A Hub-and-Spoke Model for Content-Moderation-at-Scale on an Information-Sharing Platform
This addresses the scalability problem for platform operators by reducing moderation costs, though it is incremental as it builds on existing statistical learning methods.
The paper tackles the high cost of content moderation on information-sharing platforms by proposing a hub-and-spoke model that leverages user ratings as features and editorial labels for a statistical learning algorithm, achieving significantly cheaper moderation in cases where all messages are public.
One of the most expensive parts of maintaining a modern information-sharing platform (e.g., web search, social network) is the task of content-moderation-at-scale. Content moderation is the binary task of determining whether or not a given user-created message meets the editorial team's content guidelines for the site. The challenge is that the number of messages to check scales with the number of users, which is much larger than the number of moderator-employees working for the given platform. We show how content moderation can be achieved significantly more cheaply than before, in the special case where all messages are public, by effectively platformizing the task of content moderation. Our approach is to use a hub-and-spoke model. The hub is the core editorial team delegated by the management of the given platform. The spokes are the individual users. The ratings of the editorial team create the labels for a statistical learning algorithm, while the ratings of the users are used as features. We have implemented a primitive version of this algorithm into our open-source DimensionRank code base, found at "thinkdifferentagain.art".