Recommendation system for workers and tasks Recommending the optimal assignment of workers to tasks

Abstract

This thesis tries to solve the problem of matching workers with tasks when unknown parameters are involved. Looking at the trend where outsourcing tasks to previously unknown parties is becoming more common, a need is definitely there to solve this problem in an efficient way. The problem can be described as a list of workers, each with an unknown list of skills, and a list of tasks, each with a known list of requirements. Any method assigning all tasks to workers, while maximizing the reward given for doing so, must be able to accurately estimate the skills of every worker to provide good results. To solve this problem when each worker only has a single skill has been shown to be possible with an algorithm called Bounded Epsilon First. This algorithm is used as a starting point for testing data with single-skill workers and single-requirement tasks, before moving on to multi-skill workers and multi-requirement tasks. No real world data was available for multi-skill matching, which is why all experimentation is done on synthetic data, generated uniformly at random. After the first phase, different matching algorithms and methods of rating worker performance were implemented and tested, producing varying results. Testing all implemented methods on real world data would surely produce interesting results, but overall, the results presented in this thesis show good promise. Our best solution, given time to estimate each worker’s skills, give results approaching 85% of the result produces by matching with all parameters known.