Research On Task Allocation Algorithm For Team Crowdsourcin

Crowdsourcing has become an effective measure to solve the problems that are hard for automated algorithms by integrating anonymous workers from the Internet,which has been widely concerned in many fields such as human-computer interaction,artificial intelligence,and has become a new research hotspot.With the continuous development of the crowdsourcing technology,more and more Internet users register as crowdsourcing workers.However,these workers come from all over the world and have different cultural backgrounds and ability levels,resulting in great differences in the quality of completing the same task.And crowdsourcing is not free,the workers must receive certain financial rewards or other forms of rewards for each crowdsourcing task they complete.In addition,the feedback speed of tasks result may be slow due to various reasons such as untimely worker response.Therefore,the key objectives of crowdsourcing result quality,cost and task completion time must be optimized and balanced in crowdsourcing.Recent studies proved that workers compete against each other can improve crowdsourcing efficiency.However,when individual workers compete,they may be demotivated by poor performance.Workers often form teams and seek to collaborate with each other through social connections in order to achieve a competitive advantage.To this end,we use team mechanism in the crowdsourcing process,utilizing a combination of competition,collaboration,and motivation to improve crowdsourcing efficiency.Specifically,the contributions of this thesis are as follows:(1)To address the problem of ability differences among workers,we propose a algorithm named Team Crowdsourcing Algorithm based on Individual Working-Capability and Social Activity.Firstly,the algorithm uses the team as the unit of commission payment to stimulate competition between teams.Two factors,the number of team members and the balance of workers’ ability,are taken into account in the team formation to avoid the extreme situation of uneven number of people or uneven workers’ ability.Secondly,we also design an incentive mechanism based on workers’ social activity to encourage synergy between team workers.Experimental results show that the algorithm can effectively alleviate the impact of crowdsourcing workers’ ability differences,reduce crowdsourcing costs while ensuring the quality of task results.(2)Aiming at the problem that forced teaming among unfamiliar workers may lead to inactive cooperation and lack of overall team competitiveness,we propose a algorithm named Reliable Team Crowdsourcing Algorithm based on Workers’ Social Network.The algorithm leverages social connections among workers and allows workers to independently weigh team size,membership strengths and communication costs to complete team formation.In addition to commission rewards,a reputation reward mechanism is designed to correlate the value of the team’s markup and the quality of the crowdsourcing results with the performance of each team member to fully exploit the team’s potential.Experimental results show that this strategy can effectively mitigate the problem of decreasing worker motivation and ensure high quality annotated data at a lower cost.(3)We design and implement a data annotation system based on team crowdsourcing.In addition to the basic modules such as login authentication,task management and user management,the system also designs and implements a team management module.On one hand,it can be used as an experimental platform for running the team crowdsourcing in this thesis.On the other hand,it can provide data support for research workers’ cooperation by collecting information exchange.