A Framework for Optimal Worker Selection in Spatial Crowdsourcing Using Bayesian Network

Nor Aniza Abdullah,Md Mujibur Rahman,Khairil Imran Ghauth,Mohammad Mustaneer Rahman

doi:10.1109/access.2020.3005543

Nor Aniza Abdullah, Md Mujibur Rahman + Show 2 more

Open Access

https://doi.org/10.1109/access.2020.3005543

Copy DOI

Journal: IEEE Access	Publication Date: Jan 1, 2020
Citations: 46	License type: CC BY 4.0

Affiliation: University of Malaya, Multimedia University

Abstract

Spatial Crowdsourcing (SC) is a new paradigm of crowdsourcing applications. Unlike traditional crowdsourcing, SC outsources tasks to distributed potential workers, and those who accept the task are required to travel to a predefined location to complete it. Currently, the primary aim of SC is to maximize the number of matched tasks or to minimize the travelling distances of the workers. However, less focus is given in matching the right tasks to the right workers, particularly in a heterogeneous tasks environment. To address this lacking, our work provides an efficient framework for selecting optimal workers for every task with various specification (geographical proximity, domain types, and expiration times), based on workers’ attributes (task domain-specific knowledge, expertise or performance history, distance to task location, and task workload distribution). We introduce the use of Bayesian Network in modelling and selecting optimal workers, and use k-medoids partitioning technique for tasks clustering and scheduling. Our experimental results on both synthetic and real-world large datasets have shown that our proposed approach has outperformed other baseline approaches, in terms of low average error rate and fast execution time.

Highlights

Crowdsourcing refers to an emerging distributed problemsolving paradigm that incorporates the power of both human computations and machine intelligence
It is imperative to model an optimal worker selection mechanism that will depend on three of the most fundamental factors based on the dedicated task requirements: i) domain-specific knowledge and expertise to identify the expert workers in a particular domain, ii) distance to task location to minimise the workers’ travelling cost, and iii) task workload distribution to improve the workload balancing among the platform workers
We found that when the Greedy algorithm was first executed on the workers’ expertise, and executed it again on the outcome from the first match but this time the matching is based on the task workload scores, the resulting output was not optimised for both expertise and workload

Summary

INTRODUCTION

Crowdsourcing refers to an emerging distributed problemsolving paradigm that incorporates the power of both human computations and machine intelligence. It is imperative to model an optimal worker selection mechanism that will depend on three of the most fundamental factors based on the dedicated task requirements: i) domain-specific knowledge and expertise to identify the expert workers in a particular domain, ii) distance to task location to minimise the workers’ travelling cost, and iii) task workload distribution to improve the workload balancing among the platform workers. Besides these factors, the mechanism should be computationally efficient to meet the real-time matching demands on SC platform.

RELATED WORK

PROPOSED FRAMEWORK

EXPERIMENT

EXPERIMENTAL SETTINGS

EVALUATION METRIC

Findings

CONCLUSION