The challenge data is processed directly within the web browser, including removing duplicated/repeated posts, removing spams (sale promotions), extracting keywords based on an earthquake taxonomy, and aggregating posts by the hour. To determine the problems across the St. Himark, we first built the earthquake taxonomy for related events and resources needed for the situation:

Events:

• Earthquake: seismic, earthquake, quake, quaking, shake, shaking, wobble, wobbling, quiver, epicenter
• Ground damage: mudslide, rupture, landslides, liquefaction
• Flooding: tsunami, flood
• Aftershock: aftershock
• Fire: fire, smoke, burn

Resources:

• Food: food, hungry, eat
• Water: water, drink, thirst, dehydrate, reservoir
• Sewer: sewage, discharge, drain, irrigation, sewer, reservoir
• Power/Energy: blackout, electric, candle, energy, flashlight, fuel, gas, generator, nuclear, power, radiant, radiation, radio rays, valve
• Medical: ambulance, blood, bruise, emergency, escape, evacuate, fatal, first aid, fracture, hurt, illness, infection, injury, kill, lump, medic, red cross, rescue, respiratory, suffering, swollen, urgent, victim, wound
• Shelter: shelter, housing, building, collapse, construction, house
• Roads_and_bridges: bridge, traffic, congestion, avalanche, highway, lane, logistic, jammed, route, street, transportation

Our visualization has three main components:

1. Control panel includes

• Search box: To filter the main view by the inputted term
• Color legend: Such as Green for users, Blue for resources. Users can click on these names to show/hide that category in the main view
• Link filter: To adjust the threshold for link thickness.
• The St. Himark map: To allow users to select regions of interest.

2. Main view:
We use the dynamic network visualization, called TimeArc, to visualize the evolution of different topics and the correlations between them. As shown in the main view of Figure 1, TimeArc has 2 main components: the stream graphs of topics depict the popularity of these topics over time, and the arcs between topics present the cooccurrences in the text messages from social media posts. The thickness of links indicates how often the two topics mentioned together in Y*INT messages.

3. Detail view:
The detailed Y*INT messages of particular events/topics are presented to users in tabular format on demand. This allows users to investigate the content of social media posts by selecting a topic or a link in our time series view (TimeArcs).

Figure 2(A) shows the 3 earthquakes (at 2pm Mon 06, 7am Wed 08, and 2pm Thu 09). In particular, the purple stream depicts the aggregated frequency of earthquake-related terms (listed in the above taxonomy): seismic, quake, shake, shaking, wobble, quiver, epicenter. On the left, St. Himark map is colored by the number of earthquake-related posts.
As shown in Figure 2(B), the 2nd pulse is strongly connected to Downtown and West Parton while the 3rd pulse has many more connections to Weston, Southton, Cheddarford, Southwest, East Parton, and Boardview. In Figure 2(B), the map is color-coded by only posts associated with these connections (arcs).

Now, we are going to focus on the following resource allocations: roads and bridges, sewer and water, power and energy, and medical.

We can focus on roads and bridges by input this category into the search box. We inserted the pink annotated text on top of our time series for explanation purpose. As depicted in Figure 3(A), the concerns on roads and bridges are posted as soon as (2nd and 3rd) earthquakes occur. The posts at 1 hour after the 2nd earthquake are announcements of roads/bridges closures, as shown in Figure 3(b). At 5 hours after the 2nd earthquake, some further announcements are posted. In particular, Magritte Bridge and Wilson Forest Highway are re-opened at this hour, as depicted in Figure 3(c).
User DOT-StHimark is the main organization which is responsible for announcing/updating information on roads and bridges in St. Himark. At 4 hours after the 3rd earthquake, Figure 3(d) shows important Y*INT messages regarding the activities (close/open) of Tranky Doo Bridge, Wilson Forest Highway, and 12th July Bridge (connecting outside to Safe town).

For the entire time span, people constantly concern about Nuclear Plant and its efficiency as shown in the blue stream of Figure 4. The need for power significantly increase at 6 hours after 2nd earthquake and keep fluctuating 26 hours after. The possible regions in need of power include Cheddarford, Southwest, Scenic Vista, PalaceHill, and DownTown as shown by connecting arcs at various hours. The links between Power/energy and roads/bridges are due to the spams detailed in Figure 4(b) as we mouse-over the arcs.

We now investigate sewer problems across St. Himark, by filtering only the posts that contain: sewage, discharge, drain, irrigation. We annotate the critical sewer events directly in Figure 5(A) and list the detail messages in that hour at the bottom. Figure 5(b) displays the discussions on broken sewer pipes in Old Town and other regions (Safe Town, Scenic Vista, Broadview, Chapparal and Easton) at 5 hours after the earthquake as reported by HealthDept. This leads to contaminated water: we can easily notice a peak on the water stream at the same time.

