Visualizing the "Heartbeat" of a City with Tweets

TL;DR

This study addresses how to quantify urban human activity dynamics using large-scale geolocated Twitter data. By analyzing over 6 million geolocated tweets from the NYC metro area collected between 2013-08-19 and 2013-12-31, the authors build 168 hourly slices on a 90,000-cell grid and compute per-hour deviations with $d^i_{ ext{hour}} = \tanh\left[ \alpha \left( n^i_{ ext{hour}} - \bar{n}^i \right) \right]$ (where $\alpha = 0.04$), producing heat maps and interactive movies of activity. They uncover a diurnal 'heartbeat' driven by wake/sleep cycles and commuting patterns, with Manhattan dominating work hours and bedroom communities contributing in mornings and evenings; differences between weekdays and weekends include a later weekend wake time and absence of a pre-commuting peak. The paper also identifies high-activity locations such as airports, Meadowlands, and the Statue of Liberty, and shows that anomalous individual activity can dominate local patterns, which motivates separating collective dynamics from outliers. A global metric based on the average distance to a downtown reference point is computed via the Haversine distance to summarize city-scale shifts, highlighting the potential of social media analytics for urban planning and attention modeling in near real time.

Abstract

Describing the dynamics of a city is a crucial step to both understanding the human activity in urban environments and to planning and designing cities accordingly. Here we describe the collective dynamics of New York City and surrounding areas as seen through the lens of Twitter usage. In particular, we observe and quantify the patterns that emerge naturally from the hourly activities in different areas of New York City, and discuss how they can be used to understand the urban areas. Using a dataset that includes more than 6 million geolocated Twitter messages we construct a movie of the geographic density of tweets. We observe the diurnal "heartbeat" of the NYC area. The largest scale dynamics are the waking and sleeping cycle and commuting from residential communities to office areas in Manhattan. Hourly dynamics reflect the interplay of commuting, work and leisure, including whether people are preoccupied with other activities or actively using Twitter. Differences between weekday and weekend dynamics point to changes in when people wake and sleep, and engage in social activities. We show that by measuring the average distances to the heart of the city one can quantify the weekly differences and the shift in behavior during weekends. We also identify locations and times of high Twitter activity that occur because of specific activities. These include early morning high levels of traffic as people arrive and wait at air transportation hubs, and on Sunday at the Meadowlands Sports Complex and Statue of Liberty. We analyze the role of particular individuals where they have large impacts on overall Twitter activity. Our analysis points to the opportunity to develop insight into both geographic social dynamics and attention through social media analysis.

Figures (5)

• Figure 1: NYC geographical region with the locations of 6 million tweets shown. The sharp land-sea boundary is apparent as is the boundary of land area with high population density.
• Figure 2: Twitter activity of weekdays compared to weekends, where the difference in the urban life can be clearly seen during the late night (top panels) and late afternoon (bottom panels). Colors as in Fig. \ref{['fig:day']}
• Figure 3: Examples of times of high activity at locations listed in Table \ref{['tab:spots']}. The left panel shows annotated activity plots using the colors as in Fig. \ref{['fig:day']}, and the right panel shows the corresponding height as a three dimensional surface.
• Figure 4: Single user anomaly: the annotated peak corresponds to a single user tweeting more than 10 times the average of other users.
• Figure 5: Average distance from Central Park at the different hours of the day for each day of the week. The working days of the week, Monday to Friday, are shown in different shades of blue (light blue for Monday and the darkest blue for Friday), while the weekends are shown in shades of red (red for Saturday and dark red for Sunday).