Research On User Behavior Analysis And Task Scheduling Of Crowdsourcing Based On Uncertain Task Environment

Crowdsourcing is a working mechanism that schedules Internet distributed user resources to collaborate to complete a large number of tasks.In a crowdsourcing task,workers of different professional skills fulfill different tasks through a certain platform in an uncertain environment.Due to this open working mode,the result of crowdsourcing tasks often contains a lot of low-quality data.At present,the commonly used quality control method is usually to improve the quality of tasks by increasing the number of workers and assessing the capabilities of workers in a defined work environment.Although these methods can improve the effect,there are still some problems that cannot be ignored: On the one hand,they ignore many uncertainties in the process of crowdsourcing interaction,which affects their effectiveness in the real environment;on the other hand,they have also led to a linear increase in crowdsourcing costs while improving the quality of the tasks.In order to solve the above problems,this project firstly analyzes and understands the behavior of crowdsourcing user interaction in the real environment,and proposes relevant suggestions for future crowdsourcing platform construction and interaction mechanism design,and based on this,it designs a crowdsourcing scheduling mechanism in uncertain task environment to achieve an effective balance between crowdsourcing quality and cost.The main work and contributions of this article are as follows:(1)Qualitative analysis of crowdsourcing interaction based on grounded theory.In order to fully understand the interaction of crowdsourcing in the actual environment,this article takes the current largest crowdsourcing platform,Ali crowdsourcing,as a specific research example,adopts a behavior analysis method based on grounded theory,and carries out a comprehensive survey of popular platforms in China,combined with crowdsourcing environment,we analyze Ali's crowdsourcing implementation process and user behavior and explore the impact of task environment on crowdsourcing implementation process and user behavior.At the same time,combined with our research results,we propose innovative suggestions for future crowdsourcing platform construction and interaction mechanism design in order to play a positive role in the research and development of domestic crowdsourcing technology.(2)Discovery of the difficulty characteristics of crowdsourcing task based on uncertain task environment.Existing studies have proved that the quality of crowdsourcing results is affected to a great extent by the difficulty of the task.Even the same batch of data,the difficulty of the task is not exactly the same.This article first uses the inherent properties of the task data and unsupervised clustering method as the grouping method of task datasets which are used to group datasets.Then,we propose a minimum sample coverage method based on greedy algorithm,and uses this method to select the sample data for each grouping to the crowdsourcing worker for annotation.Finally,we collect crowdsourcing results of the sample dataset,calculate the labeling difficulty of each subgrouping data under each grouping method using the result data,and select the optimal approximation of which the grouping feature is difficulty,corresponding to the grouping method with the biggest difference in the subgrouping difficulty.This article takes the difficulty of crowdsourcing in the actual environment into consideration and combines the actual situation of crowdsourcing to adaptively estimate the task difficulty.(3)Design of crowdsourcing scheduling mechanism based on uncertain task environment.This paper uses the difficulty feature determined by the uncertain task environment to group the crowdsourcing task sets and assign different number of workers according to the task difficulty under each subgrouping in order to improve the quality of crowdsourcing results.In addition,without affecting the quality of the crowdsourcing task results,the use of a semi-flood voting method will dynamically reduce the number of workers per task during the implementation of crowdsourcing activities,thereby reducing the actual cost of crowdsourcing.We used three data sets and conducted relevant experiments in a lab environment and a real crowdsourcing environment.The experimental results show that our crowdsourcing can effectively identify the task difficulty,improve the quality of crowdsourcing results,and can significantly reduce the cost of crowdsourcing,thus demonstrating the effectiveness of the task difficulty as a task modeling attribute and the rationality of approximation method.