At 10 hours after the 2nd earthquake, the sewer problem becomes worst and leads to flooding, mainly in Old Town, as shown in Figure 5(c). We recommend the sewer crew to be assigned in Safe town and nearby to mitigate earthquake effects.

At 24 hours after the 2nd earthquake shown Figure 5(d), the sewer problem has not been fixed. The temporal solution was proposed: drink boiled water and use portable toilets. When the 3rd earthquake hits, the sewer problem is not significant since the sewer system has not been repaired (can not be worse) and the city is better prepared for the 3rd hit (volunteers were deployed to provide food, water, and blankets to citizens before the 3rd earthquake).

We filter medical and shelter discussions by using the search box. As depicted in Figure 6(A), the demand for medical is very high immediately after the 2nd earthquake and during 5 to 15 hours after that. Some medical-related reports are presented in Figure 6(b). In term of location, the rescue teams are needed in the regions near the earthquake center, such as Old Town, Safe town, Scenic Vista, and DownTown.

Additionally, the medical needs are linked to the shelter demand from 6 hours to 15 hours after the 2nd earthquake with a slight delay of less than 1 hour. By mousing over a link, Figure 6(c) displays some medical vs. shelter demand at 15 hours after the 2nd earthquake.

At 24 hours after the 2nd earthquake shown Figure 6(d), most of the discussions are about evacuation as some of the main roads have been re-opened.

Figure 7 illustrates 5 time points that St. Himark conditions have been change, from left to right: (a) 9:00am Wed 08, (b) 1:00pm Web 08, (c) 6:00pm Web 08, (d) 7:00pm Thu 09. The focused resources are listed top-down. In particular,

• At (a), all routes have been closed as announced by DOT-StHimark , and no resource is needed at this point except some emergency medical issues.
• At (b), the water (the 3rd resource in Figure 7) is contaminated due to broken pipes in Old Town, Safe Town, Scenic Vista, and Downtown. Moreover, the people start to call for help from Scenic Vista. Two routes have been open at this time. After that, the shelter, medical, and power remain high in almost all regions of St. Himark.
• When flooding happens at (c) (sewer and flooding connection at 10 hours after the 2nd earthquake), the shelter demand reaches the peak. Consequently, the medical demand increases after 2 hours in Palace Hills, Northwest, Downtown, Weston, and Southton.
• Then, all demand (except power) decreases until other bridges are re-opened at (d) to reduce the traffic jam at Magritte Bridge.
• After (d) 23 hours after the 2nd earthquake, people need food, water (links between food and water); some reports concerning roads and bridges problems in Oak Willow and medical issues in Northwest, Southwest, Northwest, and Downtown.
• Since all route re-open at (e), no further resources are needed.

Besides the flood described in Figure 5(c), the St. Himark community experiences other events outside the realm of the earthquakes, including food, water, and fire.

We filter the visualization (using the search box) by food-related demand: food, hungry, eat. As shown in Figure 8(A), food demand isn't the major concern during the observed period, yet in some places (such as Oak Willow), the food demand is signified around 24 hours after the earthquake by traffic delay and the closed routes as showed in messages Figure 8(b).

At 21 hours after the 3rd earthquake, there are more discussions about food, as shown in Figure 8(c).

Figure 9 shows the water stream and its related topics. Due to the problem with the sewer system at 5 hours after 2nd earthquake, the drinking water needs to be boiled before use and citizens are very worried about that as shown in messages in Figure 9(b).

The fire-related posts are displayed in Figure 10. Overall, a limited number of posts reporting fires in St. Himark, including announcements from the fire department. About 5 hours (Figure 10 (b))to 9 hours after 2nd earthquake (Figure 10(c)), there are several messages report fire/ smoke/ burning issue in Old Town, Downtown, Scenic Vista, Southwest and Terrapin Springs. Since the power went off and everybody using candles to replace a light resource, the risk of fire has been posted at 17 hours after the 2nd earthquake as shown in Figure 10(d).

The text data in this visualization is inputted as an array of messages and processed directly within a web browser using javascript and d3.js. In the real scenarios where the incoming data is large and dynamic, it is better to deploy a web server which performs data pre-processing and then sends the extracted terms/topics to the client (web browsers) via web api. The performance of this method can be improved using server rather than using the client browser alone.

Whether using a server or not, our web application still has the ability to analyze both static and dynamic data. As the new streaming data coming, the new chunk of data is accumulated by the hour. An obvious limitation of our technique is that if the emerging topics or connections might not be highlighted until accumulated to a certain threshold, such as the link strength filter which sets the minimum concurrences of two terms/events.

In this case, the aggregated time interval can be refined to 10 minutes or a minute to capture the hot topics/events and their associations promptly.