The Design Of Task Allocation Mechanism For Spatio-temporal Crowdsourcing

With the development of technology,smart devices equipped with various sensors(such as GPS,TP)and wireless mobile networks(such as 4G,5G),are more and more capable to process data.People can easily participate in crowd tasks that need to be performed at specific locations near the worker's location,such as uploading traffic information,taking photos / videos,uploading weather conditions,express pick-up and reporting waiting times for popular restaurants.Therefore,a new framework,spatial crowdsourcing,has emerged,which enables workers to move to the area where the task is located to perform crowd tasks.Meanwhile,there are more and more smart phone users worldwide,which means that a large number of crowd workers and publishers of crowd tasks will be poured into the crowdsourcing system in the future.How to allocate appropriate crowd tasks to crowd workers based on the properties of crowd workers and crowd tasks to maximize the effectiveness of the platform and improve the quality of task completion has become an important issue.This issue is also a key issue in the research of spatial crowdsourcing.Aiming at the above problems,we design a task allocation model based on social relations of crowd workers and a task recommendation model based on probability.The main research contents are shown as follows:(1)In order to improve the completion quality of crowd tasks and reduce the completion cost of crowd tasks,this thesis deeply explores the relationships between friends in social networks and applies them to task allocation in spatial crowdsourcing.A method for calculating the strength of temporal relationship based on the historical interaction information of friends and calculating the the strength of location relationship using GeoHash coding is proposed.While protecting the privacy of workers,this thesis designs an algorithm for task allocation based on the strength of social relationships.Finally,the comparison experiments are conducted using real dataset and synthetic dataset to verify the accuracy and effectiveness of the proposed algorithm.(2)In order to improve the recommendation accuracy,we proposes a task recommendation model based on probability.Firstly,the characteristics of the recommendation properties of the crowdsourcing task are considered to determine the similarity between crowd workers.Then,based on Markov model and Collaborative Filtering model,this paper considers the similarity,trajectory prediction,dwell time and trust to calculate the recommendation probability matrix.Finally,using Yelp real dataset,the effectiveness and the accuracy of the task recommendation model proposed are verified through comparison experiments.The influence of the weight of the recommendation transition probability for the recommendation results is analyzed.