PaperlayerView exact revision on arXiv

Source-linked AI summary

Quality of Information in Mobile Crowdsensing: Survey and Research Challenges

Francesco Restuccia, Nirnay Ghosh, Shameek Bhattacharjee, Sajal Das, Tommaso Melodia

arXiv:1709.03583v1cs.CY

TL;DR

Mobile crowdsensing depends on human participants and therefore raises unresolved questions about obtaining reliable, useful information at scale. The paper proposes a comprehensive QoI framework, surveys existing approaches to truth discovery and trustworthiness, and identifies research challenges and future directions. Its conclusion is a roadmap for improving QoI research in mobile crowdsensing, while existing work remains limited by scalability, behavioral, claim-type, and multimodal-data challenges.

  • Problem

    Mobile crowdsensing needs reliable and useful information from human participants, but existing literature lacks a comprehensive framework covering QoI and its research challenges.

  • Method

    The paper proposes a comprehensive QoI framework and surveys existing truth-discovery and trustworthiness approaches, their limitations, and related research challenges.

  • Results

    The paper provides a framework for defining QoI, analyzes challenges in enforcing and estimating it, and surveys the relevant state of the art.

  • Takeaways & Limitations

    The paper offers a roadmap of possible future research directions for improving QoI in mobile crowdsensing.

  • Takeaways & Limitations

    Existing work often lacks scalability analysis, dynamic human-reliability models, nonbinary claims, and methods for multimodal crowdsensing data.

Abstract

from arXiv · show

Smartphones have become the most pervasive devices in people's lives, and are clearly transforming the way we live and perceive technology. Today's smartphones benefit from almost ubiquitous Internet connectivity and come equipped with a plethora of inexpensive yet powerful embedded sensors, such as accelerometer, gyroscope, microphone, and camera. This unique combination has enabled revolutionary applications based on the mobile crowdsensing paradigm, such as real-time road traffic monitoring, air and noise pollution, crime control, and wildlife monitoring, just to name a few. Differently from prior sensing paradigms, humans are now the primary actors of the sensing process, since they become fundamental in retrieving reliable and up-to-date information about the event being monitored. As humans may behave unreliably or maliciously, assessing and guaranteeing Quality of Information (QoI) becomes more important than ever. In this paper, we provide a new framework for defining and enforcing the QoI in mobile crowdsensing, and analyze in depth the current state-of-the-art on the topic. We also outline novel research challenges, along with possible directions of future work.

1 INTRODUCTION

Mobile crowdsensing combines pervasive smartphones, embedded sensors, connectivity, and human participation to monitor diverse phenomena, but human unreliability creates distinctive QoI challenges. The paper surveys this field, proposes a unified QoI framework, analyzes enforcement and estimation challenges, and identifies future research directions.

  • Smartphones combine widespread use, multimodal sensors, wireless connectivity, and processing capabilities for mobile crowdsensing applications.
  • Mobile crowdsensing introduces research challenges because human participants may choose whether and how to submit information.
  • Mobile crowdsensing applications require different notions of useful information, such as frequent traffic updates but less frequent pollution measurements.
  • The paper addresses a literature gap by proposing a unified QoI framework and analyzing challenges in defining, enforcing, and estimating QoI.
  • The paper examines existing work’s limitations and provides a roadmap for future research on QoI in mobile crowdsensing.

2 WHAT IS MOBILE CROWDSENSING?

Mobile crowdsensing uses smartphones to acquire timely, fine-grained information about locations or events through participants, applications, and a crowdsensing platform. The platform processes, stores, filters, redistributes, and coordinates the collected data.

  • Mobile crowdsensing acquires up-to-date and fine-grained information about locations or events using widespread smartphones.
  • Its architecture generally consists of participants, a sensing application, and a mobile crowdsensing platform.
  • Participants use smart devices to capture sensor data including location, images, sound, accelerometer readings, biometric data, and pressure.
  • The platform filters, processes, stores, redistributes, and coordinates sensed data and system operations.

3 WHY IS THE QOI PROBLEM IN MOBILE CROWDSENSING SO CHALLENGING?

QoI in mobile crowdsensing cannot be defined solely through traditional network-oriented quality measures because applications handle diverse information and rely on voluntary human participation. Human behavior also creates unstable data flows and opportunities for malicious disruption.

