Modeling the Impact of Human Mobility: Mobile Devices as Sensors and Content Vectors

Abstract

Cellular technology has had a profound impact on modern computing. As an expanding computing paradigm, mobile computing presents many challenges and opportunities. Cell phones give researchers an unprecedented glimpse into the way that humans move, but records of a phone's movement are precise neither in time, nor space. The ease of deploying the devices offers connectivity solutions in areas that previously had little hope of receiving Internet access, but the mobility of the devices means

users may be difficult to locate.

The ubiquity of mobile devices gives researchers a chance to understand how and when people move. By analyzing regional mobility patterns, researchers can understand the impacts of human mobility on communications networks. This information can also be used in a wide range of other fields: social science, urban planning, and biology, among others. Understanding movement can help model human mobility, which may provide better ways to test new communications protocols or experiment with planned changes to an urban area. However, challenges come from the variance in how many calls different users make, and in cell tower density between urban and rural areas. Additionally, the datasets containing records of people's movements are often proprietary and thus unavailable.

Using an unprecedentedly large dataset of over 3 billion cellphone data records for hundreds of thousands of users in Los Angeles and New York, this dissertation demonstrates techniques to quantitatively characterize and model movement patterns within the cities. By selecting appropriate metrics and validating against a set of volunteers, I show that differences between regions can be accurately observed (e.g., Angelenos, in general, take day-to-day trips that are 34%-53% longer than those of New Yorkers). Further, this thesis presents algorithms based on logistic regressions that allow the discovery of home and work locations of users. Such estimates can be used to calculate average commute distances for a region that are within 1 mile of census estimates. Further, this thesis shows how to combine the macroscopic properties of cities with user profiles (generated via machine learning techniques) to create a fully synthetic city with user behavior that mimics the behavior of actual users. These synthetic cities can be used to model user behaviors in simulations of mobile systems.

The second half of my dissertation explores using mobile devices such as laptops and cell phones to connect developing regions to the Internet. Access to the Internet is an increasingly powerful way to make available tools for healthcare, education, and more. However, the world has a "digital divide"; parts of the world have 80% of their population connected to the Internet while other parts have less than 15% Internet penetration. Fortunately, although adoption of traditional computing platforms is slow, cell phone adoption in these areas is very high. Thus, using these and other mobile devices to deliver content is a promising avenue of research into narrowing the divide.

Despite the promise of mobile devices, many challenges exist. The

mobility of the devices means that it is hard to know where a user will be when she wants to access Internet content. Further, the devices may be connected to the Internet through a poor quality Internet connection. Satellite or cellular Internet connections are costly and multi-hop, delay-tolerant connections may have hours of latency.

To address the challenges to connectivity in developing regions, my

dissertation proposes and evaluates collaborative caching techniques. In these techniques, a group of peer-nodes (e.g., laptops or mobile phones) cooperate to store and retrieve data, minimizing how often users need to use the expensive or slow Internet connection. However, because data are shared by all members of the cache, user mobility plays a large role in the efficacy of such a cache. I show that even with user mobility, collaborative caches reduce latencies by over 35%. As a demonstration of the efficacy of the techniques, this thesis showcases a real-world

deployment of a system called C-LINK which uses collaborative caching. C-LINK was deployed in rural Nicaragua on top of a multi-hour, delay-tolerant network.

Overall, this thesis opens up an interdisciplinary body of work

exploring the mobile phone as the next generation of computing. Specifically, how mobile devices can help answer questions on topics as far-ranging as carbon footprints and Internet connectivity.