Task navigation panel for Amazon Mechanical Turk

Abstract

Amazon Mechanical Turk (Mturk), as the world's largest micro-task crowdsourcing platform, provides task search services to thousands of users every day. However, the lack of navigation panel in the platform's existing user interface makes task searching difficult and tedious, resulting in users rarely finding acceptable tasks among the countless crowdsourced assignments. To lower the task search threshold of Mturk, this study creates an automated task navigation structure by combining the unsupervised task clustering algorithm and the topic recognition algorithm, which can be fine-tuned according to task characteristics and help users focus on the task type quickly and precisely. Since the navigation panel is a non-existent element of Mturk, this paper develops a well-designed questionnaire to investigate users' perceptions of the improved interface. The results show that users were dissatisfied with Mturk's current UI, preferring the navigation interface outlined in this study. In conclusion, this work provides theoretical guidance for building similar automated task navigation panels for other crowdsourcing platforms while addressing the practical problems of Mturk.