  • Traditional sensor-network quality measures focus on QoS attributes such as bandwidth, delay, jitter, and packet loss.
  • These network-oriented measures do not fully capture application-oriented quality in mobile crowdsensing.
  • Mobile crowdsensing systems handle diverse information types and application purposes, requiring quality notions suited to each system.
  • Voluntary human participation prevents systems from assuming a steady flow of sensing information.
  • Malicious participants can disrupt sensing applications by submitting low-QoI information for rewards, spamming, or influence.

4 A COMPREHENSIVE FRAMEWORK FOR QOI IN MOBILE CROWDSENSING

The paper defines QoI through application-specific quantity and accuracy constraints and organizes their achievement into a QoI loop. It surveys related quality concepts and methods for collecting, estimating, enforcing, and extracting useful information.

  • The framework defines QoI broadly enough to cover diverse mobile crowdsensing applications and existing QoI research.
  • QoI constraints comprise information quantity and information accuracy, while QoI enforcement and estimation are the activities used to achieve them.
  • The QoI loop requires applications to quantify constraints before deployment and uses participant recruitment and incentives during collection.
  • Information accuracy covers objectivity, precision, completeness, and timeliness, while information quantity specifies received information over time.
  • QoI estimation checks whether collected data meets application constraints, while enforcement supports useful inference from conflicting reports through truth discovery or trust frameworks.
  • The framework is situated within prior QoI work spanning enterprise data management, information fusion, telecommunications, and sensing.

5 TRUTH DISCOVERY

This section surveys truth discovery in mobile crowdsensing, formalizes its problem and background, reviews existing frameworks, and identifies research challenges.

  • 5 TRUTH DISCOVERY: The section defines truth discovery, introduces the concepts needed to understand existing frameworks, surveys related work, and discusses research challenges.It proceeds through an overview and problem definition, mathematical background, detailed framework discussion, and concluding challenges.

5.1 Overview, Problem Definition and Related Work

Truth discovery addresses conflicting participant reports by jointly inferring entity truths and source reliability, using iterative methods adapted to mobile crowdsensing conditions.

  • Problem motivation: Conflicting reports about the same event create the central truth-discovery problem: identifying which value best represents reality.The paper illustrates this with three differing gas-price reports and notes that errors can arise from devices, environments, users, or networks.
  • Problem motivation: Participant reliability is usually unknown beforehand, so truth discovery must infer reliability while improving the accuracy of aggregated results.A participant with privileged knowledge may be more reliable than others, but that reliability is not directly available a priori.
  • Problem definition: The formal problem is to infer an optimal truth value for each monitored entity from observations submitted by multiple sources.The paper represents entities, sources, and source reports explicitly before stating this general objective.
  • Related work: Existing approaches include fact-finders, reputation-based methods, MLE and Kalman-filtering frameworks, and machine-learning techniques for improving data quality.Fact-finders iteratively estimate claim credibility from source credibility and source credibility from claims, while other work targets low-quality labelers and data aggregation.
  • General framework: General truth-discovery algorithms alternate between computing entity truths from source weights and updating source weights from the current truths until convergence.The framework takes source observations as input and returns optimal truths together with estimated source reliabilities.

5.2 Background Concepts

This section introduces likelihood-based statistical foundations, explaining how likelihood, MLE, MAP, and EM formulate and solve parameter-estimation problems.

  • Statistical foundations: The background models observations as i.i.d. samples from a parametric family with an unknown parameter vector to be estimated.The parameter θ is treated as the unknown true value of the model’s parameter vector.
  • Likelihood: For independent observations, the joint density factors into the product of the individual conditional densities, which becomes the likelihood when observations are fixed.The likelihood varies the parameter while holding observed values constant; the log-likelihood is often algebraically more convenient.
  • Maximum likelihood estimation: Maximum likelihood estimation chooses the parameter value that maximizes the likelihood or equivalently its log-likelihood when a maximum exists.The equivalence follows because the logarithm is monotonically increasing; the formulation also extends beyond i.i.d. data when a joint density can be written.
  • Maximum a posteriori estimation: Maximum a posteriori estimation maximizes the posterior probability by combining the likelihood with a prior distribution over the parameter.The MAP estimate coincides with the ML estimate under a uniform prior, while nonuniform priors add information about the quantity being estimated.
  • Expectation maximization: Expectation maximization estimates parameters in models with unobserved latent variables by alternating an E-step and an M-step until convergence.The E-step computes an expected log-likelihood under the current parameters, and the M-step updates parameters to maximize that quantity.

