News: I am now a Lecturer (Assistant Professor) at the School of Geographical Sciences, University of Bristol. Please refer to my new website for the most updated info.
I am Rui Zhu, a postdoctoral scholar at the Center for Spatial Studies, Department of Geography, University of California, Santa Barbara. My research interests include geospatial semantics, spatial statistics, as well as their broader interactions in geospatial artificial intelligence (Geo-AI). More concretely, I combine theory-informed approach and data-driven appraoch to address geospatial challenges such as geospatial data interoperability, spatial predictions, qualitative spatial reasoning, and so on. My work has been applied to a wide range of applications, indluding urban studies, global health, environmental modeling, as well as humanitarian aids. I am currently working on the KnowWhereGraph (KWG) project funded by NSF Convergence Accelerator Program.
I obtained my Ph.D. degree from the University of California, Santa Barbara in 2020 under the advise of Dr. Krzysztof Janowicz and Dr. Phaedon Kyriakidis. I hold a M.S. degree in Information Sciences from the University of Pittsburgh working with Dr. Hassan Karimi.
Ph.D. in Geography, University of California, Santa Barbara, Santa Barbara, CA
M.S. in Information Sciences, University of Pittsburgh, Pittsburgh, PA
B.S. in Management, Shanxi University of Finance and Economics, China
We apply spatiotemporal thinking into knowledge graph generation, embedding and analysis. Specific research questions include how to represent and incorporate remotely sensed images and maps into knowledge graph, as well as how to leverage geographic distance and direction knowledge into state-of-the-art models of deep learning. Applications include soil science, oceanography and ecology. [This project is in progress and is funded by NSF]
This project investigates the role of directions in GIS. We claim that anisotropicity is the norm in geographic phenomena, not just an exception. Tobler's first law of geography has been generalized accordingly, and multiple new spatial operations are introduced to incorporate directional information with distance. Related paper
A product knowledge graph was generated from technical documents of Siemens Corporation. We further leveraged natural language processing techniques (e.g., BERT) and knowledge graph embedding techniques (e.g., ConvE and GCNN) to answer meaningful and complicated questions based on the built knowledge graph. Another contribution of this project is the introduction of literal-based embedding approach, which incorporates not only object-based nodes but literal information (e.g., countries & dates) into question answering models. [Funded by Siemens Corporation]
We propose the conceptualization and development of a pattern-based, multi-modal framework for OWL ontologies to aid subject matter experts in developing the necessary axiomatization for semantic applications. The graphical end-user components will be matched to SHACL (Shape Constraint Language), that can operate on a domain ontology and whose arguments are selectable and composable, via a graphical user interface. You can check the Demo and Video demo. [Funded by Siemens Corporation]
I interned at Oak Ridge National Laboratory (ORNL) in the summer of 2017. Under the supervision of Dr. Robert N. Steward, I focused on researching the spatio-temporal-thematic patterns of African conflicts from 1997-2017 (one sub-project under the World Spatiotemporal Analytics and Mapping Project-WSTAMP). I also built a self-exciting point process model to predict the occurence of conflicts in Africa. An interactive interface was developed to allow users to explore detected pattern. [Funded by DOE]
Attractiveness of places is modeled to help understand the long-distance travel decision makings of citizens in California. Multiple indicators, including the diversity of the neighborhood in terms of place types, the neighborhood's spatial structures and citizens' perception of neighborhoods, are taken into account in the model. Experiments found that these proposed attractiveness indicators have significant influence on understanding citizens' travel behavior in respect to long-distance travel in California. Related paper
Semantic spatial signatures are proposed to characterize the semantics of place types (e.g., restaurants, hotels, and bars). A series of spatial statistics, such as nearest neighbor distance, Moran's I, and place-based statistics were applied to comprise the spatial signatures. Experiments were conducted to show that spatial signatures are capable to align place types across different gazetteers and location-based social networks. Related paper
Conventional spatial statistics are based on first- or second-order stationarity/statistics. However, due to the complex nature of geographic phenomena, their spatial patterns are frequently beyond simple low-order statistics. Therefore, this project proposed a new conceptualization named as Geo-multipole to model higher-order spatial interactions. A specific approach, multiple-point geostatistics, was employed as one example of Geo-multipole to generate complex urban structures from remotely sensed images. Related paper
In this research, several supervised learning models were applied to select “salient” objects as landmarks. Features were extracted from various sources, including traditional databases, social networks and LiDAR data. The features did not only represent static aspects of the object, but also the interactions between objects and humans. Related paper
Personalized Accessible Maps is a project focusing on providing navigation services to a wide range of users on campus. The prototype of this project was implemented at the University of Pittsburgh. The service provided three kinds of route for different users, including shuttle path, shortest path and personalized path. For personalized path, factors like slope, traffic, and width of the road were considered. In terms of navigation guidance, both turn-by-turn and landmark-based instructions were implemented in this service. Different POIs, service areas, accessible entrances and sidewalks were demonstrated on the maps as well. I was one of the developers for this project. Link
PNS aims to provide navigation guidance for pedestrians. In this service, the sidewalk, rather than road networks, was utilized to generate the route. Due to the smaller granularity of pedestrian navigation compared to vehicle navigation, more challenges were imposed on the techniques of map matching. It required the results of map matching to be highly accurate. To overcome this issue, a new algorithm, chain-code map matching, was implemented in this system. This technique largely improved the accuracy of positioning pedestrians. Relevant paper