Russian knowledge graph approach aims to help AI understand context
Researchers in Novosibirsk say a new method can reduce neural network errors and improve the accuracy and reliability of AI-generated answers by organizing information as a knowledge graph.
Scientists in Novosibirsk have developed a method designed to cut down on neural network mistakes, according to a university press office. To boost the precision and dependability of responses, the team structures information as a knowledge graph – not a table or a narrative description, but a system that explicitly shows how elements are connected.
The neural network analyzes these patterns to arrive at a more accurate answer. Knowledge graphs can be combined with large language models to support AI training. Researchers at Novosibirsk State University have developed their own graph library, RAGU (Retrieval-Augmented Graph Utility), built around a step-by-step process for constructing a structured network.
Step-by-step toward accuracy
One of the authors, Ivan Bondarenko, said that traditional approaches to building knowledge graphs efficiently require extremely large language models, with up to around 32 billion parameters.