5.3 Truth Discovery in Mobile Crowdsensing

The surveyed mobile-crowdsensing truth-discovery literature is dominated by MLE-based EM methods, alongside MAP, optimization, privacy-preserving, and matrix-factorization approaches.

  • Taxonomy and approaches: Truth-discovery research in mobile crowdsensing is organized into MLE-based EM, MAP, optimization-based, privacy-preserving, and miscellaneous approaches.The paper presents a taxonomy and comparative tables covering these methodological families.
  • MLE and MAP methods: Most existing methods use MLE-based EM to estimate source reliability and infer truth values, while fewer use MAP with an informative prior.MAP is described as closely related to MLE but with an augmented objective incorporating prior information.
  • MLE and EM methods: EM-based truth discovery commonly assumes independent sources, although later work incorporates social-sensing dependencies and other data characteristics.The first EM formulation quantified measurement correctness and participant reliability under a no-correlation assumption.
  • MLE and EM methods: Convergence and computational efficiency are key EM performance indicators because mobile crowdsensing requires timely truth estimation.Comparative analysis examines these properties across MLE-based EM algorithms.
  • Research limitations: Most surveyed EM studies omit convergence proofs or computational-complexity analysis, and binary claims can create an exponential search space for n sources.Some studies reach a stable error rate after 3-6 iterations, but the reviewed works do not discuss approximation methods for obtaining sub-optimal solutions in reasonable time.
  • MAP methods: MAP-based QTF models quantitative annotation and counting tasks, while spatial-event models preserve participation privacy and avoid treating missing reports as negative evidence.QTF-A models annotation and QTF-M models counting; TSE and PTSE model spatial events using MAP estimation.
  • Optimization-based methods: Optimization-based methods jointly estimate source reliability and truth values, sometimes incorporating correlated entities, topic-specific expertise, redundancy, or sparsity.Examples include block coordinate descent, correlated-entity objectives, FaitCrowd, and RST.
  • Other methods: Other approaches address privacy-preserving truth discovery and missing information, including encrypted distance computation and matrix-factorization-based inference.These methods expand truth discovery beyond the dominant likelihood and optimization formulations.

5.4 How to Improve Existing Work

The surveyed truth-discovery literature has unresolved computational, adversarial, behavioral, data-model, scalability, and streaming limitations. The paper identifies these gaps as directions for improving QoI estimation in mobile crowdsensing.

  • Truth-discovery approaches span MLE-based EM, MAP-based unsupervised, optimization-based, privacy-preserving, and missing-information methods.
  • Most existing approaches are iterative and computationally expensive, while most EM schemes lack convergence guarantees.
  • Current methods do not adequately address malicious participants who collude or compromised reliable users that generate false observations.
  • Reliability is often modeled as static even though human participants may behave irrationally, motivating dynamic reliability models.
  • Most models assume binary claims, although crowdsensing data can include text, audio, video, and images, making truth discovery non-trivial.
  • Open crowdsensing systems require scalable truth finding as participant and entity counts grow, while many EM methods remain designed for static datasets and rerun from scratch on updates.

6 TRUST FRAMEWORKS

Trust frameworks assess the reliability of mobile-crowdsensing contributions through trust scores and a recurring trust loop. The surveyed framework collects evidence, computes participant trust, and uses it to infer submitted-data QoI while facing scalability and evidence-reliability challenges.

  • Trust expresses whether interactions with an entity are dependable based on evidence from prior or current behavior, while technological trust quantifies that trustworthiness.
  • A trust function maps each entity to a trust score, and reputation aggregates beliefs about an entity across multiple parties.
  • The trust loop comprises activities for collecting trustworthiness evidence, computing trust scores, and assessing the QoI of participant-submitted data.
  • Trust computation uses a participant’s previous trust value and current evidence to produce a current score, which supports QoI inference through trust aggregation.
  • Mobile crowdsensing trust frameworks must address scalability and other challenges that existing frameworks alone cannot resolve.
  • Evidence-collection approaches include ground-truth data, contextual factors such as location and time, participant expertise, and outlier or similarity detection.

