Towards a deep learning powered query engine for urban planning

Abstract

Urban planning is crucial to sustainable growth. In order for the planners to make informed decisions, data from multiple sources have to be retrieved and cross-referenced efficiently. We discuss the implementation of a query engine which accepts natural language as input, using machine learning and NLP techniques namely word embedding, CNN, rule-based system and NER to produce accurate output enriched with geographical insights to facilitate the planning process. The query engine classifies the query into one of the planning domains, as well as determines the category, location and the size of buffer. Processed results are presented on the ePlanner, which is a map service on the GIS implemented by the Urban Redevelopment Authority (URA) of Singapore.