On Leveraging Dynamic Processes in Large Social Networks for Smart Cities

Abstract

The concept of smart city which began as being synonymous with electronically networked community underwent significant changes with the growth in mobile devices and social networking. This expanded the outlook of smart city to encompass different perspectives namely Smart People, Smart Governance, Smart Mobility, and Smart Economy. While the diverse sensing capabilities of the mobile devices and social networking have influenced the smart city in several ways and continue to do so, this thesis attempts to make a contribution on leveraging dynamic processes in large social networks represented by the Social Media and Telecom Call Data Records (CDRs) in order to understand human behavioral patterns and information cascade dynamics in the context of smart cities. A thorough study of the advances in communication, the behavioral pattern of usage of technologies and the demands of smart cities in today’s context has revealed that the Social Media data and the CDRs can be used effectively in understanding and identifying the key stakeholders and influencers in the society, the mathematical limitations on the predictability of the influences by the influencer, the retention capability and the behavior to churn of the influencers and the predictability of their everyday mobility patterns. These form the major focus and contribution of this thesis. The content of this thesis includes four chapters covering Smart cities from four different perspectives namely People, Governance, Mobility, and Economy using social networking and mobile technologies. The thesis provides technical solutions based in machine learning, graph and data analytics to relevant problems in these four directions. With regards to Smart People, an efficient algorithm based on the state-of-the-art semi-definite programming solution technique has been proposed for solving the problem of influence maximization when the network structure over which the information propagates is unknown for identifying community leaders and key players that encourage development in digital communities for maximizing the dissemination of information in a smart city. With the vast majority of people adopting online social media for communication, it becomes highly lucrative for viral marketers, political campaigners, local governing authorities, security agencies etc. to understand and be able to foresee the cascade xii dynamics, in order to affect social media movements for desired results. A mathematical framework for characterizing the predictability of social media cascade volumes based on entropy has been proposed and the theoretical “maximal predictability” achievable has been derived for augmenting emergency response and city incident management services towards Smart Governance. Understanding the complexities underlying the emerging behaviors of human mobility patterns in a metropolitan city is essential toward making informed decisionmaking pertaining to urban infrastructures. This thesis addresses the problem of predicting the next location of users for large population in a metropolitan city based on their past trajectory. A novel Temporally-Aware Matrix Factorization approach used in Recommender systems that leverages the scalability of matrix factorization methods and can effectively handle the temporal dependencies and the sparsity present in the users’ mobility patterns found in CDRs has been proposed. In the area of Smart Economy, business models for keeping customers satisfied is one of the primary capital invested areas for the economic growth of these service providers. To this end, this thesis develops churn prediction models for which there exists very little data except for CDRs. Such practical limitations certainly make the problem more challenging. Novel hybrid feature sets based on mobile call usage and social network groups have been proposed in order to improve customer churn prediction models for businesses in a Smart Economy. In all the problems addressed in this thesis, experiments have been carried out on various synthetic and real-world networks and the proposed approaches have been compared with relevant methodologies currently existing in the literature. The results demonstrate the superior performance of our proposed approaches for problems addressed. In conclusion, by leveraging recent advances in real time data analytics, machine learning and visualizations, mapping and location enablement in mobile phones and the rise of social media, the methodologies proposed in this thesis provide new avenues for development of smart city services that can be integrated in current smart city frameworks and thus provides impetus to the development of a Smart City.