Mobile Crowdsourcing

Inhaltsverzeichnis

1. Frontmatter

2. Introduction

   1. Frontmatter

   2. Crowdsourcing as a Future Collaborative Computing Paradigm

      Jie Wu, Chao Song, Wei Chang

      Abstract

      Crowdsourcing coordinates a group of people called crowd, to perform small jobs that solve various problems (which computer systems or a single user could not easily solve). Crowdsourcing involves three components: tasks that need to be completed, workers that are recruited to complete tasks, and a platform that assigns workers to tasks and facilitates the completion of tasks. Since crowdsourcing is becoming more prevalent, in this chapter we try to understand and characterize crowdsourcing better through reviewing historical events, followed by some existing platforms that support crowdsourcing, as well as various crowdsourcing applications, and we end with a discussion of several algorithmically and theoretically challenging issues. We take a unique viewpoint of treating people in crowdsourcing as Human Processing Units (HPUs) in contrast to Central Processing Units (CPUs) in traditional computer systems. The comparison of HPU and CPU is analogous to the human vs. machine debate nowadays. We will provide our own insights on this debate.

   3. Urban Mobility-Driven Crowdsensing: Recent Advances in Machine Learning Designs and Ubiquitous Applications

      Suining He, Kang G. Shin

      Abstract

      Driven by explosively growing urban big data, increasing network connectivity, and enhanced mobility of human and transportation modalities, Urban Mobility-driven CrowdSensing (UMCS) has been widely used for various urban and ubiquitous applications, including urban event monitoring, city planning, and smart transportation system. This chapter examines and analyzes the recent advances and application of UMCS learning algorithm design and emerging use cases. In particular, we first overview the recent advances of machine learning (ML) techniques and algorithms for crowdsensing signal reconstruction and mobility learning to understand the importance of signal learning and mobility characterization for UMCS. We then review the emerging applications for UMCS, including indoor crowd detection and urban mobility system reconfiguration. For each category, we identify the strengths and weaknesses of the related studies and summarize the key research insights and representative studies, which can serve as a guideline for new researchers and practitioners in this emerging and important research field.

3. Key Technical Components: User Recruitment and Incentive Mechanisms

   1. Frontmatter

   2. Unknown Worker Recruitment in Mobile Crowdsourcing

      Mingjun Xiao, Yin Xu, He Sun

      Abstract

      Mobile CrowdSourcing (MCS) has emerged as a cheap yet effective paradigm by which a platform can recruit a group of mobile workers to jointly accomplish some complex tasks by using their smart mobile devices. In this chapter, we focus on the unknown worker recruitment problem in MCS, where workers’ qualities are unknown in advance. Meanwhile, we also consider the budget constraint and privacy-preserving issue, i.e., protecting each worker’s quality information from being revealed to others. In order to tackle the unknown worker recruitment challenges, we model the problem as a multi-armed bandit game by regarding each worker as an arm and the task completion quality contributed by each worker as the reward of pulling the arm. Then, we propose an unknown worker recruitment scheme, where we extend the Upper Confident Bound (UCB) bandit strategy to balance the exploration and exploitation processes. Furthermore, we also extend this scheme to support the budget constraint and adopt the differential-privacy technique to provide the privacy-preserving functionality, respectively. Finally, we analyze the regret bounds of the proposed schemes through rigorous theoretical proofs.

   3. Quality-Aware Incentive Mechanism for Mobile Crowdsourcing

      Haiming Jin, Lu Su

      Abstract

      Recent years have witnessed the emergence of mobile crowdsourcing (MCS) systems, which leverage the public crowd equipped with various mobile devices for large-scale sensing tasks. In this chapter, we study a critical problem in MCS systems, namely, incentivizing user participation. Different from the existing work, we design two quality-aware incentive mechanisms, and we incorporate a crucial metric, called users’ quality of information (QoI), in the first quality-aware incentive mechanism and consider the preservation of users’ bid privacy in the second quality-aware incentive mechanism for MCS system. Due to various factors (e.g., sensor quality, noise, etc.), the quality of the sensory data contributed by individual users varies significantly. Obtaining high-quality data with little expense is always the goal of a quality-aware incentive mechanism for MCS system. Besides, the data from users usually contains the private information that should not be disclosed. A quality-aware incentive mechanism should consider the preservation of users’ bid privacy. Technically, we design the first quality-aware incentive mechanism based on reverse combinatorial auctions. We investigate both the single-minded and multi-minded combinatorial auction models and design two computationally efficient mechanisms that the one for single-minded models can approximately maximize social welfare and the one for multi-minded models can achieve close-to-optimal social welfare. We design the second quality-aware incentive mechanism based on the single-minded reverse combinatorial auction that preserves the privacy of each workers bid against the other honest-but-curious users. Specifically, we design a private, individual rational, and efficient mechanism that approximately minimizes the platforms’ total payment and satisfies the desirable economic properties of approximate truthfulness and individual rationality.

