Document Object Model (DOM) Graph RAG

The Document Object Model Graph RAG project introduces an innovative approach to Retrieval-Augmented Generation (RAG) in artificial intelligence, overcoming the limitations of traditional vector databases. By leveraging Document Object Model (DOM) structures and knowledge graphs, it grounds the large language models used in intelligent applications and enhances the accuracy and reliability of AI-generated content, especially in high-stakes environments. Unlike vector-based systems that struggle with context loss, the DOM Graph RAG model preserves content integrity and enables multi-hop reasoning. 

Utilizing the DITA schema, DOM Graph RAG ensures effective retrieval and response generation. Moreover, the model incorporates neuro-symbolic reasoning, combining neural networks’ pattern recognition with logical reasoning. This allows for continuous updates to dynamic content, maintaining its relevance and accuracy.

This paper showcases a sample chatbot that was built to assist in a real-life enterprise scenario. This chatbot, created for a tax software company, provides accurate, context-aware answers while reducing reliance on large language models, thus lowering computational costs. The DOM Graph RAG application has capabilities such as assisted prompt engineering, question-answering follow up, semantic search, content recommendations, and summarize takeaways.

Read this paper to understand Document Object Model Graph RAG, how it is built, its unique advantages in comparison to a purely vector-based RAG, and what a sample application looks like for an enterprise. 

Authors:
Michael Iantosca
Senior Director of Knowledge Platforms and Engineering
Avalara Inc.

Helmut Nagy
Chief Product Officer
Semantic Web Company GmbH

William Sandri
Data & Knowledge Engineer
Semantic Web Company GmbH