Understanding how people see, feel, and think about their surroundings (i.e., locality characteristics) is important, as this directly affects well-being, an essential component of sustainable development, and impacts levels of physical activity, which is crucial for public health. Efficient ways for identifying locality characteristics (e.g., disorder, safety) allow stakeholders, including policymakers and urban planners, to improve quality of life.
This 4-hour workshop offers a hands-on opportunity for participants to explore methods using language models, including BERT and our previous work SpaBERT and GeoLM, to capture shared locality characteristics through online text descriptions of places. Additionally, we will introduce in-context learning methods with vision large language models to understand locality characteristics using street view images.
Participants will learn to 1) process text descriptions with locations or street view images and then train tools to learn locality characteristics, 2) visualize localities using the learned characteristics, enabling comparative analysis between results from text and images, and 3) fine-tune tools to predict locality indices such as greenness or other human perception indexes.
This workshop is designed for researchers, practitioners, and anyone interested in learning how to capture locality characteristics from text and images.
Participants are encouraged to bring their own location-based text datasets they would like to test. Basic Python knowledge is required to complete the experiments.
Authors Jina Kim, Guanyu Wang, Michelle Pasco, Yao-Yi Chiang1 University of Minnesota, Minneapolis, MN, USA
