Assessing Spatio-Temporal Street Name Evolution Using Natural Language Processing and Geospatial Analysis
Topics:
Keywords: street name, language model, spatiotemporal analysis, semantics
Abstract Type: Paper Abstract
Authors:
Yuankun Jiao, University of Minnesota, Minneapolis, MN, USA
Jina Kim, University of Minnesota, Minneapolis, MN, USA
Min Namgung, University of Minnesota, Minneapolis, MN, USA
Johannes H. Uhl, University of Colorado Boulder, Boulder, CO, USA
Keith Burghardt, University of Southern California, Los Angeles, CA, USA
Yao-Yi Chiang, University of Minnesota, Minneapolis, MN, USA
Stefan Leyk, University of Colorado Boulder, Boulder, CO, USA
Kristina Lerman, University of Southern California, Los Angeles, CA, USA
,
,
Abstract
Street names fossilize local culture and topography, providing a nuanced understanding of urban history. It is challenging, however, to automatically extract the context and etymology of street names. We propose an automated pipeline to capture the semantics and origins of street names in over 2,600 counties in the United States and analyze their spatiotemporal development. Given that the semantic meaning of the same street name in different geographical or social backgrounds may vary, we use BERT pretrained on a large corpus of Wikipedia and BookCorpus to extract all potential semantic meanings of individual street names. After extracting the contextualized embeddings of each street name, we identify the shared semantic meanings which co-occur among nearby street names. For example, 'Washington St' refers to the president's name in the politics category when there is 'Lincoln St' nearby. Inspired by the concept of mutual information, we quantify the semantic changes across different periods by the Average Mutual Information Index, which also accounts for the semantic meanings of neighboring streets. In this way, we explore distinctive patterns of semantic changes in the United States and identify relationships between socio-cultural factors and semantic meanings of streets among four census regions over time.
Assessing Spatio-Temporal Street Name Evolution Using Natural Language Processing and Geospatial Analysis
Category
Paper Abstract
