The principles of knowledge representation that underpin the BIBFRAME RDF/OWL ontology provide a powerful foundation for reasoning over bibliographic data and inferring new connections utilizing linked open data resources across the web. Earlier Semantic Web research relied on reasoning-based approaches from "Good Old-Fashioned AI" (GOFAI), employing symbolic logic and aspects of first-order logic for reasoning. Much of what was once considered science fiction in the era of symbolic logic-based agents is now poised to become science fact with advances in generative AI that can be used for powering the powering agents of the semantic web.

The departure point of this work is in bridging aspects of symbolic AI with generative AI for improving BIBFRAME Discovery Services.

This presentation explores how knowledge representation and reasoning principles can be fully exploited by integrating Generative AI into BIBFRAME-based discovery within libraries and beyond. Specifically, the talk demonstrates how retrieval-augmented generative chat models (RAG) enhance bibliographic exploration by surfacing the interconnections within BIBFRAME resources along with linked open data endpoints (demo video and screenshot). The RAG approach is designed to drive use of collections, services, and expertise within the library.

Current experimentation utilizes GraphQL endpoints from Share-VDE (BIBFRAME data) and SPARQL queries for grounding LLM knowledge with Wikidata. This is done by careful curation of prompts that direct the RAG agent in the LLM to 1) refer follow up questions to librarians and the librarian-based library chat service; 2) acknowledge limited access to data (both Share-VDE and Wikidata may be incomplete); 3) utilize the user context of the author page where the chat widget is embedded; and 4) use the context of the chat widget as the starting point for BIBFRAME graph exploration.