Research Of Iterative Strategy In The Crowdsourcing Quality

With the rapid development of network information technology, the crowdsourcing appear as a new service mode on the Internet platform, which is attracting more and more workers to participate in. In recent years, the crowdsourcing has been widely applied to many fields in the business, scientific research and has achieved good results. However, the crowdsourcing relies on the complex online trading platform and the task of release is geared to the needs of all users of the Internet, because accept workers with the identity of the anonymous, each worker ability size, different attitude, it lead that the result of crowdsourcing task has greater uncertainty, which cannot meet the requirements of task demanders. Therefore, how to effectively improve the quality of crowdsourcing results is the current needs to solve the hot issues at domestic and foreign crowdsourcing study.This article first introduces the concept of crowdsourcing model and the research status at domestic and abroad based on reading the large number of the related literature of crowdsourcing at domestic and abroad, and then summarizes the crowdsourcing model in the business and the application of scientific research. After it introduces the current research work about the package quality control and analyses the deficiency of the existing crowdsourcing quality evaluation methods and then gives the corresponding improving methods. It is of great significance to improve the package quality evaluation result. In the paper, the main work is as follows:First of all, the paper proposes a crowdsourcing quality evaluation architecture including crowdsourcing task distribution, workers classification and iterative detection strategy three modules. In the architecture, it firstly use the classification algorithm to classified workers participate in the task of crowdsourcing and select the excellent workers to join the candidate crowd. Then select workers involved in crowdsourcing task from the candidate crowd. According to the task crowdsourcing workers complete, we can use the principle of the minority is subordinate to the majority to assess the results of tasks. The result sets of the assessment task in which candidate option is not unique will be regarded as new task to release in the crowdsourcing platform. Candidated workers will be chose to participate in the iterative detection operation until the optimal result of each task has been determined. It can effectively identify the assessment tasks that exist in the similar results through crowdsourcing iterative detection strategy and improve the accuracy of crowdsourcing quality evaluation. Experimental results show that compared with classic quality assessment algorithm based on entropy, the mechanism can get better effect.On the other hand, according to the present most crowdsourcing quality evaluation algorithm is just only from the accuracy taking into consideration to improve the quality of crowdsourcing result, it usually rarely consider crowdsourced workers pay cost problem, the paper proposes a iterative approximation crowd consistent algorithm. The algorithm firstly set up the quality evaluation model for workers, according to the evaluation results workers complete, we select the high quality workers as the represent to compose the minimal subset. Then the subset of results are weighted sum, if the result can’t response the majority of the views, the candidate workers will be selected to join the minimal subset to weighted summation until the task results close to the majority opinion of the crowd. In order to reduce the cost of crowdsourcing to pay, algorithm can accurately choose the typical representatives of the crowd after several iterative operation. By comparing different data sets and benchmark algorithms, the results show that the algorithm can achieve better effect.