PISCES: Participatory Incentive Strategies for Effective Community Engagement in Smart Cities

Arpita Biswas, Deepthi Chander, Koustuv Dasgupta, Koyel Mukherjee, Mridula Singh, Tridib MukherjeeHCOMP 2015 · OpenReview

TL;DR

This paper introduces PISCES, a novel incentive framework for participatory sensing in smart cities. It addresses the challenge of motivating citizens to contribute data by dynamically adjusting rewards to meet varying demands for reports while minimizing costs for city agencies, even with unknown reporter availability and behavior.

What Problem Does This Paper Address?

Smart cities increasingly rely on participatory sensing – using citizens' mobile devices to collect data about urban environments. This data is invaluable for urban planning, traffic management, and identifying areas for improvement. However, a key challenge is designing effective incentive mechanisms to encourage citizens to contribute data. Requesters (like city agencies) often have varying levels of utility for different types of reports based on their spatio-temporal context (e.g., a traffic congestion report during rush hour is more valuable than one at midnight). This leads to a non-uniform demand for reports.

Existing incentive mechanisms often struggle with:

  • Unknown reporter availability: Citizens are mobile and their availability can change unpredictably.
  • Unknown reporter behavior: It's hard to know in advance what rewards individuals expect or if they will report even if their expectation is met.
  • Minimizing costs: Requesters want to meet their report demand without overpaying.
  • Guaranteed report quantity: Many systems don't guarantee a specific number of reports will be collected.

This paper tackles these issues by proposing a framework that can adapt to dynamic conditions and ensure a desired number of reports are collected at the lowest possible cost.

Methods

The core of the proposed solution is the STOC-PISCES algorithm. It's a closed-loop incentive framework that operates in trials (e.g., daily).

  1. Reward Declaration: For each event (a specific type of report needed at a certain place and time), STOC-PISCES declares a reward.
  2. Observation: After a trial, the system observes how many reports were actually collected for each event.
  3. Adaptation: Based on the observed effectiveness, the reward for the next trial is adjusted. This process uses a Multi-Armed Bandit (MAB) framework combined with binary search. The MAB helps learn the optimal reward by balancing exploration (trying new rewards) and exploitation (using rewards that have worked well). Binary search efficiently narrows down the reward range.

Key aspects of the method:

  • Posted Price Mechanism: All reporters reporting the same event in a trial receive the same reward.
  • Stochastic Rationality: Reporters are assumed to be "stochastically rational," meaning their probability of accepting a reward increases with the reward value, but it's not deterministic. This accounts for real-world variability.
  • No Prior Knowledge: The system doesn't need to know individual reporter availability or their exact reward expectations beforehand.
  • Optimality Guarantees: STOC-PISCES provides probabilistic guarantees on achieving the desired number of reports while minimizing costs, even in dynamic and uncertain environments.

For comparison, the paper also introduces OPT-PISCES, a simpler algorithm that provides theoretical optimality under highly simplified assumptions (deterministic reporter availability and behavior). This serves as a baseline to highlight the advantages of STOC-PISCES in realistic scenarios.

Datasets

The experiments were conducted using:

  • Real-world mobility traces: GPS data from 143 employees commuting to and from a BPO facility in Bangalore, India, collected over a month. This data was used to model reporter mobility patterns and availability.
  • Simulated reporter profiles: The paper uses sigmoidal acceptance models to simulate both "greedy" (highly reward-sensitive) and "altruistic" (less reward-sensitive) reporters.
  • Simulated events and demands: Twelve distinct events were defined, mapped to specific locations and times, with varying demands (low, normal, high) to represent different levels of utility for requesters.

Key Insights and Ideas

  • Dynamic Adaptation is Crucial: The paper strongly emphasizes that static incentive mechanisms fail in dynamic urban environments. The ability to learn and adapt rewards based on real-time feedback is essential for effective participatory sensing.
  • Stochasticity is the Norm: Real-world reporter behavior and availability are inherently uncertain. The STOC-PISCES framework's strength lies in its ability to handle this stochasticity, unlike simpler models that assume deterministic conditions.
  • Cost-Effectiveness Through Learning: By iteratively adjusting rewards, STOC-PISCES can converge to the minimum reward needed to meet demand, leading to significant cost savings for requesters compared to simpler or non-adaptive methods.
  • Scalability with Supply: The experiments suggest that as the number of available reporters (supply) increases, the cost per report decreases, and convergence to optimal rewards happens faster. This is a positive sign for large-scale smart city deployments.
  • OPT-PISCES as a Baseline: The comparison with OPT-PISCES highlights that theoretical optimality under simplified assumptions doesn't translate well to real-world complexity. STOC-PISCES, while not strictly optimal in a deterministic sense, is more practical and effective.

Key Results

  • Lower Costs: STOC-PISCES consistently achieved lower average payments (costs) for requesters compared to OPT-PISCES, especially under high demand scenarios.
  • Guaranteed Report Quantity: The system demonstrated that it could achieve the desired number of reports with high probability (e.g., 80% confidence with ε=0.2).
  • Consistent Rewards: STOC-PISCES provided more stable and less variable reward values across different runs compared to OPT-PISCES, which showed drastic fluctuations.
  • Scalability: The framework showed good scalability with an increasing number of reporters, requiring fewer trials to converge.
  • Efficiency: The algorithm converges to near-optimal rewards in a polynomial number of trials.

Limitations and Future Work

  • Data Veracity and Reputation: The paper acknowledges that a critical extension is incorporating data veracity and reporter reputation into the incentive mechanism. Ensuring reports are valid and from trusted sources is vital for city agencies.
  • Collusion and Malicious Reporters: The authors plan to explore how to discourage collusion among reporters and prevent malicious actors from gaming the system by sending multiple similar reports to meet demand.
  • Estimation-Theoretic Approach: Future work will leverage estimation theory to compute report veracity as part of the incentivization process.
  • Real-World Deployment: While simulations are extensive, real-world deployment and testing would be the next logical step to validate the framework's performance in practice.

Summary updated

BibTeX
@inproceedings{biswas2015pisces,
  author = {Arpita Biswas and Deepthi Chander and Koustuv Dasgupta and Koyel Mukherjee and Mridula Singh and Tridib Mukherjee},
  title = {{PISCES: Participatory Incentive Strategies for Effective Community Engagement in Smart Cities}},
  booktitle = {HCOMP},
  year = {2015},
  url = {https://openreview.net/forum?id=F_M3KDtYUM-},
}
Figure