Rijul Magu Reputation System Idea
1. Reputation scores are calculated for both the workers and the requesters using a combination of rating methods, opinion mining of comments and statistical modelling/machine learning based on indicators such as average wage and past performance on similar tasks. However, the reputation scores are not public. This addresses some core issues of traditional numeric rating methods:
A) Users begging for 5 stars (as noted by @arichmondfuller on http://crowdresearch.stanford.edu/w/index.php?title=Summer_Milestone_9_Evaluations_of_reputation_systems_on_other_crowdsourcing_platforms )
B) Prior ratings affecting subsequent ratings (Muchnik et al. 2013)
C) Star ratings being inflated (as noted by @arichmondfuller on http://crowdresearch.stanford.edu/w/index.php?title=Summer_Milestone_9_Evaluations_of_reputation_systems_on_other_crowdsourcing_platforms )
2. For a given task, the scores of workers and requesters are matched. A list of available workers which fall below (or are equal) to the score of the requester is shown in descending order of scores. A similar process occurs on the worker side. This ensures that bad requesters/workers do not have easy access to good requesters/workers. The workers/requesters can choose their counterparts based on their textual feedback, past work and relative position on the list.
EDIT: A diagram that represents my idea.