   4. Incentive Mechanism Design for Mobile Crowdsourcing Without Verification

      Chao Huang, Haoran Yu, Jianwei Huang, Randall Berry

      Abstract

      This chapter studies the design of incentive mechanisms for mobile crowdsourcing systems in which verifying the underlying ground truth is not possible. Namely, we consider a crowdsourcing platform that seeks to incentivize a group of workers to put in effort and truthfully report solutions to a given task. Challenges in this setting include that the workers may have heterogeneous capabilities and may have an incentive to collude in order to deceive the platform. The platform itself may have incomplete information regarding the workers’ capabilities, which it could attempt to learn over time. Furthermore, there may be asymmetries in the information available to the platform and to the workers. We will survey approaches to dealing with such problems using game-theoretical and online learning-based approaches.

4. Key Technical Components: Task Allocation

   1. Frontmatter

   2. Stable Worker–Task Assignment in Mobile Crowdsensing Applications

      Fatih Yucel, Murat Yuksel, Eyuphan Bulut

      Abstract

      Mobile crowdsensing (MCS) is an emerging form of crowdsourcing, which facilitates the sensing data collection with the help of mobile participants (workers). A central problem in MCS is the assignment of sensing tasks to workers. Existing works in the field mostly seek a system-level optimization of task assignments (e.g., maximize the number of completed tasks or minimize the total distance traveled by workers) without considering individual preferences of task requesters and workers. However, users may be reluctant to participate in MCS campaigns that disregard their preferences as this can cause the majority of users to find their assignments dissatisfying and consequently to cease participating in the campaign, putting the long-term success of the campaign in jeopardy. Moreover, dissatisfying task assignments may hinder the effective functioning of the campaign, as unhappy users may refuse to fulfill the assignments made by the platform. Thus, it is critically important to consider user preferences during the task assignment process of MCS campaigns.

      In this book chapter, we review the recent studies in the mobile crowdsensing domain that seek preference-aware assignments between the workers and tasks in the system. While these studies build their designs on the Stable Matching Theory, existing methods to find stable preference-aware assignments cannot be applied directly due to the type of the MCS scenario (e.g., participatory, opportunistic, or hybrid) as well as the constraints (e.g., budget of task requesters, Quality of Service (QoS) required from completed tasks, capacity of workers (i.e., the maximum number of tasks they can perform), and rewards given to each task) given in MCS setting and the assignment types allowed between workers and tasks such as many to one and many to many with additive or non-additive utility of workers. We highlight these differences in each MCS scenario studied together with how the stability is defined. We then discuss if stable solutions are possible in each and provide a brief summary of corresponding solution approaches. Finally, we provide a set of open problems that need to be studied to find stable work–task assignments.

   3. Spatiotemporal Task Allocation in Mobile Crowdsensing

      Honglong Chen, Guoqi Ma, Yang Huang

      Abstract

      With the rich sensing ability and extensive usage of various sensors, mobile crowdsensing (MCS) has become a new paradigm to collect sensing data for various sensing applications. In the MCS system, task allocation is one of the most significant issues, which refers to that the platform assigns the most suitable users to each task according to the attribute characteristics of users and tasks. Reasonable and effective task allocation is essential to improve the efficiency of the MCS system. Generally, the time and space requirements of sensing tasks have a great influence on task allocation. Therefore, in this chapter, we take the spatiotemporal tasks into consideration, where mobile users are required to sense in specific areas, and each task’s collective sensing time should be no less than the specified time duration. We consider two scenarios: one is that the task requires sensing data with a specific duration and mobile users with limited time budgets, and the other is that the spatiotemporal characteristics and sensor requirements of the task are different and the mobile users are heterogeneous (e.g., personal preferences, carrying sensors). Finally, we give the corresponding task allocation algorithms and verify the effectiveness of the proposed algorithms through both theoretical analysis and extensive simulations.