6.3 Computation of Trust Scores

The surveyed trust-score models include Gompertz, belief-based, Beta, Dirichlet, and fuzzy-logic approaches. They differ in how they represent evidence, participant dynamics, and uncertainty, while each has stated applicability or limitations.

  • 6.3.1 Gompertz’s Function Model: Gompertz’s function models trust over time with slow initial and final growth and asymmetric approaches to its lower and upper asymptotes.
  • 6.3.1 Gompertz’s Function Model: A watchdog can convert distance similarity between reports and consensus into ratings, then use Gompertz’s function to convert past cooperation into cumulative reputation.
  • 6.3.1 Gompertz’s Function Model: The Gompertz approach is easy to implement and models dynamic participant behavior, but it does not account for substantial numbers of malicious participants.
  • Jøsang’s belief model combines reliability and participation opinions to produce trust scores used for incentives and sensing-report acceptance.
  • Beta-distribution trust scoring represents binary positive and negative evidence, with trust modeled from counts of successful and unsuccessful interactions.
  • Dirichlet models represent multinomial evidence by estimating subjective probabilities for multiple competing outcomes from observation counts.
  • 6.3.5 Fuzzy Logic Model: Fuzzy-logic approaches compute trust from collected evidence in publish/subscribe sensing systems built around online social-network participants.

6.4 How to Improve Existing Work

Trust frameworks are useful for implementing QoI because they are easy to implement and can adapt to dynamic participant behavior, but important robustness and scalability questions remain unresolved.

  • Trust frameworks offer ease of implementation, flexibility, and adaptability to participants’ dynamic behavior.
  • Contextual information can be forged by malicious participants, so its appropriate use for trust assessment requires further study.
  • The effects of different participant behaviors on trust scores remain unclear, including how to choose parameters that improve robustness and encourage good behavior.
  • Most existing work does not examine the scalability or computational complexity of trust frameworks, despite the large scale of mobile crowdsensing systems.
  • Rigorous theoretical analysis and implementations on parallel frameworks such as MapReduce are needed to make trust frameworks viable in real-world scenarios.

7 OPEN RESEARCH CHALLENGES AND DIRECTIONS

The paper identifies open challenges in assessing complex information, validating gamification, and modeling participant behavior through game-theoretical trust frameworks.

  • Assessing Trustworthiness of Complex Information: Existing work largely addresses binary or multinomial information, leaving complex multimedia and qualitative information insufficiently handled.
  • Assessing Trustworthiness of Complex Information: Multimedia QoI requires algorithms that can compute and compare quality across audio, video, and image inputs, but remains substantially unexplored in mobile crowdsensing.
  • Assessing Trustworthiness of Complex Information: Opinion-based systems require new ways to model and manage opinions, potentially using trust or fuzzy-logic mechanisms.
  • Gamification for QoI Improvement: Gamification may motivate reliable contributions, but its effectiveness and applicable conditions in mobile crowdsensing remain uncertain.
  • Gamification for QoI Improvement: Future work should develop modeling languages and integrate trust or truth-discovery frameworks to capture gamification’s complex interactions.
  • Game-theoretical Trust Frameworks: Behavioral game theory and bounded rationality are proposed to model participants whose information, strategies, and learning processes may be imperfect.
  • Game-theoretical Trust Frameworks: The rationality parameter λ controls action selection from random behavior as λ →0 to choosing higher expected-payoff actions as λ →∞.
  • Game-theoretical Trust Frameworks: Additional research is needed to model λ and obtain more realistic equilibria, potentially using large-scale campaigns with sufficient participants.

8 CONCLUSIONS

Mobile crowdsensing can reduce infrastructure costs and provide detailed information, but human participation introduces unresolved challenges for guaranteeing QoI. The paper surveys QoI, proposes a framework, analyzes research challenges, and offers a future-work roadmap.

  • Mobile crowdsensing can reduce infrastructure costs while providing detailed information about monitored phenomena.
  • Human dynamics and predictable behavior are fundamental to guaranteeing QoI because people remain in the sensing loop.
  • The paper surveys QoI, proposes a comprehensive framework, analyzes challenges in enforcing and estimating QoI, and reviews related work.
  • The paper’s analysis of existing work produces a roadmap for future research, while energy consumption, security and privacy, and efficiency remain additional challenges.
Loading 1709.03583v1…