TaskRec: Using Causal Graph to Debias in Task Recommendation System

Sifeng Wang, Anran Li, Zhaoan Dong, Shengxiang Li

Abstract: In recent years, crowdsourcing has gradually become an emerging business model. Requesters from different backgrounds on the network put their requests on the crowdsourcing platform and hand them to non-professionals from different fields to complete, which not only saves working time, but also greatly reduces the economic cost. However, with the increase of crowdsourcing tasks and the expansion of platforms, the task information of crowdsourcing platforms tends to be abundant and redundant, which makes workers face the problem of information overload. It is very difficult to match the tasks they need from thousands of tasks. Therefore, task-based recommendation system emerges as The Times require, which can help workers quickly select the tasks they want to do from the massive tasks. However, there are still some problems in the current task recommendation system. For example, the existing task recommendation system tends to recommend simple and easy common tasks to workers, while the exposure probability of those complex and difficult tasks is greatly reduced, so exposure bias appears in the task recommendation system. Deviation brings great hidden dangers to crowdsourcing platforms. For example, complex tasks are hidden in the long tail for a long time and no one pays attention to them, so the task completion rate will be greatly reduced and the satisfaction of requesters will also be reduced. In this paper, an unbiased task recommendation method based on causal graph analysis is proposed to remove the bias in task recommendation system and make tasks better match.

External IDs:dblp:conf/dasc/WangLDL22