Research On Task Matching Of Spatial Crowdsourcing Based On Team Diversity

Spatial crowdsourcing refers to a crowdsourcing model based on the Internet and mobile devices,which connects people and computer resources distributed in different locations through network connections to provide services for task publishers.Unlike traditional crowdsourcing,spatial-temporal crowdsourcing focuses on the spatial location and time constraints of tasks,as well as the geographical location and mobility of participants.It can be applied in many fields such as traffic monitoring,map making,photo annotation,environmental monitoring,etc.Team-based spatial-temporal crowdsourcing is an extended form of spatial-temporal crowdsourcing,which introduces a team of multiple task performers to collaborate in completing tasks.Unlike traditional spatial-temporal crowdsourcing,team-based spatial-temporal crowdsourcing emphasizes the efficiency and quality of team collaboration in completing tasks,while also being more flexible and scalable.Previous research has mainly focused on simple task scenarios,such as one-toone task assignment in the field of ride-hailing services.In such cases,task modeling is relatively simple,and many studies focus on optimizing matching algorithms,dynamic scenarios,and optimization objectives.However,these studies did not consider complex task modeling,such as some complex spatial-temporal crowdsourcing tasks that not only require a team to complete but also consider the diversity of team members.In fact,considering team diversity can not only cope with more complex tasks but also improve the quality of task assignment.Team diversity can be reflected in various aspects in spatial-temporal crowdsourcing,for example,1)skill diversity: team members have different skills and professional knowledge.In a software development team,team members can be responsible for different modules and work together to complete a complex software project.2)cultural diversity: team members come from different cultural backgrounds.In a multinational company’s marketing team,team members can consider the product positioning and promotion strategy from different cultural perspectives.The main research content of this thesis is as follows:(1)This thesis proposes a spatial-temporal crowdsourcing task assignment problem based on team diversity.In addition to considering traditional constraints such as range,time,skills,and budget,this thesis also emphasizes the necessity of diversity constraints and introduces submodular functions to measure team diversity.The thesis first adapts a diversity greedy algorithm to solve this problem.However,the diversity greedy algorithm only obtains a local optimal solution,and there are many constraints in task assignment,which can easily lead to conflicting allocation schemes.Therefore,this thesis proposes a diversity divide-and-conquer algorithm,which decomposes the large-scale problem into multiple sub-problems and solves the conflict situation while solving the sub-problems.(2)This thesis further analyzes the problem of time and space crowdsourcing task assignment based on team diversity,focusing on the impact of current task allocation results on the future spatial distribution of unmanned vehicles.This thesis proposes the Deep Matching Network(DMN)algorithm,which connects the spatial distribution of unmanned vehicles with the distribution of tasks,shortening the time for unmanned vehicles to reach the next task location.The core of this algorithm is a globally shared value function,which uses online experience generated from real-time systems for continuous updates and can quickly adapt to real-time dynamic environments while maintaining high performance.The DMN algorithm can quickly adapt to new changes in the environment and can effectively deal with situations where the task volume at individual locations suddenly changes.For example,if a large number of tasks suddenly appear in a certain area,this framework can quickly adjust the task allocation strategy to better meet the demand.This thesis validates the correctness,feasibility,and performance of the algorithms using the spatial-temporal crowdsourcing task dataset provided by Didi Chuxing’s Gaia initiative.Experimental results show that both diversity greedy algorithm and diversity divide-and-conquer algorithm can effectively improve task allocation quality and meet diversity constraints in the first experimental setting.Among them,the diversity divide-and-conquer algorithm performs the best among the four algorithms,with task completion rate and team diversity of 94% and 0.76,respectively.In the second experimental setting,the online learning algorithm based on value function can quickly adapt to dynamic environments,reduce total travel distance,improve task completion rate,and enhance the working efficiency of unmanned vehicles.The task completion rate of this algorithm is 96%,the team diversity is 0.76,and the average efficiency is390.Compared with other algorithms,the average efficiency has increased by 8.3%.