Know Where Graph Logo
Back to All Prototypes

Humanitarian Relief Graph

Background Imagery - County Map of the USA with a vectored web of lines and nodes wrapping across the country, signifying connections

Disasters are complex and dynamic situations requiring humanitarian organizations to evaluate and respond rapidly to many different issues simultaneously. Often what is most needed to improve effective response is quick access to the right experts at the right time. To assist in identifying people with expertise in humanitarian aid and relief, with a particular focus on health and the health care impacts of disasters, we are working with Direct Relief to showcase how our knowledge graph can give them rapid access to area briefings, including previous events, and physical properties of the affected regions, including climate, transportation infrastructure, and other key data points.

The 2020 hurricane season provides a good example. Many people residing in the state of Louisiana became internally displaced as a result of the impact of Hurricane Laura, a category 4 storm which made landfall in late August. Only a few weeks later, Hurricane Delta made landfall in nearly the same location. What would have been a serious series of events under any circumstances, was made additionally challenging due to COVID-19. The pandemic affected people's willingness to evacuate or seek refuge in emergency shelters. In complex situations like this, a relief organization has to acquire situational intelligence rapidly about the affected region, its population, its previous impactful events (here Hurricane Laura), its local COVID-19 community spread forecasts, and so on. Then, based on an understanding of the particular situation, relief organizations must identify experts on the ground. Each of these steps takes multiple hours and meetings when time is of the essence. With KnowWhereGraph, we can provide county-level forecasts from over fifty COVID-19 models within minutes, link them to population data, and provide information about all recent and previous relevant events that affected the area within seconds. These events do not need to be of the same type. For instance, in some regions of the world because disasters can compound the spread of diseases, it is essential to identify previous endemic disease outbreaks whenever a major storm hits,. Since our data is integrated spatially, any information about climate, epidemiology, events, transportation, loss, and affected soils and crops, is just a few graph links away. In the following section, we will focus on the identification of experts and their expertise.

The KnowWhereGraph contains semantically-defined, graph-ready data on several topics: including contact information for individuals with (1) expertise in specific emergency medical techniques and diseases; (2) expertise in types of hazards and disasters that could lead to specific medical concerns (e.g., floods leading to outbreaks of water-borne disease); and (3) experience (when relevant) in specific geographic regions, where cultural, religious, and political conditions may become significant factors in providing effective disaster relief. Such data do not yet appear to exist in organized, accessible formats over the Web or distributed sources, beyond personal social and organizational networks. Currently, our data is restricted to the United States. We aim at a global coverage (at varying spatial resolutions) by the end of 2022.

The typical types of expertises that a humanitarian organization attempts to find are diverse, but all have a close relation with the targeted disaster or its type. For example, in the face of Hurricane Katrina, experts at a local government agency (e.g., Smith in Figure \ref{fig:diagmra}) who have first-hand data about the disaster event and are capable of contacting leaders in the impacted community, are invaluable resources for humanitarian organizations to consult with in order to respond effectively to the disaster. In this case, the expertise is about the concrete disaster itself, i.e., Hurricane Katrina. In contrast, expertise can also be for a disaster topic, in general. For example, as shown in Figure 1, Wang (green box) is a professor who researches hurricanes in general. She has studied the mechanisms and impacts of most major global hurricanes in the past 20 years. Due to her systematic investigation of hurricanes, humanitarian aid specialists are able to rely on her expertise to estimate, for instance, the impact of a hurricane on the local public health system, thus facilitating a more precise and timely allocation of medical relief packages to the right group of people.

In addition, humanitarian aid expertise is needed in topics that are frequently associated with specific types of disasters, such as drought in developing nations often leading to famine. For a different type of example consider a graduate student, Johnson, studying human geography. His dissertation covers human migration, as well as its economic, social and cultural impact, as a result of Hurricane Katrina. Unlike Smith and Wang, Johnson's expertise does not help with the immediate disaster response, but instead provides insights on its long term impacts so that humanitarian aid and local governments, if necessary, can be planned in advance.

Interestingly, regardless of type, expertise can also be spatiotemporally scoped. It is only relevant within some period of time and for specific regions. For instance, an expert may have worked on a topic in the past, but changed their research direction and interests. Similarly, many experts restrict their area of interest spatially, e.g., working on vector-borne diseases in Brazil. An example of a combination of these aspects may be an expert that worked on droughts in Eritrea but has now shifted their study area to China given its rapidly growing need for water. These aspects of relevant expertise are formally modeled in our Knowledge Graph, as depicted in Figure 1.

Examples of experts and their expertises in KnowWhere-Graph
Figure 1: Examples of experts and their expertises in KnowWhere-Grap

To enable disaster relief specialists to access and explore easily the graph of experts, expertises, and disasters, we also provide a similarity search interface and a follow-your-nose interface, which are demonstrated in Figure \ref{fig:something}. A similarity search can be useful when it is known that some person's expertise would be ideal for consulting on some Disaster Relief incident, but they are unavailable. In the similarity interface (Figure \ref{fig:something}), users can type in that expert's name into the search box and the system will return the top 15 experts who are most similar to the input expert. The similarity score is computed using a combination of Doc2Vec and knowledge graph embeddings \citep{le2014distributed, mai2018combining} based on the most cited three and most current three papers for each expert as well as the relations of experts to other nodes in the graph. Figure \ref{fig:something} (Right) shows an example of the similarity search. From there, users can directly explore information about the experts, their area of expertise, and events that they have worked on. Conversely, users can also start by selecting a certain event or a geographic region, learn about previous events, their impacts, and experts that could be contacted. In fact, this ability to seamlessly navigate between physical events, event topics (expertise), affected regions and people is one of the key strengths of knowledge graphs such as our KWG.

Left: Similarity interface of experts. Right: Follow-your-nose interface
Figure 2: Left: Similarity interface of experts. Right: Follow-your-nose interface