QADM: A quality-aware data model for freelancing applications based on recommendation systems

Abstract

Freelancing systems such as Freelancer and Upwork have recently been increasing in popularity due to their ability to provide efficient solutions. Nevertheless, the success of freelancing depends on the delivery of high-quality freelanced output. Indeed, quality control is a major concern and is affected by the uncertainty stemming from human factors relating to the diversity of workers’ skills and the workers’ changing behaviour over time. This article proposes a quality-aware freelancing data model (QADM) that takes into account both the diversity of worker skills and workers’ changing behaviour. The QADM is comprised of three submodels: skill, task and worker. The main goal is to model worker quality appropriately, and it does so by effectively modelling (1) worker suitability for new tasks, (2) worker reputation, (3) worker accuracy in completed tasks and (4) worker expected accuracy in new tasks. To improve the worker accuracy estimation, a task-to-task similarity algorithm is developed that achieves higher accuracy than Cos(topic), Cos(tf-idf) and Jaccard similarity methods. The quality-aware task assignment decision problem is solved as a top-k task recommendation problem. The results achieved in this article show that the QADM accomplishes a high recommender system mean average precision for the assignment decisions.