| Crowdsourcing is a new pattern of software engineering.It engages a workforce to accomplish complex tasks regardless of geographical limitation,therefore can take advantage of various human and material resources and is now growing rapidly in a variety of areas.Currently,crowdsourcing websites attract much attention and have a large group of users,such as upwork,Amazon Mechanical Turk,Topcoder,CrowdFlower,Taskcn etc.While workers on these websites have to search tasks they are interested in or matched with their capabilities all by themselves,then apply for the task.Correspondingly,task publishers will evaluate the applicants and select the task undertakers.In this task assignment pattern,tasks are described entirely in natural language which is not machine-readable and makes task assignment inefficient since task and worker cannot be matched automatically,especially when the number of workers is large.However,describing the task requirements and worker skills simply with tags is not sufficient to articulate the task publishers' needs.All the skills tagged are in the same weight and the constraints on certain skills are ignored.Moreover,the requirements of the tasks are not exhaustive and the matching rules may also vary on skills.Furthermore,if there is no single suitable worker meeting all the required skills of the task,the matching process will fail,and there is no alternative solution given to handle this issue.In this paper,we present STWM,a solution to self-adaptive task-worker matching in software crowdsourcing to solve the above task-worker matching problems.An extensible meta-model is proposed to support description of both worker skills and task requirements.Task publishers can distinguish the priority of different requirements by giving weight on every property of the task.And they will also be capable to customize their requirements and task assignment rules flexibly with this model thus to make the matching results more accurate.Moreover,the worker profile data will be periodically corrected according to the feedback data provided by the task publishers,which will reduce the subjectiveness of the description data of the workers and improve the quality of the matching result.Based on this meta-model,we define an algorithm that allows self-adaptive matching of the task requirements against the worker skills.Workers possessing the skills required by the task will be discovered automatically and recommended to task publishers in a certain order according to the preference of task publishers such as ranking workers by their skill proficiency.And we also take some measures to ensure the efficiency of the matching progress,such as clustering workers on the property dimension and performing the matching procedure concurrently.Further,several workers will be chosen to form a team according to the worker composite rule and related constraints of the task defined by the user once a single individual doesn't meet the requirements of the task.A full experimental validation with several tasks and thousands of workers has been done showing the validation of our solution.The innovation of STWM includes the following two aspects.First,it proposes an extensible model for task and worker capability description.Second,it provides a team formation mechanism to deal with the issue when there is no suitable individual worker for the task.Currently,once a task publisher can't find a worker meeting all his skill requirements from the task applicants,he would cancel the task or wait until such a worker appears,which is inappropriate.While a worker team formed by several workers may completely satisfy all the skill requirements and is quite up to the task.In our solution,such a worker team will be discovered,formed and recommended to the clients.The meta-model,adaptive matching algorithm and the team formation strategy introduced in STWM will have a great reference to the improvement of the accuracy and efficiency of the task-worker matching in the real crowdsourcing platform. |
PDF Full Text Request