5. Key Technical Components: Data Inference

   1. Frontmatter

   2. Joint Data Collection and Truth Inference in Spatial Crowdsourcing

      Xiong Wang

      Abstract

      In recent years, people have witnessed the proliferation of mobile devices equipped with powerful processors and plentiful sensors (GPS, microphone, camera, etc.). Spatial crowdsourcing is recognized as a newly emerged paradigm to leverage the power of crowds who carry mobile devices for large-scale tasks. Crowdsourcing enables a wide spectrum of intelligent applications, where much attention is drawn to the fundamental problems of data collection and truth inference. Researches have devised manifold techniques to discover truth from collected noisy data, but they frequently ignore various expertise of workers and dynamic information of crowdsourcing system, thus leading to error-prone estimated truth and unqualified collected data. In this chapter, we develop a framework for joint data collection and truth inference to accurately estimate truth and efficiently assign tasks. Specifically, we unify diverse types of numerical and categorical tasks based on probabilistic graphical model and then propose unsupervised learning methods that can dynamically infer ground truth and various worker expertise at the same time. Furthermore, we devise online task allocation schemes to gradually gather high-quality data considering location awareness and estimated worker expertise. Extensive evaluations show the superiority of our algorithms over the baseline approaches.

   3. Cost-Quality Aware Compressive Mobile Crowdsensing

      Yong Zhao, Zhengqiu Zhu, Bin Chen

      Abstract

      With ubiquitous mobile devices, mobile crowdsensing (MCS) emerges as a promising paradigm for monitoring the overall status of a large-scale area. However, the MCS applications have yet to be widely adopted in practice because of high sensing and communication costs as well as insufficient participants. To deal with these problems, compressive sensing is introduced into mobile crowdsensing, where it is used to deduce the missing data of unsensed locations by exploiting the inherent correlations of sensory data. This new paradigm is the so-called sparse mobile crowdsensing or compressive crowdsensing (CCS). In this emerging approach, compressive sensing not only can be used after the data are collected, but also before or during the data collection process. Two main questions lying in CCS are: (1) Where to sense? (2) How to recover the unsensed data accurately? Considering the practical factors (e.g., diverse sensing cost and importance disparity of different cells), it would affect the selection strategies design and further determine the recovery accuracy. In this chapter, we will have a close look at recent advances about CCS and provide formulations to solve the above-mentioned problems.

6. Key Technical Components: Security and Privacy

   1. Frontmatter

   2. Information Integrity in Participatory Crowd-Sensing via Robust Trust Models

      Shameek Bhattacharjee, Sajal K. Das

      Abstract

      In this chapter, we propose a recipe for event truthfulness scoring and user reputation scoring framework that is immune to the cold start problem in participatory mobile crowd-sensing applications, while being robust to attacks that target operational and AI-based weaknesses in mobile crowd-sensing and existing trust models. Our method does not need to depend on the knowledge of ground truth nor the existence of a prior user reputation, both of which are impractical assumptions during the cold start phase. Specifically, we first show subtle variations of dishonest intent in terms of fake event reporting attacks as an operational vulnerability. Additionally, we show threats that weaponize design provisions that help assess event truthfulness, in the form of feedback weaponizing attacks as an AI-based vulnerability. Furthermore, we show how existing methods of trust and reputation should be modified to jointly mitigate the effects of both fake event reporting and feedback weaponizing attacks, during the cold start, by using a vehicular crowd-sensing application as a proof-of-concept. Our design modifications are inspired by cognitive psychology, behavioral economics, and symbolic AI, and how they can be seamlessly embedded into known approaches for trust and reputation scoring.

