Dr. Dalia Varanka, USGS
Large Language Models (LLM) depend on the availability of sufficient text data and primarily statistical processing techniques aided by applied ontologies to improve information search and extraction over large volumes of data sources. The extraction of relevant concepts and relationships of the input data consisting of feature type definitions can be structured as graph property patterns based on dependency relations among parts of speech. Dependency relations determine the domain and range classes forming patterns for information retrieval. Among these patterns are relative spatial and topological relations, nominal and ordinal data, and more. The application of topographic concepts organized by grammar patterns support retrieval augmented generation to improve LLM query results. The body of work involving semantic ontology and a syntactic lexicon serve similar traditional functions as topographic base maps have in cartography by utilizing the the user’s input content of topographic features and relations within their spatial context among each other while allowing extensions to other thematic data from the trained LLM.
Please register to join the presentation
If the link doesn't work, try this URL: https://docs.google.com/forms/d/e/1FAIpQLSc7LS2WFRG0ONVAWSXlXMW7Kw-FOAafLxdULgYBijjN6AjOHw/viewform?usp=header