Research

Our research currently focuses on three interrelated areas:

GeoAI Methods and Applications: Our work in this area aims to develop and apply machine learning and deep learning methods for addressing various geographic problems and discovering knowledge from big geo data. Our existing work includes the NeuroTPR model for extracting locations from social media messages posted during a natural disaster, urban areas of interest derived from geotagged Flickr photo data, local place names obtained by mining geotagged housing posts, and the sentiments of people toward their living environment extracted from online neighborhood reviews. Various AI and data science methods are involved in these projects, such as deep neural networks, spatial clustering, topic modeling, and image analysis.

Geo-text and Information Retrieval: Our work in this area aims to develop methods and tools that can facilitate the retrieval and use of geo-text data. Many datasets nowadays contain links between geographic locations and texts (geo-text data for short). In addition, people nowadays often interact with smart devices through natural languages (e.g., Amazon’s Alexa), and there is an increasing demand for extracting spatial information from texts. Our research has produced computational methods for place name disambiguation and EUPEG: a unified platform for evaluating geoparsers. We also developed methods for active information retrieval based on information value theory to support daily tasks and prioritize information for disaster response.

SDIs and Knowledge Graphs: Our work in this area aims to build knowledge graphs and search methods for spatial data infrastructures (SDIs) to support intelligent queries and data sharing. SDIs are playing an even more important role nowadays in supporting the development of smart cities. Meanwhile, knowledge graphs, such as Google’s Knowledge Graph, provide machine understandable knowledge for enhancing the capabilities of today’s GIS and enabling question answering (QA). So far, our research has developed methods for harmonizing heterogeneous metadata and enabling semantic search, and we are exploring more topics in this area.