Milestone 3 TuringMachine TrustIdea 1: Distribution of work according to workers' interests

From crowdresearch
Revision as of 05:48, 19 March 2015 by Avrimmit (Talk | contribs) (Incentive/Interests Categories)

Jump to: navigation, search

Background and Motivation

The Future of Work, Kittur etal 2013 has shown that Incentives and motivation play a big role in developing sustainable crowd sourcing communities. Workers like to follow their passion and perform the tasks that suits to their skill-sets. This approach helps workers build their resume, career, earn money, and work on projects of their interests.

Design Goal

Match me with the work I am passionate about.

Task Categories

Crowd sourcing systems can be designed to solve large number of problems. The eco-system can involve tasks and work that require different kind of expertise and skills. These tasks can be broken down into various categories as shown in the figure below


Incentive/Interests Categories

Workers who participate in these tasks have different motivation as indicated in the figure below. We envision automated hiring/task recommender algorithm that can match workers with their areas of interest. This will give them a freedom to choose the projects they care about. Previous research in the field of organizational behavior has shown that individuals are more committed to work if the tasks they are performing fall under their area of interests.


Distribution of Work according to interests of workers:

  • Maintain the Master Profile Map of Tasks according to functionalities.
  • Maintain Master Profile Map of Workers' skills, experience, salary history, salary expectations, and interests.
  • Develop Clustering/Classification algorithm to maintain categorize of profiles.
  • A perfect matching/recommender algorithm to match workers profile with tasks
  • As discussed above: The two phase recommendation system using Single Value Decomposition or Collaborative Filtering can be developed. In first phase, we use rating of reviewers against workers hence recommending closest match between requestors and workers. In second phase we design workers against the task ratings. The optimization objective function can be designed using stochastic gradient descent algorithm. Furthermore, we can optimize the algorithm using baseline prediction. Finally, we can aggregate the results depending on users' preferences.