are equipped with confidence intervals, which are calculated based on past observations and events. Moreover, instead of sending a notification, you can create a ‘possible event’ in your graph. This is an event equipped with a probability that represents the likelihood of it occurring or of having occurred. This event is then treated like any other by the AI system, except that it only has a certain likelihood of having happened. This means that AllegroGraph’s AI can produce reasoning based on both known and unknown events.
Finally, AllegroGraph also features native, near real-time multi-master replication and management; multi-modal input from RDF, CSV, JSON, JSON Lines and JSON-LD files; built-in document storage (comparable to MongoDB) with support for graph algorithms and semantics; natural language processing (NLP) and textual analysis, including entity extraction; and extensive support for a variety of data science tools. It is also optimised for use with the newly released Intel Optane line of memory and storage products.
