A visual analysis system and method for improving the quality of crowdsourcing annotated data

The invention discloses a visual analysis system and method for improving the quality of crowdsourcing annotated data, and the system comprises a confusion matrix visualization module which displays the confusion degree between different types so as to select an analyzed confusion type according to the confusion degree; an example virtualization module which displays the mutual influence information between the uncertain class label of each instance and the instance; an annotator behavior visualization module which is used for displaying the annotation accuracy and the invalid annotation degree score of each annotator on the selected category through the scatter diagram so as to determine invalid annotators; And an interactive progressive confirmation module which is used for propagating the class label of the instance and confirmation information of the reliability of the annotator by the user so as to recommend another instance needing to be annotated and the annotator. The four modules of the system are tightly combined with the crowdsourcing learning model, so that a user is helped to confirm an uncertain instance and an unreliable annotator in an interactive progressive mode,and the data annotation quality is improved.