GPTKB 2.0: Direct Construction of Disambiguated Knowledge Bases from Large Language Models
GPTKB 2.0 is a methodology for constructing disambiguated knowledge bases directly from large language models, incorporating on-the-fly disambiguation of entities, relations, and classes. It was executed at scale, producing a materialized KB with over 1 million disambiguated entities and 38.4 million triples, representing the first million-scale LLM-native KB with explicit internal canonicalization. The system is available at https://gptkb.org/.
Automated Knowledge Base Construction (AKBC) is a core NLP task, and recent work proposes generating knowledge bases directly from large language models (LLMs), treating the model itself as the knowledge source. However, LLMs natively possess no representation of entities, leading to duplicate entries as well as conflations. GPTKB 2.0 is a methodology for constructing disambiguated KBs directly from LLMs. It incorporates on-the-fly disambiguation of entities, relations and classes, and is meticulously designed to satisfy both scalability and disambiguation accuracy. The central design decisions are analyzed and the trade-offs between accuracy, scale, and cost are characterized. GPTKB 2.0 was executed at scale, obtaining a materialized KB containing over 1M disambiguated entities and 38.4M triples. This represents the first million-scale LLM-native KB with explicit internal canonicalization of entities, relations, and classes, a significant departure from prior Wikimedia-centric works. GPTKB 2.0 is available at https://gptkb.org/.
GPTKB 2.0 introduces on-the-fly disambiguation of entities, relations, and classes during KB construction from LLMs, addressing the lack of native entity representation in LLMs. The methodology balances scalability and disambiguation accuracy, with design trade-offs analyzed between accuracy, scale, and cost. The resulting KB is the first million-scale LLM-native KB with explicit internal canonicalization, moving beyond reliance on Wikimedia-based knowledge sources.
This work signals a shift in knowledge base construction from curated sources like Wikipedia to direct extraction from LLMs, potentially reducing dependency on human-curated knowledge and enabling more dynamic, scalable KBs. The availability of a disambiguated, million-entity KB could accelerate downstream NLP applications and enterprise knowledge management.
GPTKB 2.0 offers a scalable, automated approach to building high-quality knowledge bases, which can enhance search, question answering, and enterprise AI systems. Its disambiguation capabilities reduce manual curation costs and improve data reliability, making it valuable for companies needing large-scale, up-to-date knowledge graphs.
Observable next signals include adoption of GPTKB 2.0 in research and commercial applications, further scaling to larger KBs, integration with retrieval-augmented generation (RAG) systems, and comparisons with traditional KB construction methods. Potential improvements in disambiguation accuracy and cost efficiency may follow.