Athens University Economics and Business (AUEB)
Real-time, reliable and privacy-aware mobile crowdsourcing: algorithms and practical systems
Abstract
dc:descriptionThe recent massive prevalence of mobile devices along with the advances in mobile networking have led to the development of "Mobile Crowdsourcing Systems". In these systems the ubiquitous human users participate actively in data collection and sharing in order to identify events of interest in real-time. The data contributed by the participants in mobile crowdsourcing can be either extracted from the sensors which are embedded on the users' mobile devices or obtained by asking questions to the ubiquitous human users. Such information can be can be used to identify events of interest in real-time in a wide range of application domains from entertainment, to transportation, to environmental monitoring. Mobile Crowdsourcing is a novel application paradigm that reflects unique opportunities mainly due to: (i) the ubiquity of the users, (ii) the low-cost and easy to deploy infrastructure, (iii) the dynamic nature of the queries and (iv) the various connectivity media such as 3G, 4G, and WiFi. However, there exist several important challenges that need to be addressed in order to use Mobile Crowdsourcing systems in their full potential. These challenges include, but are not limited to, extracting data from appropriate users to maximize the information gain, privacy protection of the participants, dealing with real-time challenges and response delays, quality assessment of the collected data, data unavailability and effectively incentivizing people to participate. The purpose of this thesis is to propose practical methods to address some of these issues so as to improve Mobile Crowdsourcing Systems. The first part of the thesis improves Mobile Crowdsourcing Systems by tackling the problem of assigning crowdsourcing tasks to the most appropriate users in order to maximize the benefit for the task requester. The problem is difficult since human users have different skills and behaviors that depend on the human factor, such as objectiveness, reliability and response delays which are often difficult to estimate in advance. Hence, selecting the most appropriate users for each task in real-time is a challenging problem. This thesis first contributes to this problem by providing an approach that assigns each incoming task to the most appropriate group of human workers taking into account their reliability and real-time performance. The proposed system systematically investigates the search space for Pareto optimal solutions with respect to the defined objectives. The algorithm executes in polynomial time and improves the amount of tasks processed under the defined constraints over 71% compared to the state-of-the-art approaches like Amazon Mechanical Turk. Then, we focus on the problem of exploiting Mobile Crowdsourcing to detect ongoing events effectively and in real-time. The problem is that querying every available user is infeasible due to the volume and costs of crowdsourcing. Thus, we should be able to identify and query the optimal sample out of the available users. In this thesis we contribute to mobile crowdsourcing with a set of event detection approaches. Our first approach uses a novel sampling algorithm that selects a subset of the ubiquitous human users to provide real-time information so as to identify the state of a major emergency event. Our proposed approach captures well emergencies in large geographical regions and takes into consideration the individual characteristics of the human users to maximize the probability of receiving accurate responses in a timely manner. Our second approach exploits the sensor-based readings of the mobile devices of the users to extract important events with minimal cost. The proposed approach determines clusters of users dynamically to identify the locations where important events take place and then performs sampling to retrieve the most representative data streams in every cluster to extract additional information for the event. In the second part of the thesis we contribute on the problem of preserving user privacy. The problem relies on the fact that mobile crowdsourcing systems require from the users to share location-embedded information that can lead to exposing private user information. Hence, sophisticated attacks can expose sensitive information such as the user identity and the places that the user visits frequently. First, we contribute with a distributed data exchange approach that aims to protect the user sensitive data by making the attacker consider all user trajectories as equiprobable to contain sensitive data, so that the leak of sensitive data is prevented. Second, we examine to the privacy effect that arises when the anonymous user mobility traces can be linked with publicly available data and reveal personal user information that was not available in advance. For this problem we present an approach that uses a novel feature-based Entity Resolution technique to determine similar user identities and react to preserve user privacy on the shared data. Finally, another privacy problem arises when users share publicly geo-located information through crowdsourcing applications. The issue is that the majority of the data will reside along routes mostly visited by the individuals. Hence, as the amount of user-contributed content increases, they gradually expose the user's mobility. We address this problem with a highly-efficient privacy preserving approach for Crowdsourcing applications that runs locally on mobile devices. The approach relies on the geometric approximation solution of coresets to preserve user privacy, by evaluating the privacy exposure of the users before sharing their geo-located data. Throughout the thesis, our methodologies have been evaluated on real-world datasets and scenarios. Our experimental results have shown that our proposed algorithms systematically outperform the competing approaches and that they constitute practical approaches which can be employed in commercial Mobile Crowdsourcing Systems.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2017
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Boutsis, Ioannis
- Μπούτσης, Ιωάννης
Subjects
dc:subject × 15Rights
dc:rights- Statement dc:rights
-
- BY_NC
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/40268