Scalable and Cost-Effective Assignment of Mobile Crowdsensing Tasks Based on Profiling Trends and Prediction: The ParticipAct Living Lab Experience

Paolo Bellavista, Antonio Corradi, Luca Foschini, Raffaele Ianniello

TL;DR

This paper introduces ParticipAct, a large-scale mobile crowdsensing platform that tackles the challenge of efficiently assigning tasks to volunteers. By leveraging user profiling, mobility prediction, gamification, and timely geo-notifications, ParticipAct significantly improves task acceptance and completion rates, paving the way for more effective smart city sensing.

What Problem Does This Paper Address?

The proliferation of sensor-equipped smartphones presents a massive opportunity for collecting valuable data in urban environments through Mobile Crowdsensing (MCS). However, effectively managing and assigning data collection tasks to a large, diverse, and voluntary user base is a significant challenge. Key issues include motivating participation, ensuring data quality, minimizing user burden, and efficiently matching tasks to the most suitable individuals. Existing approaches often lack real-world validation or focus on only a subset of these problems. ParticipAct aims to address these technical and social challenges by providing a comprehensive platform and evaluating novel task assignment strategies in a realistic, long-term living lab.

Methods

ParticipAct employs a client-server architecture. The client, running on users' smartphones, manages task reception, data collection, and uploading. The server handles data storage, processing, and the crowdsensing management interface. Key innovations include:

  • Participant Selection Algorithms: The paper explores and evaluates four task assignment policies: random, recency, frequency, and dbscan. These policies aim to predict user behavior and assign tasks to those most likely to accept and complete them, based on historical mobility data.
  • Gamification: To boost user engagement and retention, ParticipAct incorporates reputation systems, rankings, badges, and an "aging" mechanism for points.
  • Task Co-Creation: Users are empowered to define and propose their own MCS tasks, fostering a sense of ownership and creativity.
  • Timely Geo-notification: Tasks can be geo-localized, with notifications delivered only when users enter the designated area, improving relevance and reducing task abandonment.
  • Mobile Sensing Technology (MoST): An open-source Android sensing library is used to manage sensor access efficiently and minimize intrusion on user devices.

Datasets

The study utilizes a large-scale, real-world dataset collected over more than a year from the ParticipAct living lab. This dataset comprises mobility traces from over 170 students at the University of Bologna, covering various urban areas. The data includes GPS and Wi-Fi location data sampled every 180 seconds.

Key Insights and Ideas

  • Geo-notification is Crucial: Notifying users only when they enter the relevant geo-execution area significantly boosts task completion rates and precision, likely by reducing the chance of users forgetting about the task.
  • Task Complexity Matters: Users are more willing to accept and complete passive tasks or simple active tasks. Complex tasks requiring multiple actions see lower participation.
  • Gamification "Level" Strategies Excel: "Level" based point assignment strategies (e.g., Average Level, Sum Level) are more effective than "Reputation" based ones in driving user response and leaderboard climbing, though they risk widening the gap between high and low-ranked users.
  • User-Generated Tasks Drive Engagement: Allowing users to create tasks increases their involvement and understanding of the platform, leading to suggestions for improvement.
  • Social Connections Influence Participation: While users form limited friendship groups, these social ties can be leveraged to encourage participation and cooperation.

Key Results

  • Task Assignment Policy Performance: The recency and frequency policies perform similarly and are superior to random. The dbscan policy shows high accuracy and stable behavior, achieving these results with a very low number of assigned users.
  • Geo-notification Impact: Geo-notification improves acceptance rates by 52% and completion rates by 423% compared to non-geo-notified tasks.
  • Acceptance/Completion Rates: Passive and simple active tasks have higher acceptance and completion rates than complex tasks.
  • Gamification Boost: After introducing gamification features, ParticipAct saw acceptance rates rise to 25.5%, compared to a lower rate before.
  • User Participation: Most users completed around 30 tasks, with a significant number completing over 70, and some exceeding 100 tasks.

Limitations and Future Work

  • Generalizability: While the dataset is large, the authors acknowledge that results from a student population at specific university campuses might not be directly generalizable to all citizen volunteers in a large city. They plan to address this by considering dynamic weights based on population diversity.
  • Bias: The majority of volunteers (135 out of 170) are from the School of Engineering and Architecture, which could introduce bias. Future work will involve smoothing these effects with correction metrics.
  • Future Work: The team is developing new smart city services and applications using the ParticipAct platform and extending the task co-creation facility to support more complex task workflows, including subtask delegation. They also aim to further refine user profiling and task assignment strategies.

Summary updated

Abstract

Nowadays, sensor-rich smartphones potentially enable the harvesting of huge amounts of valuable sensing data in urban environments, by opportunistically involving citizens to play the role of mobile virtual sensors to cover Smart City areas of interest. This paper proposes an in-depth study of the challenging technical issues related to the efficient assignment of Mobile Crowd Sensing (MCS) data collection tasks to volunteers in a crowdsensing campaign. In particular, the paper originally describes how to increase the effectiveness of the proposed sensing campaigns through the inclusion of several new facilities, including accurate participant selection algorithms able to profile and predict user mobility patterns, gaming techniques, and timely geo-notification. The reported results show the feasibility of exploiting profiling trends/prediction techniques from volunteers’ behavior; moreover, they quantitatively compare different MCS task assignment strategies based on large-scale and real MCS data campaigns run in the ParticipAct living lab, an ongoing MCS real-world experiment that involved more than 170 students of the University of Bologna for more than one year.
BibTeX
@inproceedings{bellavista2015scalable,
  author = {Paolo Bellavista and Antonio Corradi and Luca Foschini and Raffaele Ianniello},
  title = {{Scalable and Cost-Effective Assignment of Mobile Crowdsensing Tasks Based on Profiling Trends and Prediction: The ParticipAct Living Lab Experience}},
  booktitle = {Sensors},
  year = {2015},
  url = {https://openreview.net/forum?id=bcfzEeRqTn},
}
Figure