Karamshuk, Dmytro (2013) Modeling and understanding the role of human mobility in the cyber-physical world. Advisor: Conti, Dr. Marco. Coadvisor: Lenzini, Prof. Luciano . pp. 135. [IMT PhD Thesis]
Text
Karamshuk_phdthesis.pdf - Published Version Restricted to IMT staff and National library only Download (3MB) |
Abstract
Modeling human mobility is important in the context of smart cities as it can assist design of pervasive systems and intelligent services in the city. In synthetic mobility models dynamic processes in the city are modeled by means of either simulation or mathematical analysis. Traditional synthetic approaches are usually limited by the state of the art findings in human mobility analysis and fail to update when new results come up from trace analysis. Moreover, the understanding of the connection between different mobility characteristics is missing from the existing synthetic models. This implies that there is no direct way to control the output of the models (e.g., statistics of contacts between people) using the input parameters (e.g., human mobility patterns). In this work we propose a mobility framework that can be instantiated to the required mobility settings and produce controllable output. The framework is built around the three dimensions of human movements, namely, social, spatial and temporal. The social environment in the framework is customized by taking the social graph as input. Then the spatial dimension is added by distributing communities of tightly connected users across common meeting places and assigning them to physical locations. The temporal dimension of human arrivals to places is modeled with stochastic point processes. We demonstrate the flexibility of the framework by showing that it can reproduce realistic mobility behavior observed in the mobility traces collected from online locationbased social networks. Additionally, we show that the framework can produce controllable output by providing a thorough mathematical analysis of the contact statistics in different mobility settings. Alternatively, data-driven models are used when the system under analysis is not well formalized but its behavior can be traced and further studied from the traces. In data-driven models relations between properties of the system and patterns of human movements are mined directly from the data with the help of machine-learning. In the second part of this work we develop a data-driven methodology to study the impact of human mobility on the retail quality of locations in the city. With respect to existing work in this direction we aim to assess the extent to which the new layers of information available in location-based social networks can assist geographic retail analysis. We study co-location patterns of various venues in the city and propose a methodology to assess the flows of the users between them. We exploit the result of this analysis to tackle the optimal business placement problem for three different retail chains in New York. We formalize this problem as a data-mining task where we aim to predict potential popularity of a store if placed in a given area. We devise a number of signals to describe the area including place-geographic features, e.g, density, heterogeneity of places, and mobility-based features, e.g., flows of users towards and inside the area. We show that the presence of place-attractors (e.g., airport, train station) and competing venues in the area are strong indicators of the popularity across all considered chains. However, the best performance is achieved when we consider the fusion of mobility and place-geographic features.
Item Type: | IMT PhD Thesis |
---|---|
Subjects: | Q Science > QA Mathematics > QA75 Electronic computers. Computer science |
PhD Course: | Computer Science and Engineering |
Identification Number: | https://doi.org/10.6092/imtlucca/e-theses/145 |
NBN Number: | [error in script] |
Date Deposited: | 20 Oct 2014 09:18 |
URI: | http://e-theses.imtlucca.it/id/eprint/145 |
Actions (login required, only for staff repository)
View Item |