Winter Milestone 5: Samarth Sandeep

From crowdresearch
Revision as of 22:19, 14 February 2016 by Samarthsandeep (Talk | contribs) (Created page with "A research proposal for each of the three areas of research are in this milestone submission. All three revolve around using data, either through machine learning or simple st...")

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

A research proposal for each of the three areas of research are in this milestone submission. All three revolve around using data, either through machine learning or simple statistical programming, to make Daemo more intuitive.

Proposal #1: Gravitate, a system for improved task ranking


Gravitate is a task ranking system that uses behavioral statistics and machine learning to ensure maximum retention from the worker.

Problems Solved

The Gravitate task ranking system aims to solve the problem of worker retention. Right now, crowdsourcing sites, such as Amazon’s Mechanical Turk and CrowdFlower, are mainly used as second or third sources of income for workers. Daemo is trying to change this by making crowdsourced tasks something that is more stable, with the goal of creating a new way for people to earn a living. To achieve this goal, however, workers need to feel as if they are receiving tasks that help lead to further career development, make them a reasonable amount of money, and above all, give them a sense of accomplishment from the tasks they complete.

Module #1: Improved task altering

Problem: Tinder is a dating application that allows for instantaneous approval or disapproval of matches. Due to this, more people are using Tinder to start relationships than much older dating sites, like eHarmony and It's "secret sauce" is this simple interface for flipping through potential matches that would be a match for the user. Crowdsourcing platforms do not have this yet. While they do have filters for search results, there is no connection between the results that appear and the user. This requirement to search for a task most likely frustrates the user, and results in them selecting tasks that do not actually help them, and take away time from other, quicker, more satisfying activities.

Module Details: Data of some sort needs to be collected about the worker. This could be done using a questionnaire as a part of onboarding, asking questions such as "What are your key interests?" and "How much time are you willing to give tasks?." This can also be employed by using the worker's LinkedIn, Twitter, or Facebook. Using this data, Google's TensorFlow can be employed to make vectors of these words, and analyze the overlap of these words to come up with a few task categories that the worker would enjoy completing, before matching this up with tasks that meet these criteria, and placing them in a Tinder-like navigation to allow for quick, easy access to tasks that the worker would actually like to complete. To help tailor this system further, random questions about lifestyles, media, or ideologies will be asked for every 5 tasks that are left without acceptance by the worker. This will add to the pool of data that exists for the worker, and help give them better task recommendations.

Module #2: Tasks as a more complete learning pathway

Problem: Any form of work can allow for some amount of learning. Completing tasks in certain areas can serve as a great learning experience for the worker, and help give them access to further educational opportunities that they would not have had otherwise. This is especially considering the role that Daemo aims to play in the future, a platform where one can earn a stable source of income from, and would like to come to continually for future enrichment, and the role that it could play as the provider for jobs to unskilled laborers, especially those in India, the country whose workers and requesters make up 36% of Amazon MechanicalTurk users.

Module Details: Create a pathways program for each set of skills. Use the worker's LinkedIn and a questionnaire to see which pathway the worker would be interested in. Use natural language processing to turn each task into one objective that leads to completion of a pathway. For example, say a worker is interested in psychology, as indicated by his/her LinkedIn, they could complete the "Psychology Basics" pathway by participating in a study, and earn the "understanding of double blind procedure" objective as a part of this pathway. This system would help enrich the workers and provide them a reason to come back to Daemo, as well as provide the requesters a way to find the most qualified workers.


To experiment with Gravitate, a group of 20 recently enrolled Daemo workers will be given option of using Gravitate to better organize their tasks. The number of tasks they complete in a week will be compared to the number of tasks completed in a week prior to this features appearance to see if these features do lead to greater retention.