   3. AI-Driven Attack Modeling and Defense Strategies in Mobile Crowdsensing: A Special Case Study on Fake Tasks

      Didem Cicek, Murat Simsek, Burak Kantarci

      Abstract

      Technological advancements in information and communication have brought up ubiquitous sensing as a key concept to obtain valuable crowd data. Mobile crowdsensing (MCS) has emerged as an essential data collection opportunity through a wide variety of smart device sensors. This technology has increased its popularity due to low data collection and sensing cost through open platforms where smart device users can reach out to the offers coming from MCS platforms. MCS campaigns aim to ensure that all resources are used for the required amount of time with the desired capacity; hence, trustworthy services can be offered to all stakeholders in a MCS platform. However, it is a big challenge to protect smart device users and MCS platform against cyber-attacks. Therefore, it is vital to implement intelligent attack and defense mechanisms before launching a MCS campaign. Anticipating fake task injections is one of the crucial strategies to be considered through artificial intelligent empowered attack modeling to forecast the impact of the attack on draining computational resources. Self-organizing feature map (SOFM) has been leveraged on fake task injection modeling to increase the impact of the attack strategy through launching fake sensing tasks in locations with the highest impact. Moreover, defense mechanism for legitimacy detection is a paramount factor to mitigate fake task submission effects on the users and platforms. Machine learning strategy should monitor the task generation process to eliminate malicious activities in the campaign; thus trustworthy environment for data collection can be achieved for task submission and sensing via smart devices. This chapter presents the state of the art of MCS security from the anticipatory and defensive perspectives through AI-enabled schemes. Furthermore, it unveils the open issues and challenges that remain roadblocks against wide adoption of MCS-enabled systems.

   4. Traceable and Secure Data Sharing in Mobile Crowdsensing

      Jinwen Liang, Song Guo

      Abstract

      In mobile crowdsensing, a crowd of mobile devices collect data and utilize cloud infrastructures to share data with others. However, as the cloud server is not always honest, sharing the collected data without protection may arise privacy concerns such as privacy leakage, unauthorized access, and secret key abuse. As a result, it is desirable to design data sharing schemes with traceability and security. In this chapter, we first introduce the background of security issues in mobile crowdsensing. Then, we review and compare several existing secure data sharing schemes in mobile crowdsensing. After that, a Traceable and privacy-preserving non-Interactive Data Sharing (TIDS) scheme is described as an example. Finally, we show the future research topics of traceable and secure data sharing schemes in mobile crowdsensing.

   5. User Privacy Protection in MCS: Threats, Solutions, and Open Issues

      Zhibo Wang, Xiaoyi Pang, Peng Sun, Jiahui Hu

      Abstract

      Mobile crowdsourcing, as an effective paradigm that completes data collection and processing with human involvement and powerful mobile devices, has become a promising area of interest for research and application in recent years. However, users involved in crowdsourcing systems are at high risk of privacy leakage since their sensing data is generally tagged with spatio-temporal information as well as personal information. This discourages users from playing an active part in crowdsourcing tasks. The privacy leakage issues arouse widespread concerns among researchers, and many privacy protection approaches have been proposed. To better understand these privacy challenges and solutions, this chapter first highlights user privacy concerns and requirements that exist in mobile crowdsourcing systems. Then, it summarizes some effective methods to address these concerns and requirements. Last but not least, it provides a thorough discussion of open issues for future research.

7. Applications

   1. Frontmatter

   2. Crowdsourcing Through TinyML as a Way to Engage End-Users in IoT Solutions

      Pietro Manzoni, Marco Zennaro, Fredrik Ahlgren, Tobias Olsson, Catia Prandi

      Abstract

      Tiny machine learning (TinyML) is a new field aimed at miniaturizing machine learning algorithms to the point that app developers can integrate them into IoT devices. Since TinyML delivers AI capabilities to embedded devices, it is also known as edge AI or embedded AI.

      TinyML allows bringing AI to devices like smartphones, tablets et al., too. Since these mobile devices have currently surpassed desktop computers as the primary computing device for most users, it allows the possibility to engage even more end-users in crowdsourcing data for the IoT world.

      In this chapter, we will review the current status of TinyML by illustrating its underlying technologies and methodologies and showing some relevant examples where this new area is being used to provide novel applications, thanks to crowdsourcing as a way to engage the final user.

