Natural Language Processing of User Queries using Post-Wildfire Disasters Concept Maps and Graphs to Derive Relevant Variables

Abstract:

Extracting geophysical variables that are related to a user query is essential to developing an effective search engine for Earth Science data. Currently, the NASA Goddard Earth Sciences Data and Information Services Center (GES DISC) search engine does not return all relevant datasets when search terms are input. Thus, it is desirable to link these search terms to related external (to the GES DISC) variables that can then be linked to their corresponding internal GES DISC variables. In this work, search terms related to disasters and specifically, post-wildfire events, were studied. By establishing hierarchical relations between wildfire-related events, including post-fire hazards, concept maps were created to serve as a domain model. We showed how such domain models can be used to contextualize user search queries. We applied machine learning technologies, including graphs and natural language processing (NLP), to the concept maps to provide better correspondence among concept map entities and related geophysical variables. Wildfire-related user searches could be successfully connected with external GES DISC variables using NLP. This correspondence is a key step towards providing more relevant results in response to user search queries. By expanding the graphs, this work can easily be generalized to all natural disasters.