Research On Incentive Mechanism Design For Task Allocation In Crowdsourcing

Crowdsourcing is a novel paradigm where a company(or organization)that uses the Internet to allocate tasks to an unspecified public in a free and voluntary manner instead of being performed by internal employees,so as to achieve a multilateral winwin situation.Since the concept of crowdsourcing was proposed by Howe Jeff in 2006,it has been successfully applied in the domain of human-computer interaction(HCI),databases,natural language processing(NLP),machine learning and artificial intelligence,information retrieval,computer theory and other disciplines.Crowdsourcing has the following advantages: finding unusual solutions,reducing management costs,thinking diversity,and faster problem solving,enrich more user data.The above-mentioned advantages have made crowdsourcing technology widely concerned by academia and industry.The majority of researchers pay close attention to the theories and methods of crowdsourcing.At the same time,many crowdsourcing commercial platforms have emerged,such as Zhubajie,Alibaba crowdsourcing,etc.The essence of crowdsourcing is a distributed problem solving and production organization model on the Internet,and it is developed from the swarm intelligence and outsourcing.It uses some specific strategies and methods to guide workers to complete some micro-tasks in order to achieve an accumulated result.Crowdsourcing allocates micro-small tasks to workers on the Internet.A common problem faced by the task allocation technology of crowdsourcing is that there are uncertainties of workers on the Internet.There is no prior experience and knowledge for the platform user to make decisions.Incentive mechanism design is a branch of crowd intelligence and game theory,which is the most effective method for solving the problem of cold start.Mechanism design focuses on the integration of humans,machines,and networks.Under the conditions of voluntary choice,free exchange,incomplete information,and decentralized decision-making,the scattered people,machines,and networks on the Internet are integrated into a whole.Mechanism design focuses on the integration of humans,machines,and networks.Under the conditions of voluntary choice,free exchange,incomplete information,and decentralized decision-making,the scattered people,machines,and networks on the Internet are integrated into a whole.Thus,crowd intelligence is an indispensable key technology in all aspects of the crowdsourcing process.This thesis aims to tackle shortcomings of existing game theory and mechanism design technology in the process of crowdsourcing,and apply the research results to platforms in practise to improve the quality of crowdsourcing services.Due to the diversity and complexity of crowdsourcing tasks,the existing technology of mechanism design have many shortcomings to be applied for crowdsourcing in practise.In this thesis,we will study the application of game models and incentive mechanism design in the process of on the Internet.For different types and requirements of tasks,the research is mainly divided into following modules: homogeneous task crowdsourcing incentive mechanism,heterogeneous task crowdsourcing incentive mechanism,online task crowdsourcing incentive mechanism and cross-platform task crowdsourcing incentive mechanism.While studying how to use mechanism design to improve the efficiency of crowdsourcing,the thesis also enriches the theoretical aspects of mechanism design.In this thesis,we will use mechanism design technology and game theory models to build task allocation auctions for four types of tasks: homogeneous,heterogeneous,online and cross-platform.Specifically,the main contributions and research innovations of this thesis can be summarized as follows:1)Incentive mechanism for homogeneous task crowdsourcing.Based on the assumption of micro-tasks(less coupling between tasks,simple capability requirements,and easy quality control),we define the problem as a mechanism design for homogeneous tasks.For the scenarios of profit maximization and budget limitation,based on the idea of profit extraction and greedy sorting,we propose a task allocation incentive mechanism with the submodular utility function.The experimental results show that the effectiveness of the proposed mechanisms,which is significantly higher than the theoretical lower bound.2)Incentive mechanism for heterogeneous task crowdsourcing.Based on the assumption of the complexity of tasks,we define the problem as a mechanism design for heterogeneous tasks.For the scenarios of multi-class tasks and topological structure based tasks,based on the ideas of cooperative game and core-selecting,we proposed an incentive mechanism for the utility sharing and and a computational accelerating algorithm.The experimental results show that the costs of the proposed mechanisms are significantly less than VCG mechanisms.3)Incentive mechanism for online task crowdsourcing.Based on the uncertainty of tasks,we define the problem is as a mechanism design for online tasks.For the scenarios of the profit maximization and budget limitation,based on the ideas of online learning and profit extraction,a two-stage sampling and multi-stage sampling frameworks are proposed.The experimental results show the utility of the proposed online mechanisms is close the utility of offline setting.4)Incentive mechanism for cross-platform task crowdsourcing.Based on the task's cross-platform needs,we define the problem as a mechanism design for cross-platform task.For the scenarios of weighted average maximization and minimum maximization,based on the idea of greedy sorting and ”divide and conquer”,a feasible budget mechanism and a budget allocation framework with platform weights are proposed.The experimental results show that the utility of the proposed mechanisms are higher than the benchmarks.