   3. Health Crowd Sensing and Computing: From Crowdsourced Digital Health Footprints to Population Health Intelligence

      Jiangtao Wang, Long Chen, Xu Wang

      Abstract

      Population health monitoring and modelling is important and fundamental for public health operations for the control and intervention of Non-Communicable Diseases (NCD). Healthcare administrators often perform data collection for population health monitoring either by integrating records of hospital visits or conducting survey among a sample of residents, but both approaches are of high cost and time-consuming, which results in limited spatial coverage. The proliferation of devices embedded with multimodality sensors and digital health applications in our daily lives generates data at an unprecedented scale, providing valuable crowdsourced information about personal health status or health-related context. In this book chapter, we propose a new vision, called Health Crowd Sensing and Computing (HCSC), which leverages opportunistic and crowdsourced digital health footprints within a full lifecycle of data collection, linkage, integration, augmentation, and analytics, to realise the goal of more intelligent population health monitoring for NCD. Specifically, our own case study called Compressive Population Health will be introduced, where we combine AI techniques with HCSC to achieve cost-effective public health monitoring. Finally, existing gaps will be discussed with future research opportunities and proposal in this interesting and novel research area.

   4. Crowdsourcing Applications and Techniques in Computer Vision

      Miloš Stojmenović

      Abstract

      Rapid advancements in artificial intelligence have recently revolutionized computers’ capabilities to understand the world just by observing it through a camera lens. Technologies such as automated driving emphatically show that computers can learn to do just about anything with a large enough labelled dataset. Deep learning applied to large, curated datasets has produced remarkable results, but its potential is yet to be unlocked in areas that do not enjoy an abundance of catalogued data. Gathering such datasets can be a very difficult task in situations where data is expensive to generate or labelling such data requires expert knowledge, or both. Examples of these circumstances can be found in fields such as object detection in geospatial imagery, or 3D instance segmentation of medical imagery. This is most notably applicable to various forms of nanometer scale microscopy imagery that is both difficult and slow to produce, and requires very specialized domain knowledge for labelling purposes. Crowdsourcing from previously disjointed expert communities is the approach taken to solve the data cataloguing problem in the mentioned fields. Experts are first evaluated on their ability to correctly recognize objects of interest on a pre-labelled dataset. This test of their abilities yields a confidence score which is used on future data they categorize. Aggregating labelled data from various experts, of differing knowledge levels, is also a notable challenge, as the confidence score plays a more or less important unification role depending on the complexity of the assigned segmentation task. Task difficulty can range from simple object detection, to instance segmentation, to finally panoptic segmentation where each pixel of the image needs to be categorized as belonging to an individual instance of a class of object, or to uncountable formless regions of identical texture such as the sky, sand, or cell plasma. We explore the many applications of crowdsourcing that attempt to solve computer vision tasks, and dive into the details pertaining to each approach. A top down perspective of crowdsourcing techniques is explored, and tied to each downstream application.

   5. Mobile Crowdsourcing Task Offloading on Social Collaboration Networks: An Empirical Study

      Liang Wang, Yong Cheng, Dingqi Yang, Haixing Xu, Xueqing Wang, Bin Guo, Zhiwen Yu

      Abstract

      Mobile Crowdsourcing (MCS), a human-centric promising paradigm for performing location-based tasks, has drawn rising attention from both academia and industry. In MCS applications, the outsourced tasks are allocated by a management platform to a group of recruited workers. However, during real-world task implementation, various types of unpredictable disruptions are usually inevitable, which might result in task execution failure, and subsequently hinder the development of MCS applications. Facing the task execution failure issue, centralized task reassignment approaches thus become ineffective and inefficient in practice. Against this background, by exploring the underlying social relationship between workers, we consider a distributed MCS task offloading scheme, i.e., the workers autonomously offload the unexecuted MCS tasks to their social acquaintances. However, to efficiently design such offloading mechanisms, we are facing several challenges, including investigating the relevant influential factors in task offloading, designing offloading patterns and incentive mechanism to accommodate it. To address these challenges, in this paper, we conduct an on-campus empirical study on MCS task offloading on social collaboration networks. Firstly, we conduct a survey covering over 1000 workers to capture the preliminary understanding of it. Based on the survey results, we then conduct a “field experiment” over a deployment period of 8 weeks, to comprehensively examine the intrinsic characteristics and behavioral patterns in task offloading, including the effectiveness of task offloading scheme, the offloadee selection, the impact of punitive measures, the adopted task offloading patterns, and reward-sharing incentive mechanism. By analyzing the collected operation logs of the workers, we summarize several important findings on the design of task offloading scheme in MCS applications, which we believe, can serve as useful guidelines for future research work on MCS task offloading.