Semantic document indexing with generative AI

Abstract

This paper presents new methods for semantic indexing of reference information using generative artificial intelligence. A GPT language model was used to automatically extract descriptors and relationships between them from architectural history documents. A semantic network was created from the extracted descriptors and relationships. A prototype was then developed that made it possible to find relevant documents using the semantic network. Finally, it is shown how the quality of semantic networks can be improved with the help of a swarm of virtual experts. The use of generative artificial intelligence can reduce the workload and costs of semantic indexing. Semantic indexing and the semantic network can contribute to more effective use and dissemination of scientific information by enabling semantic search, easy navigation, and user-friendliness.